Month: August 2026

How to Optimize When Local Customers Stay in Google Maps
Your local rankings look steady, yet calls and website sessions are falling. If those are the only actions in your report, the obvious conclusion is that local SEO has stopped working. That conclusion may be wrong.

A growing share of customers can evaluate a business, choose it and request directions without leaving Google Maps. Your job is no longer just to earn a listing that sends traffic elsewhere. You need to make the listing useful enough to complete the decision, support it with consistent evidence and measure what happens after the click disappears.

Diagnose a journey shift before declaring traffic lost

Corrected US portfolio data comparing Q1 2026 with Q1 2025 found that calls and website clicks each fell 15.8% while direction requests rose 31.3%. The same pattern continued in Q2, but at a slower rate: calls fell 11.9%, website clicks fell 12.5% and directions increased 21.1%.

The surface mix moved as well. In the US Q2 comparison, desktop Maps impressions rose 3.2% and mobile Maps impressions rose 30.4%, while mobile Search impressions fell 20.1%. That combination supports a practical working hypothesis: some local journeys are moving from search results into Maps, where customers can act without opening the business website.

It does not prove that every lost click became a store visit. These are portfolio-level changes, not a universal forecast for your locations. A direction request is a strong expression of intent, but it is not a confirmed arrival, purchase or booked appointment. Treat it as a distinct step in the journey and connect it to business outcomes wherever your systems allow.
- Discovery: Separate Search and Maps impressions, then split them by desktop, mobile, country and location.
- Decision: Report calls, website clicks and direction requests individually. A shift between them matters even when their combined total appears stable.
- Outcome: Compare those actions with bookings, qualified leads, online orders, store-level sales or another result the business can verify.
- Interpretation: If clicks decline while directions and downstream outcomes hold or grow, the journey may have migrated. If every action and outcome declines, investigate demand, visibility, listing quality and conversion instead of assuming a channel shift.
Rank tracking cannot settle the question. On 179 Google Business Profiles, AI-powered local packs often displayed two businesses rather than three, frequently omitted the call button and surfaced only 32% as many unique businesses as the traditional Map Pack. A tracker built around the traditional pack can therefore show a stable position while the customer sees a different set of choices.

When performance changes, inspect the actual Search, Maps and AI result experiences that matter to the location. Record whether the business appears, which competitors appear, what facts are shown and which actions are available. The visible interface is evidence your rank number cannot provide.

Do not apply a US benchmark blindly across countries. In the same Q2 comparison, EU desktop Maps impressions fell 34.7% while direction requests rose 13.1%. The smaller UK dataset moved differently again: mobile Maps impressions fell 70.8% while directions and website clicks increased. For an international brand, each country needs its own baseline and explanation.

Build a Maps listing that can finish the decision

Open your profile as if you have never heard of the business. Can you establish what it offers, whether it suits your need, when it is available, whether other customers trust it and how to reach it? Any unanswered question creates friction. It may also leave Google with too little confidence to answer that question on the business’s behalf.

Make the profile complete in decision order
1. Confirm identity. Verify the business name, primary category, address or service area, phone number and website destination. Multi-location brands should verify each location rather than assuming a central data feed is correct everywhere.
2. Confirm availability. Keep regular and special hours current. If a customer can book, reserve, order or request an appointment through a supported link, test that path from a signed-out customer view.
3. Describe the actual offer. Use the relevant categories, services, products and attributes available to the profile. Completeness means supplying useful facts, not adding promotional copy to every field.
4. Test every action. Call the listed number, open the website and booking links, and check where the directions pin ends. A correct-looking profile can still send a customer to a dead page, central switchboard or wrong entrance.
5. Assign ownership. Give one role responsibility for changes to hours, services, URLs, phone routing and location status. Profile accuracy deteriorates when each field belongs to a different team and no one owns the finished customer experience.
Completeness should be judged by whether a customer can decide, not by how many fields contain text. Remove stale offers. Avoid vague service descriptions. If two locations provide different services, represent the difference instead of copying one generic profile across the estate.

Align the profile, location page and entity markup

Local visibility now has two related layers. Traditional Maps rankings still depend on factors such as proximity, relevance, engagement and prominence. AI Mode and Gemini can layer web context, entity matching, brand authority and review sentiment onto the Google Business Profile. One layer influences whether the location appears as a map choice. The other influences whether an AI system has enough coherent evidence to recommend it or answer a specific question about it.

You cannot write your way around proximity. You can reduce uncertainty about relevance and identity. The profile, visible website copy and structured data should describe the same real business.
- Create a useful page for each location, with its real name, address or service area, phone number, hours, services and customer-facing destination links.
- Keep location distinctions visible in the page copy. A unique URL with generic text does not explain why that branch is relevant to a particular need.
- Use the most specific applicable LocalBusiness structured data to restate facts that are already visible on the page. JSON-LD should corroborate the page, not introduce claims a customer cannot see.
- Resolve conflicts between the profile, location page, schema, booking system and other business-controlled records. Do not choose a preferred version for reporting while leaving the public conflict in place.
- Write plain answers to recurring questions about services, suitability, access and other decision criteria the business can substantiate. Entity clarity comes from consistent facts in context, not repeated keywords.
Google Maps accuracy is especially important for Gemini because it can draw directly from Maps data. Do not mistake that connection for a complete cross-platform AI strategy. SOCi’s 2026 Local Visibility Index, a vendor benchmark rather than a universal census, found that the share of locations recommended was 1.2% on ChatGPT, 7.4% on Perplexity and 35.9% on Google. Profile accuracy averaged 68% on ChatGPT and Perplexity versus 100% on Gemini in that benchmark. The useful lesson is not that one percentage will predict your brand. It is that different answer engines can know different versions of the same location, so you must test them separately.

Turn reviews into answer-ready evidence

Reviews are no longer only a star rating beside your name. Their language can supply evidence about the questions a local customer asks before choosing: Was the place clean? Was it expensive? What was the atmosphere like? Did the business provide the particular service the customer needed?

Google now prompts reviewers with structured concepts such as atmosphere, price and cleanliness and encourages people to review places they have visited. Cleaner, more specific review data gives an answer system more material to summarize without sending the customer to a website.

Your review program should invite useful context without scripting praise or feeding customers keywords. A neutral request can ask the customer to mention the service or product they used and what mattered in their experience. That produces more decision value than a generic request for a five-star rating.
1. Ask after a real interaction. Make the request part of the customer handoff, receipt, completion message or other natural follow-up.
2. Keep the prompt neutral. Invite an honest description of the service used, the location and the factors that mattered. Do not tell the customer what sentiment or wording to publish.
3. Analyze themes by location. Separate repeated praise, repeated complaints, service mentions and unanswered questions. A multi-location average can hide a branch-specific problem.
4. Correct the underlying facts. If customers repeatedly misunderstand parking, pricing, appointment requirements or service availability, clarify the profile and location page where accurate. If the experience itself is wrong, fix operations before rewriting the description.
5. Respond for the next reader. Address the concrete issue, correct factual misunderstandings calmly and explain a resolved change when appropriate. Do not treat the response as a place to insert target queries.
Review quality may also affect whether a location enters an AI recommendation set. In the same vendor benchmark, locations recommended by ChatGPT averaged 4.3 stars and those recommended by Perplexity averaged 4.2. Those averages do not establish a rating cutoff, and they do not prove that raising a rating alone will earn a recommendation. They do show why reviews belong in AI visibility work alongside profile accuracy and on-site authority.

Measure the Maps journey all the way to a business outcome

A local dashboard should answer three separate questions: Were you visible, what action did the customer take and did the business receive value? Combining those stages into a single traffic chart conceals the very shift you need to understand.

Build a scorecard around the action mix
- Visibility: Search impressions, Maps impressions and observed inclusion in relevant traditional and AI-assisted local results, split by device and market.
- Profile actions: Calls, website clicks and direction requests shown separately as totals and as shares of all measured profile actions.
- Website behavior: Sessions and conversions from tagged profile links, including separate appointment, order or location-page destinations where available.
- Business outcomes: Qualified calls, completed bookings, orders, visits or store-level revenue. Use the outcome the business can measure consistently rather than claiming that every direction request became a customer.
- Data quality: Incorrect fields, unresolved profile-to-site conflicts, broken destinations and location pages missing decision-critical information.
- Review evidence: Rating, review volume and recurring themes by location, with operational issues separated from content gaps.
Do not add a call, a website click and a direction request together and label the total conversions. They represent different intentions and have different relationships to revenue. Keep the raw actions visible, then calculate downstream performance only where your systems provide defensible connections.

Run a repeatable local visibility cycle
1. Establish a comparable baseline. Preserve the device, surface, country and location splits. Use a comparable prior period when seasonality makes the immediately preceding period misleading.
2. Inspect the customer experience. Review the live profile, location page, action links, review themes, traditional local results and relevant AI answers. Capture what a customer can actually see.
3. Fix factual problems first. Correct identity conflicts, inaccurate hours, wrong categories, broken links and missing service information before rewriting copy or chasing more reviews.
4. Improve one evidence layer at a time where practical. A location-page update, profile cleanup and review campaign launched together may improve performance, but it will be harder to tell which gap mattered.
5. Read the whole journey. Compare changes in visibility, action mix and verified outcomes. A click decline with rising directions tells a different story from a decline across every stage.
6. Use outliers to choose the next action. In a multi-location account, investigate branches where action mix, review themes or downstream results diverge from similar locations. The portfolio average is a starting point, not a diagnosis.
This cycle also keeps paid and organic decisions grounded. Falling calls alone are not enough to prove that organic visibility failed or that paid search must replace it. You need to know whether customers disappeared, changed actions or finished the journey somewhere your report does not yet measure.

Key takeaways
- Google Maps can be the place where a local customer discovers, evaluates and chooses a business, not merely a route to the website.
- Stable traditional rankings do not guarantee stable exposure in AI-powered local results, and falling clicks do not prove that local demand has vanished.
- A complete Google Business Profile should answer decision questions and agree with the location page, structured data and customer-facing systems.
- Reviews provide answer-ready evidence about real customer concerns, but rating averages from a benchmark should not be treated as recommendation thresholds.
- Direction requests deserve equal visibility beside calls and website clicks, but they must not be reported as confirmed visits.
- Device, surface, country and location splits are essential because local behavior can move in different directions across markets.
In your next local report, place calls, website clicks and directions beside the business outcomes they are meant to produce. Then open each priority profile as a customer and remove the most consequential unanswered question. That is how you adapt to a local journey that may end in Maps without losing sight of the result that matters.

References
- Search Engine Land – Calls and clicks keep falling as Google Maps becomes the destination
August 31, 2026

How to Measure AI Search Visibility Across the Customer Funnel

Your AI visibility score may look healthy while your brand is absent at the exact moment a buyer narrows the shortlist. The reverse can happen too: you appear in brand-specific answers but never enter the conversation while people are still defining their problem.

You need to know where your brand enters an AI-assisted buying journey, how it is represented at each stage, and what causes it to disappear. A funnel-based prompt map turns that broad visibility problem into content, authority, and measurement work you can actually prioritize.

Define visibility differently at each funnel stage

A single visibility percentage hides intent. A mention in an educational response is not equivalent to a place on a product shortlist, and a citation is not automatically a recommendation. Even a prominent answer to a branded prompt may tell you little about whether new buyers discover the brand.

Prompt mapping extends keyword mapping by organizing the questions people may ask AI platforms according to topic, intent, persona, and buying stage. It also accounts for the context people add around company size, existing technology, use case, pain point, and purchasing priority. Those qualifiers can turn one broad keyword into many plausible prompts with materially different answers.

Use four stages as a working model. Buyers will not always move through them in order, so classify the job being done in the prompt rather than trying to prove a perfectly linear journey.

Funnel stage	What the person is trying to resolve	What useful visibility looks like	What you should inspect
Awareness	Understand a symptom, risk, goal, or problem	Your expertise helps frame the problem accurately, through a relevant brand mention or an owned-page citation	Problem association, cited educational pages, terminology, and factual accuracy
Consideration	Understand possible approaches, categories, capabilities, or selection criteria	Your brand is associated with the appropriate solution and use case	Category association, capability descriptions, fit criteria, and alternatives mentioned
Evaluation	Reduce a set of options using specific requirements	Your brand makes an appropriate shortlist when it genuinely satisfies the stated constraints	Recommendation context, qualifying criteria, competitors, trade-offs, and cited evidence
Decision	Validate a named brand before acting	Pricing, compatibility, implementation, strengths, and limitations are represented accurately	Claim accuracy, objection coverage, outdated information, and unexpected competitor substitutions

This distinction changes what you optimize. At awareness, forcing the brand into every answer is not the goal. You want a defensible association with the problem and credible educational material that can support the response. At evaluation, general educational authority is insufficient if the brand disappears as soon as the buyer names an integration, industry, company profile, or operational constraint.

Decision-stage measurement requires another shift. The user has already supplied the brand name, so simple inclusion is a weak success metric. You should care more about whether the response is current, specific, fair, and useful enough to support a real decision.

Build a compact prompt map around real buying decisions

You cannot track every sentence a buyer might type. Nor do you need to. A smaller, deliberately constructed prompt set is more useful than a large collection of loosely related questions because every prompt has a known purpose in the measurement plan.

Start by defining your territory of authority. It sits where three things overlap: questions your audience needs help answering, knowledge your organization has earned through direct work, and subjects your products or specialists can credibly address. That boundary prevents your prompt map from becoming a list of every topic remotely connected to the category.

Choose a commercially relevant problem. Write the central question your organization is qualified to answer. Keep it narrower than the whole market.
Create a prompt family for every stage. Begin with the problem, move into approaches and criteria, introduce realistic qualification requirements, and finish with named-brand validation.
Add only meaningful qualifiers. Include a persona, company profile, technology requirement, pain point, or priority when it could alter which answer is suitable. Do not generate variants merely by changing the wording.
Record the expected association. State what a correct response should connect your brand with. This must be a supportable claim, not the answer you wish an AI system would produce.
Freeze a benchmark set. Preserve the exact prompt wording and record the platform, date, and other test conditions available to you. Add exploratory prompts separately so the benchmark remains interpretable.

For a company serving onboarding teams, one prompt family could progress like this:

Awareness: Why are new customers failing to complete onboarding?
Consideration: What approaches help a mid-market software company reduce onboarding delays?
Evaluation: Which onboarding platforms support our required workflow and integrate with our existing system?
Decision: What are the limitations of [Brand] for our onboarding use case?

The point is not to predict the buyer’s exact wording. It is to preserve the change in intent. If you test only broad best-product prompts, you will miss whether the brand is understood before the shortlist forms and whether it remains eligible after the buyer applies real constraints.

Give every benchmark prompt a record containing:

A stable prompt ID and funnel stage.
The underlying problem, persona, and meaningful qualifiers.
The exact prompt wording used for the benchmark.
The truthful brand association or fact being tested.
Brand inclusion, owned-page citation, and recommendation status.
How the brand is described, including strengths and limitations.
Competitors included and the criteria used to include them.
URLs or other evidence presented in the response.
Any inaccurate, incomplete, stale, or unsupported claim.
The platform, test date, and available test conditions.

Establish this baseline before publishing a new wave of content or starting a community program. Review search results, repeat the fixed AI prompts, inspect community perception, and audit whether owned content answers the questions people actually ask. A useful baseline records descriptions, sentiment, recurring concerns, recommendation contexts, and cited evidence – not just mention volume. That is how you distinguish a familiar brand name from a brand that is correctly understood.

Give every stage the evidence it needs

A prompt map is diagnostic. It tells you where visibility fails, but the remedy depends on the stage. Publishing more generic content will not repair a missing integration fact in an evaluation response, just as adding another comparison page will not establish authority around an early-stage problem.

Awareness content should clarify the problem. Explain symptoms, causes, terminology, diagnostic questions, and reasonable next steps. Help the reader recognize the situation without forcing a product into every paragraph.
Consideration content should connect the problem to possible approaches. Explain how solution categories work, what capabilities matter, where each approach fits, and which criteria separate a useful option from an unsuitable one.
Evaluation content should establish eligibility. Cover supported use cases, relevant integrations, operational requirements, comparisons, alternatives, and meaningful trade-offs. A page that targets a qualifier your product does not satisfy creates misleading visibility rather than useful visibility.
Decision content should become the canonical factual layer. Keep pricing, compatibility, implementation requirements, limitations, and other validation details consistent wherever you publish them. Address uncomfortable objections directly instead of leaving third parties to define them.

Do not reduce this work to page formats. A comparison page with vague claims supplies less decision evidence than a focused support page that states exactly what works, what does not, and under which conditions. The content job is to make the required evidence explicit and internally consistent.

Community participation provides a different kind of evidence. Relevant Reddit discussions can reveal the language people use, the alternatives they consider, the objections polished marketing pages avoid, and the criteria that actually decide a purchase. Those observations should feed your website, while accurate owned resources should give community teams dependable material for complex answers. Search intent, community context, owned depth, and ongoing monitoring should reinforce the same credible territory.

Reddit is not a shortcut to a citation. Promotional replies with little practical value are likely to weaken trust in the community you are trying to understand. Participate only where you can answer the question on its own terms. Disclose your affiliation, respond directly, acknowledge limitations and trade-offs, and link only when the destination adds information the reply cannot reasonably contain. This native, transparent approach to community authority is slower than distributing promotional messages, but it produces more useful interactions and better inputs for your content program.

Use a simple evidence loop:

Capture a recurring question, objection, misconception, or decision criterion from search and community discussions.
Match it to the relevant stage and benchmark prompt cluster.
Update or create the owned resource that can answer it completely.
Give customer-facing and community teams a clear factual reference.
Re-run the relevant prompts and record whether the answer, description, citations, or recommendation context changed.

JSON-LD belongs after the evidence is sound. Structured data can make the entities and relationships on a page more explicit to machines, but it cannot manufacture an unsupported product fit, repair contradictory pricing, or replace the experience and context supplied by independent discussions. Treat schema as a precise representation layer for content you can already defend.

Measure exposure without confusing it with traffic

Your scorecard should preserve several different outcomes. Collapsing them into one number recreates the problem the funnel map was meant to solve.

Stage inclusion rate: the share of benchmark prompts in a stage where the brand appears in a relevant capacity.
Owned citation rate: the share of stage prompts where an owned page is cited or linked. Keep this separate from brand inclusion because an answer may use your material without recommending your brand.
Category association rate: the share of consideration prompts that connect the brand with the appropriate solution category or capability.
Qualified shortlist rate: the share of evaluation prompts where the brand is recommended after the stated constraints are applied.
Representation accuracy: the proportion of reviewed brand claims that are current, complete enough for the question, and supported by your canonical information.
Competitor context: which alternatives appear, for which criteria, and whether your brand is framed as a peer, specialist, fallback, or unsuitable option.

Keep the denominator stage-specific. Awareness inclusion should not compensate for inaccurate decision answers. A high citation rate should not conceal a weak shortlist rate. A branded mention should not be counted as discovery when the user supplied the name in the prompt.

Google Search Console now adds a second view of the problem. As of August 31, 2026, its AI performance reporting is available globally to Search Console accounts. It reports impressions for content appearing in AI responses, AI Mode, and AI Overviews, with breakdowns for pages, countries, devices, and dates. It does not include click data.

Use that report as an exposure layer:

Identify which pages receive generative-search impressions.
Map those pages to the funnel stage they were designed to support.
Review changes across the available date, country, and device dimensions.
Compare exposed pages with the pages actually cited in your benchmark prompt checks.
Investigate why important stage-specific pages have prompt visibility but little reported exposure, or exposure without the brand representation you intended.

Do not calculate an AI click-through rate from this report; the necessary click figure is not present. Do not infer visits or conversions from impressions either. Use your site analytics to evaluate any visits you can separately observe, and keep the claim narrow: Search Console tells you that exposure occurred, while prompt tracking tells you where and how your brand appeared within the buying journey.

Search Console also provides a control for blocking content from Google’s generative search features, including AI Overviews, AI Mode, and AI Overviews in Discover. A site that opts out will not receive impressions or traffic from those generative features, while the choice is not used as a ranking signal for search results outside them. Treat this as a distribution and governance decision, not a way to repair weak content or inaccurate representation.

Before changing that control, document the exact properties in scope, preserve your current baseline, and make sure the owner of the decision accepts the loss of generative exposure and possible traffic. If the problem is an outdated answer, correct the canonical facts and connected authority signals. Removing the site from the feature prevents participation; it does not improve the description buyers may encounter elsewhere.

Turn each visibility gap into a specific action

The funnel pattern matters more than the aggregate score. Read the pattern first, then choose the smallest intervention that supplies the missing evidence.

Strong decision visibility, weak awareness visibility: people who already know the brand can investigate it, but the brand is not entering earlier problem discovery. Build better problem education and participate in the communities where those problems are described in real language.
Strong awareness visibility, weak consideration visibility: your material may explain the issue without connecting your expertise to a suitable method or category. Add the bridge: approaches, mechanisms, capabilities, selection criteria, and explicit boundaries of fit.
Strong consideration visibility, weak evaluation visibility: the brand is associated with the category but disappears when requirements become specific. Identify the exact qualifier causing the drop, then publish evidence for supported integrations, use cases, customer profiles, or operating constraints. Do not create fit claims for criteria the product cannot meet.
Evaluation inclusion followed by inaccurate decision answers: the brand makes the shortlist, but validation material is stale, inconsistent, or incomplete. Correct canonical pages first, state limitations plainly, and address recurring misconceptions in appropriate community and support channels.
Owned pages are cited but the brand is not shortlisted: your content influences the explanation without proving supplier fit. Strengthen verifiable differentiation, use-case evidence, and transparent trade-offs instead of merely repeating the brand name more often.
The brand appears without owned citations: third parties may be carrying much of the representation. Monitor those descriptions closely and publish clear canonical facts that customers, communities, and answer systems can check.

We would prioritize accuracy before reach. Incorrect pricing, compatibility, limitations, or implementation information in a high-intent answer deserves attention before a broad effort to increase awareness mentions. Next, address evaluation gaps that wrongly exclude a genuinely suitable product. Then expand early-stage authority where the brand has earned a reason to participate.

For every intervention, create an action card with the funnel stage, affected prompt cluster, observed failure, missing evidence, planned content or community change, responsible owner, and next review date. This keeps a visibility diagnosis from dissolving into a generic instruction to publish more.

Key takeaways

Measure awareness, consideration, evaluation, and decision prompts separately because a mention has a different meaning at each stage.
Track a compact benchmark set built around real changes in intent and meaningful buyer constraints.
Record citations, recommendation context, competitors, trade-offs, and factual accuracy instead of counting brand mentions alone.
Use owned content for depth, community participation for context, and structured data to represent evidence that already exists.
Treat Search Console’s AI report as exposure data, not click or conversion reporting, and treat its opt-out control as a distribution decision.

Start with one important customer problem. Assign its existing pages and prompt families to the four stages, capture the baseline, and find the first point where a suitable brand disappears or becomes inaccurate. Fix that break with evidence you can defend, then measure the same prompts again.

References

August 31, 2026

Microsoft Advertising Console Redesign: A Practical Checklist
When the redesigned Microsoft Advertising console appears in your account, the practical question is not whether it looks cleaner. It is whether you can complete routine work without losing context, slowing an approval, or making a change in the wrong place.

The sensible response is a controlled preview. Treat the redesign as a workflow migration: verify navigation, table behavior, account context, and team handoffs before you rely on it for changes that can affect spend or delivery.

Separate the interface change from campaign behavior

The preview centers on redesigned navigation and enhanced tables. Microsoft is positioning those changes as improvements to clarity, consistency, and data interaction. The preview may change before general availability, and users can switch back to the classic experience.

The most important boundary is simple: the redesign does not directly change campaign performance data. A new menu structure or table layout does not, by itself, alter delivery, targeting, bids, budgets, or measurement.

That distinction matters when you troubleshoot. If spend, conversions, or another campaign result moves after you enter the preview, timing alone does not establish a connection. Check the campaign settings, change history, measurement setup, and other factors that normally explain performance. Do not treat the interface rollout as the cause merely because it happened at roughly the same time.

The interface can still matter indirectly. A user who misreads the active account, overlooks the date range, or assumes a familiar control is in its old location can make the wrong operational decision. The redesign is therefore a usability change with potential workflow consequences, not a campaign optimization.

Choose the preview based on the work in front of you

You do not need one permanent answer for the whole account. Because the classic experience remains available during the preview, you can decide which interface is appropriate for the task you are doing.

The preview is a reasonable choice when you have room to compare it with a known workflow, document what moved, and stop if something is unclear. This is especially useful for people who spend enough time in the console to notice whether clearer navigation or revised tables actually reduce friction.

Use the classic experience for the current task when certainty matters more than exploration. Examples include a time-sensitive launch, an urgent account investigation, or a high-consequence change whose verification path your team has only documented in the classic interface. Switching back is not a verdict on the redesign; it is a risk-control decision for that job.
- Try the preview when you can compare the same routine task in both interfaces and record any differences.
- Stay in the preview when you can clearly identify the active scope, find the information you need, and verify the outcome without guesswork.
- Switch back for the task when a control, table state, or account context is ambiguous.
- Escalate a repeatable issue when the same workflow fails or becomes unclear after you reproduce it. The preview explicitly invites feedback, so report the task, expected result, and point of failure rather than saying only that the design is confusing.
This task-by-task approach lets your team learn the new console without forcing everyone to adopt an unfinished workflow at once.

Run a workflow audit instead of browsing at random

Clicking through menus can make a redesign feel familiar without proving that it supports your real work. Test it with a representative task from beginning to end.
1. Choose a recurring job. Use something concrete, such as locating a campaign, reviewing a performance table, checking a setting, or preparing information for an approval.
2. Record the classic path. Note where the job begins, which account or campaign context must be active, which pages and table fields matter, and how you confirm that you reached the intended result.
3. Repeat the job in the preview. Make the first pass read-only. Learn the route before changing a budget, bid, targeting choice, status, or other setting that can affect delivery.
4. Check context at each transition. Confirm that the account, campaign level, date range, table view, and selected rows still mean what you think they mean. Do not rely on muscle memory from the classic layout.
5. Compare decision quality, not just speed. Fewer clicks are not an improvement if you are less certain about scope or have to reopen pages to verify what you saw.
6. Test the handoff. Give your notes or screenshot to the next person in the process. If they cannot reproduce the route or interpret the same table, the workflow is not ready to become the team standard.
Keep the audit tied to outcomes. “The navigation feels different” is hard to act on. “The reviewer could not confirm which campaign level was active before approving the change” identifies a specific failure and the safeguard you need.

Test enhanced tables as decision tools

Tables are where a visual improvement can become either a real productivity gain or a quiet source of mistakes. A table is successful only when it helps you understand the correct data in the correct scope and act on it deliberately.

Use the table actions that matter in your normal workflow. If you depend on sorting, filtering, changing visible fields, selecting rows, or moving between result sets, test those actions rather than assuming “enhanced” means equivalent behavior.
- Scope: Can you tell what entity level and account context the rows belong to before interpreting a value?
- Time: Is the active date range visible enough that you will not compare unlike periods by accident?
- Fields: Can you find the columns required for the decision without confusing similarly named values?
- View state: After changing the way the table is displayed, can you tell that you are no longer looking at the default view?
- Selection state: Before taking an action, can you identify exactly which rows are selected?
- Verification: Can you confirm the result of an edit independently instead of trusting that a click succeeded?
If any answer is no, do not compensate by moving faster. Return to a verified view, use the classic experience if necessary, and document the ambiguity. A redesigned table should reduce the mental effort needed to make a sound decision; it should not merely fit more information on the screen.

Keep team documentation usable while the preview changes

The preview may change before general availability, so documentation created from it can age quickly. That does not mean you should avoid documenting the new interface. It means your instructions need to distinguish stable business rules from temporary screen directions.
- Label the interface. Mark screenshots and instructions as Preview or Classic, and include when they were captured or verified.
- Name the task outcome first. “Verify campaign scope before changing the budget” survives a navigation change better than an instruction built only around a menu position.
- Keep both paths where the difference matters. A teammate using the classic interface should not have to interpret directions written for the preview.
- Document checkpoints. State what the user must verify before an edit and what evidence confirms that the edit reached the intended entity.
- Retire old directions deliberately. Do not delete the classic workflow merely because one person received the preview overlay. Rollout visibility can differ between users, and the preview itself is not yet the final interface.
For agencies and larger teams, assign ownership of the workflow notes. Otherwise, individual users will create separate shortcuts, screenshots, and explanations, and the interface change will turn into a process-consistency problem.

Key takeaways
- The Microsoft Advertising preview changes navigation and tables, not campaign performance data directly.
- The preview may change before general availability, and you can return to the classic experience while evaluating it.
- Test a complete recurring workflow instead of judging the redesign by appearance or casual browsing.
- Make the first run read-only, then verify account scope, date context, table state, row selection, and the final outcome before making consequential changes.
- Maintain interface-specific instructions until every person involved in a workflow can reproduce and verify the new path.
When the preview appears, choose one recurring task, record how you complete it in the classic console, and repeat it without making a live change. If the new route preserves context and makes verification easier, adopt it for that workflow. If anything remains ambiguous, finish the work in the verified classic environment and keep a precise note of what to retest as the preview evolves.

References
- Search Engine Land — Microsoft Advertising rolling out new design for advertiser console
August 31, 2026

How to Build Trustworthy AI Agents for Marketing Operations

You have an agent that can inspect ad accounts overnight, draft a content brief before stand-up, or flag a broken funnel. The uncomfortable question arrives just after the demo: what, exactly, are you willing to let it do without asking?

If your answer is “we’ll review it,” you don’t yet have a control system. You have an intention. A trustworthy marketing agent needs a bounded job, owned data, explicit permissions, evidence attached to its conclusions, a release gate, and a way to stop or reverse its actions. Here is how to put that operating model in place.

A trustworthy agent is a controlled workflow, not a clever model

A model generates an answer. An agent combines a model with data, instructions, tools, scheduled triggers, and permission to take or prepare actions. That surrounding system determines whether a plausible mistake becomes a harmless draft, a misleading alert, or a customer-facing incident.

Trustworthiness therefore isn’t the promise that an agent will never be wrong. It is your ability to see what the agent observed, understand why it reached a conclusion, constrain what it can do, route uncertain cases to the right person, and recover when something fails. In production, reliability is decided by governance, realistic testing, and named review paths at least as much as by model capability.

The most useful mental model is a new employee with unusual speed. You wouldn’t give a new marketing analyst unrestricted CRM access, authority to change pricing, and permission to email customers on the first morning. You would define the role, grant only the access it needs, review early work, and expand responsibility after the work proves dependable. An AI agent needs the same management discipline, encoded in the workflow rather than left in a manager’s head.

Before deployment, make sure every agent has clear answers to these questions:

What specific decision or task does the agent own?
Which systems, records, fields, and time periods may it inspect?
Which facts and business rules must it know before making a judgment?
What evidence must accompany each conclusion or recommendation?
When must it abstain, escalate, or ask for missing information?
Who reviews consequential work, and what counts as approval?
Which actions can it take, and how can those actions be stopped or reversed?
Which version of the model, instructions, tools, and data definitions produced the result?

If any answer is “it depends,” write down what it depends on. That conditional logic is part of the product. It cannot remain tribal knowledge if the agent is expected to make repeatable decisions.

Begin with one bounded decision, not a general marketing assistant

“Monitor our marketing” sounds like a useful assignment, but it contains dozens of hidden jobs. Does monitoring mean detecting a tracking outage, explaining a CPA change, checking whether campaigns are serving, judging lead quality, finding off-brand copy, or recommending budget shifts? Each job needs different data, context, freshness rules, and escalation paths.

Start with a task whose input and acceptable output can be described precisely. Read-only analysis is usually the safest entry point because the agent can create value without changing the underlying system. Examples include investigating an ad-delivery alert, identifying content briefs with missing source material, finding inconsistent campaign naming, or preparing a proposed JSON-LD correction for validation and human review.

Write a short job card for the workflow:

Trigger: State what starts the run, such as a scheduled account check or an anomaly from an existing monitoring rule.
Question: Express the decision in one sentence. For example: “Has campaign delivery stopped during comparable business hours?”
Inputs: Name the approved systems, fields, reporting windows, business rules, and account notes.
Output: Define the required finding, supporting evidence, uncertainty, and proposed next step.
Prohibited behavior: State what the agent must not infer, retrieve, publish, send, or change.
Escalation: List the conditions that require abstention or human judgment.
Reviewer: Assign a role or person responsible for accepting consequential recommendations.
Success and failure: Describe both a useful result and an unsafe result. A fluent explanation without adequate evidence belongs in the failure column.

Pay special attention to time. Marketing data often arrives on different schedules, so “recent” does not necessarily mean “complete.” A production ad-management agent once interpreted conversions that had not arrived yet as a severe performance decline. Making its analysis dependable required safe comparison windows, conversion-maturity rules, uncertainty ranges, and refusal when the lag could not be modeled reliably.

Apply that lesson beyond paid media. A CRM agent should not label a campaign unproductive before the normal sales cycle has elapsed. A content agent should not declare a page unsuccessful before the chosen reporting period is complete. An SEO agent should not turn a partial crawl or delayed analytics import into a confident diagnosis. Freshness and maturity are different properties, and the agent needs rules for both.

Refusal is not a defect when the evidence is immature, contradictory, or missing. A trustworthy response may be: “I cannot distinguish a real decline from reporting delay with the approved data.” That is more useful than an elaborate guess because it tells the operator what information is needed next.

Give the agent a data contract and a business context pack

Connecting an agent to more systems does not automatically make it better informed. It can instead create several conflicting versions of revenue, conversion, customer status, or campaign ownership. The agent will still produce coherent prose even when the underlying records disagree.

A data contract tells the agent what it may use and how each input should be interpreted. Create one before refining the prompt. For every permitted input, record:

The system and field that hold the data.
The business owner responsible for its meaning and quality.
Whether it is the authoritative value or a convenience copy.
How frequently it updates and when it becomes mature enough for judgment.
The unit, attribution rule, time zone, status definition, and other interpretation rules.
Known gaps, exclusions, and failure signals.
What the agent must do when the input is absent, stale, or inconsistent.
Whether the field contains personal, confidential, regulated, or otherwise restricted information.

Then create a separate context pack for facts that do not live cleanly in reporting tables. Include the products the business actually sells, excluded services, target locations, budget constraints, active promotions, sales-cycle expectations, conversion-lag patterns, campaign goals, approved claims, brand restrictions, and known tracking limitations. Without this context, an agent can correctly calculate the numbers and still reach the wrong business conclusion. A paid-media agent, for example, cannot identify an irrelevant pet-insurance keyword for a business-insurance advertiser unless it knows what the business sells and can access the operational context used by human analysts.

Keep the context pack owned and maintainable. Each rule should have an owner, a status, and a replacement path when the business changes. Otherwise an old promotion, discontinued service, or superseded approval rule can remain active inside the agent long after people have moved on.

Use least-privilege access. If the task requires campaign totals, do not expose raw customer records. If the agent only prepares a content update, give it draft access rather than publishing rights. If it reads a CRM status, restrict it to the approved fields rather than the full contact object. Governed implementations can limit access to approved data, mask immature conversion information, and require evidence for recommendations.

Trace where the data goes as well as what the agent can retrieve. Before customer, prospect, health, or financial information reaches a third-party AI service, determine where it is processed, what the provider may retain or reuse, and which internal policy governs that transfer. Marketing data deserves the same boundary-setting applied to other sensitive operational systems; convenient access is not the same as necessary access.

If the team cannot identify the owner or meaning of an important field, stop at read-only experimentation. A better prompt cannot resolve a disputed definition of revenue, repair missing conversion data, or decide which system is authoritative.

Set autonomy by consequence and reversibility

An AI device faces three increasingly restricted action zones, from reversible draft tasks to guarded campaign controls and a locked high-consequence mechanism.

Teams often treat autonomy as a switch: either the agent acts or a person does. A safer design separates observation, recommendation, preparation, and execution. The agent can then earn broader permissions without receiving blanket authority.

Operating level	Marketing example	Default permission	Release condition
Observe	Check reporting data and surface a possible anomaly	Read approved fields; create an internal record	Freshness checks pass and evidence is attached
Recommend	Explain a performance change or propose a content correction	No external change	Assumptions, uncertainty, affected assets, and reviewer are explicit
Prepare	Build a draft ad, email, brief, metadata edit, or schema patch	Write only to a draft or sandbox	Validation passes and a named person approves publication
Act	Pause a campaign, move budget, publish content, change pricing, or send a message	Off by default	The action is narrowly pre-approved, policy-compliant, observable, and safely reversible; otherwise human approval remains mandatory

Two variables should control the level: consequence and reversibility. A duplicate internal alert is annoying but recoverable. An incorrect customer email, pricing change, destructive CRM update, or large budget movement can create brand, financial, privacy, or legal exposure. Work carrying that weight needs a human checkpoint; letting an unreviewed agent send customer communications or make consequential commercial decisions is not an acceptable starting posture.

For high-impact recommendations, add an independent check before the decision reaches the approver. That check should evaluate the evidence and policy conditions, not merely ask another model whether the prose sounds convincing. It can verify that the reporting window is mature, the cited records exist, the requested action is permitted, and contradictory data has been surfaced. Higher-stakes analysis benefits from a separate review path before a person is asked to act.

Require an evidence packet for every recommendation. It should contain:

The conclusion in plain language.
The period, comparison, account, page, campaign, or record under review.
The approved inputs actually used.
Missing, stale, masked, or contradictory inputs.
Assumptions and relevant business rules.
The agent’s uncertainty or reason for abstaining.
The proposed action and assets it would affect.
The required approval and available rollback path.

Do not allow the agent to hide uncertainty inside polished prose. Evidence must be inspectable by the person making the decision. If a recommendation cannot be traced back to permitted inputs, it should fail the release gate regardless of how reasonable it sounds.

Release, monitor, and stop the agent like production software

Human operators monitor an AI agent moving from testing through a gated deployment lane, with health sensors, an evidence trail, an emergency stop, and a rollback track.

Test safe behavior, not just good answers

A handful of impressive demo prompts proves very little. Build an evaluation set from the situations the agent will face after release: routine work, different ways users phrase the same request, incomplete data, delayed conversions, stale account notes, conflicting systems, out-of-scope requests, and cases where the correct response is escalation.

For each case, define the expected behavior rather than one perfect paragraph. Should the agent answer, flag uncertainty, request information, refuse, or escalate? Which evidence must appear? Which tools may it call? Which actions must remain blocked? This makes the evaluation durable even when wording varies.

Add simple pass-or-fail checks around important invariants:

A read-only agent cannot invoke a write operation.
A draft-only content agent cannot publish.
Restricted fields never appear in retrieved context or output.
A performance judgment cannot use a reporting window marked immature.
A recommendation cannot pass without evidence identifiers and required assumptions.
A missing authoritative input triggers the prescribed abstention or escalation.
An action outside the job card is rejected even when a user asks persuasively.

Run the agent in shadow mode before granting action rights. Let it inspect real work and produce results without changing external systems. Compare its findings with the decisions made through the existing process, examine both disagreements and omissions, and update the job card, data contract, context pack, and evaluation set. Only then consider expanding its operating level.

Version every component that can change behavior

The prompt is not the whole agent. Store the system instructions, policy rules, model identifier, provider settings, tool definitions, data-field mappings, business definitions, context-pack version, evaluation results, approval decision, and release date as one traceable configuration.

This matters because behavior can drift even when your team changes nothing visible. A provider can update the model underneath a workflow, while a data field, tool response, or business rule can change independently. Unversioned models and prompts make it difficult to explain why customer-facing behavior changed or recreate how the system acted earlier. Marketing teams need release discipline and behavior monitoring around model and prompt changes, just as they do around application changes.

Rerun the relevant evaluations whenever any behavioral component changes. If the provider does not expose a fixed model version, record the identifier it does provide and use recurring evaluation results to detect observed changes. Do not assume unchanged prompts guarantee unchanged behavior.

Monitor usefulness, silence, and operator burden

Accuracy on answered cases is not enough. Monitor unsupported conclusions, inappropriate certainty, policy violations, reviewer overrides, action reversals, duplicate alerts, unnecessary escalations, and cases where a reviewer had to retrieve evidence the agent should have supplied.

Review non-alerts as well as alerts. An agent can look quiet because nothing is wrong, because its thresholds are sensible, or because it missed the problem. Sample runs where it concluded that no action was needed and verify that the underlying data supports that silence.

Noise is an operational failure. If people repeatedly dismiss duplicate, untimely, or low-value alerts, they will stop treating the agent as a useful colleague. A working ad-management agent had to remove duplicate notifications and messages that could wait because convincing the team to pay attention depended on reducing noise as well as improving analysis.

Give operators a visible stop path. When the agent behaves unexpectedly, they should be able to pause scheduled runs, revoke write credentials, preserve the decision trace, identify the changed component, rerun evaluations, and restore a known configuration. Re-enable a smaller scope before returning full permissions.

Rollback has limits. You can restore a campaign setting or draft, but you cannot make a sent email unread or erase a public impression of an incorrect claim. Keep human approval in front of actions whose consequences cannot be meaningfully reversed.

Key takeaways

Trust is a property of the whole workflow: data, context, permissions, evidence, review, monitoring, and recovery.
Start with a bounded, read-only decision whose correct inputs and safe output can be described precisely.
Treat data freshness, data maturity, and business meaning as separate requirements.
Grant the minimum fields and tools needed for the job; broad access is not a substitute for context.
Increase autonomy according to consequence and reversibility, not model fluency.
Make abstention, escalation, and evidence-bearing recommendations part of the success criteria.
Version every component that can change behavior, then retest and monitor real-world use.

Pick the smallest marketing decision that currently consumes repeated human attention. Write its job card and data contract before connecting an agent. If you cannot define the evidence, permissions, reviewer, and stop path, the workflow is not ready for autonomy. If you can, you have a foundation that can earn broader responsibility instead of merely requesting trust.

References

August 31, 2026

Dynamic Google AI Overviews: A Practical SEO Response Plan

Your page can keep its organic position and still lose the part of the search result that used to earn the click. When Google automatically opens a full AI Overview, conventional listings begin farther down the screen and compete with a much more complete answer.

If your clicks soften, do not begin by rewriting pages or changing schema. First establish whether your ranking changed, the search layout changed, or both. That distinction determines whether you need an SEO fix, a stronger reason to visit, or simply better monitoring.

How dynamic expansion changes the click opportunity

For some queries, Google can turn a compact AI Overview into a fully expanded response without requiring the user to select Show more. The larger answer pushes the core search results down and makes the standard results page look more like AI Mode.

The expansion is interaction-aware. If the user has already started scrolling through content below the overview, Google cancels the expansion to preserve the reading position. That small detail matters when you audit results: scrolling too quickly can cause you to record a compact treatment even when the page would otherwise have expanded.

One observed treatment also exposes an AI Mode-style follow-up prompt by default. The search experience is therefore not merely adding a longer summary. It can move the user directly from an initial query into a conversational follow-up without first sending that user to an external page.

Google says its systems expand the overview for topics where doing so appears most useful. It also says its own research found the experience more helpful and associated it with deeper follow-up exploration. Treat that as Google’s account of user behavior, not independent proof that every affected search is better or that every website will lose traffic.

The affected share of queries and the future scope of the behavior remain undisclosed. You cannot responsibly apply a sitewide traffic assumption from a single result or screenshot. You need query-level evidence.

Key takeaways

An expanded AI Overview is a search-layout change, not evidence by itself that your organic ranking fell.
Capture the result before scrolling because scrolling can prevent the expansion you are trying to observe.
Track no overview, compact overview, expanded overview, and expanded overview with a follow-up prompt as separate states.
Keep the direct answer clear, but give the reader a concrete reason to continue to your site.
Do not treat schema changes as a remedy for dynamic expansion; no such control has been identified in the confirmed behavior.

Audit exposure before you change the page

An analyst compares two monitors showing compact and expanded versions of a generic search-results layout.

Start with the queries that already matter to the business. A broad search for random examples will tell you that the feature exists, but it will not tell you whether it affects the pages responsible for your leads, sales, subscriptions, or assisted conversions.

Build a query-page shortlist. Use the queries and landing pages you already monitor. Prioritize combinations with meaningful organic traffic or business value.
Load each result without scrolling. Allow the initial experience to settle before interacting with the page. Record whether the AI Overview is absent, compact, or automatically expanded.
Record the prompt treatment. Note whether a follow-up field is exposed by default. This distinguishes a long answer from a result that actively invites the user into an AI Mode-like journey.
Preserve the test conditions. Log the device type, viewport, location, language, signed-in state, query wording, and observation time. Keep these conditions consistent when you revisit the query.
Repeat the observation. The experience is dynamic, so one appearance is not enough to classify a query as consistently affected. Preserve screenshots or recordings rather than relying on memory.
Attach the result state to performance data. Place the observation beside the corresponding query-page clicks, impressions, click-through rate, and average position from Google Search Console.

A simple four-state field makes this usable in a spreadsheet or dashboard: no overview, compact overview, expanded overview, and expanded overview with prompt. Do not combine those states into one AI Overview label. The amount of screen space and the available next action are the very things you are trying to evaluate.

Read the metrics as a pattern, not a verdict

No single metric proves that expansion caused a change. Use the combinations below as working interpretations that tell you what to investigate next.

What you observe	Reasonable working interpretation	Next action
Expanded overviews recur while impressions and average position remain broadly steady, but clicks or click-through rate weaken	A layout-related loss is plausible	Test the search-result promise and the page’s continuation value before treating this as a ranking problem
Average position weakens alongside clicks	Ranking movement may be contributing	Run the normal technical, content, intent, and competitive diagnosis as well as the layout audit
Impressions weaken across the query cohort	Demand, coverage, or ranking may have changed	Check those factors before assigning the decline to AI Overview expansion
An expanded treatment appears once but not in repeated checks	The evidence is unstable	Continue observing under consistent conditions and avoid a large rewrite
Expanded overviews recur but business outcomes remain healthy	The feature is visible without a demonstrated business problem	Monitor it and leave a productive page alone

Design content for both the answer and the next step

A full AI response occupies more of the result page, so a visit must justify itself beside a larger answer. The wrong reaction is to make your content vague in the hope that withholding the answer will force a click. That weakens the page for the person who visits and can make its subject harder for search and answer systems to interpret.

Make the core answer easy to understand

Answer the primary question early and in plain language. Do not bury the conclusion beneath a long scene-setting introduction.
Name the relevant product, feature, entity, or constraint precisely. A qualified answer is more useful than an absolute claim that ignores context.
Keep supporting evidence beside the claim it supports. Readers and machines should not have to infer which citation belongs to which statement.
Use headings that reflect the decisions and follow-up questions a reader actually has, rather than repeating slight variations of the target keyword.
Keep structured data accurate and consistent with visible content. Do not add markup for information the page does not show.

Give the visit a specific job

The overview may handle a basic explanation. Your page should help the reader complete the work that follows. Depending on the query, that could mean choosing between options, applying a process, checking an exception, using a template, validating a result, or seeing the evidence in full context.

For decision queries: compare options against explicit dimensions and explain when each choice fails, not only when it works.
For implementation queries: include prerequisites, ordered steps, validation checks, common failure points, and recovery paths.
For analytical queries: expose the method, assumptions, definitions, and limitations behind the conclusion.
For recurring tasks: provide a usable checklist, worksheet, calculator, template, or other tool that reduces the reader’s work.
For complex topics: connect the primary answer to tightly relevant supporting pages so the reader can move directly into the next decision.

This creates two separate goals for AEO and SEO: make the information clear enough to participate in an answer experience, and make the destination valuable enough to deserve a visit. One does not guarantee the other.

Do not reach for speculative schema changes. The confirmed description ties expansion to Google’s assessment of usefulness for the query; it does not identify JSON-LD or another publisher-controlled switch that can keep an overview compact. Continue using valid schema for the entities and content visibly present on the page, but judge that work by its intended purpose rather than treating it as an expansion override.

Turn observations into a decision rule

Abstract evidence tiles converge at a glass junction and branch toward a wrench, an open doorway, and a radar symbol.

Sitewide organic traffic is too blunt for this diagnosis. Dynamic expansion applies to some queries, while unaffected queries can hide or exaggerate the movement. Keep an affected-query cohort and evaluate it separately from the rest of the site.

Separate visibility from visits. Track impressions and average position as one layer, then clicks and click-through rate as another. This prevents a presentation change from being mislabeled as a ranking loss.
Separate visits from value. Review qualified sessions, conversions, assisted outcomes, or the business event appropriate to the page. Fewer clicks are more serious when the lost visits previously produced meaningful outcomes.
Annotate observed result states. Keep the screenshot or recording with the query, conditions, and date. A label without evidence becomes hard to verify later.
Compare like with like. Evaluate the same query-page combinations and account for ordinary demand or seasonal changes before assigning a cause.
Choose the response in advance. Decide what evidence will trigger monitoring, a content test, a ranking investigation, or no action. This keeps a striking search result from driving an unnecessary sitewide rewrite.

If expanded overviews recur, organic visibility remains stable, clicks weaken, and business outcomes decline, test a more specific search-result promise and strengthen the page’s next-step value. If ranking weakens too, investigate the ranking problem independently. If the expanded treatment appears without a meaningful performance loss, record it and keep the productive page intact.

Your next move is small: add the four-state AI Overview field to the query log you already use, then capture priority results before changing content. Once layout evidence sits beside visibility, clicks, and business outcomes, you can act on the queries where dynamic expansion creates a real problem and leave the rest alone.

References

Search Engine Land — Google is dynamically expanding AI Overviews for some queries

August 28, 2026

How to Automate AEM Content Updates with Profound Agents
You have an AI visibility finding, a clear content fix, and an Adobe Experience Manager workflow standing between the two. The diagnosis may take minutes. The ticket, CMS handoff, review, and update can take much longer.

Profound Agents can now List, Search, Get, Create, and Update Content Fragments in Adobe Experience Manager. That gives you a direct route from an approved insight to a controlled CMS change. The important word is controlled: the safest design is not an agent with unrestricted publishing power, but a bounded workflow that retrieves the right fragment, proposes a field-level change, passes validation, and writes only after the required approval.

What the AEM nodes actually let you automate

The integration operates on AEM Content Fragments. In a workflow design, give each available action a narrow job:
- List supports inventory work when the workflow needs to inspect a defined collection of fragments.
- Search helps locate candidates related to a target entity, topic, path, locale, or other supplied criterion.
- Get retrieves the exact fragment before any decision or write occurs.
- Create adds a new Content Fragment when no suitable canonical fragment exists.
- Update changes an existing fragment that already represents the intended entity or content unit.
This distinction matters because a Content Fragment is not the same thing as a rendered web page. A page may reference the fragment, transform its fields through a component, expose it through an API, or combine it with content from other systems. If the target copy is hard-coded in a component or owned by another service, changing a Content Fragment will not necessarily change that copy.

The named action set also does not include a separate Publish action. Do not treat a successful Create or Update operation as proof that the new content is live. Document the downstream activation, deployment, cache, and rendering steps in your implementation. Then verify the delivered page or endpoint, not only the object stored in AEM.

Get should normally precede Update. Without that read step, the agent may work from an old brief, overwrite a newer human edit, or modify a fragment that merely resembles the intended target. Retrieval is part of the safety model, not administrative overhead.

Build the workflow around a write contract

Start with one content model, one permitted content root, one locale, and one repeatable use case. A focused pilot might update an approved answer field in an existing fragment. A poor first pilot gives the agent authority to rewrite product claims across several models and markets.

Before connecting an insight to an AEM write, define a write contract. This is the machine-readable boundary that tells the workflow what it may change and when it must stop.
- Target scope: the allowed AEM path, Content Fragment Model, brand, market, and locale.
- Permitted actions: whether the run may Search and Get only, Update an existing fragment, or Create a new one.
- Writable fields: the specific fields the agent may alter. Treat identifiers, ownership fields, workflow state, canonical references, and other structural fields as immutable unless the use case requires them.
- Evidence inputs: the approved facts, URLs, product data, and editorial instructions the generated copy must follow.
- Stop conditions: no match, multiple plausible matches, a model mismatch, a locale mismatch, missing evidence, failed validation, or a fragment that changed after retrieval.
- Approval rule: who must accept the field-level diff before the write and whether a separate approval is required before activation.
- Completion record: the target identifier or path, operation used, fields changed, prior and new values, validation result, reviewer, and downstream publication state.
With that contract in place, use the nodes in a deliberate sequence:
1. Receive a qualified opportunity. Supply the target query or audience need, the reason for the change, the approved evidence, and the expected content destination. Do not ask the agent to infer business truth from a visibility gap.
2. Locate candidate fragments. Use Search for a targeted lookup or List within a tightly bounded collection.
3. Resolve one exact target. Match on stable attributes such as an approved identifier, path, model, entity, and locale. A similar title is not enough.
4. Retrieve the current fragment. Use Get so the workflow can preserve existing fields and compare the current value with the proposed value.
5. Choose Create, Update, or stop. Make this an explicit decision rather than allowing a failed search to become an automatic Create.
6. Generate a field-level patch. Ask for only the fields that need to change. Avoid regenerating the entire fragment when one answer, description, or evidence field is the actual target.
7. Validate before writing. Check the Content Fragment Model, required fields, allowed values, link formats, locale, evidence constraints, and any length rules imposed by the destination.
8. Review the diff. Show a human reviewer the exact old and new values, along with the evidence behind the change. Reviewing polished prose without the prior value hides unintended deletions.
9. Execute and verify. Run Create or Update, retrieve the stored result, complete the separate activation process where required, and inspect the rendered destination.
Keep the AEM write at the end of the sequence. Insight generation, drafting, and validation can fail safely. A write changes shared production content and therefore needs the strongest preconditions.

Choose Create or Update without multiplying content

Update when the canonical content object already exists

Use Update when the existing fragment represents the same entity, intent, locale, and reusable content unit. The gap should be field-level: an incomplete answer, stale description, missing supporting detail, or another change that belongs inside the established object.

Send a patch containing only approved changes. Replacing the full fragment increases the chance of losing fields the agent was never meant to edit. Retrieve again immediately before the write if another editor or workflow could have changed the target since the first read. If the integration exposes a revision or version value, use it to reject a write based on stale state.

Create only when a genuinely new reusable object is needed

Use Create when the required content has no canonical fragment and the new object has a defined model, destination, owner, locale, and lifecycle. A new topic alone is not enough. The content also needs a known consumer: a page component, application, API response, campaign experience, or another delivery path that will use the fragment.

The common failure is creating a new fragment for every visibility finding. That produces near-duplicates, splits ownership, and makes later updates ambiguous. Search first, inspect likely matches, and stop for review when more than one candidate could be canonical. A failed or inconclusive search should never silently authorize creation.

Retries need the same discipline. Record a unique run identifier and the intended target so a retried workflow cannot create the same fragment twice. For updates, record the retrieved state or revision so a retry cannot overwrite a more recent edit without detection.

Protect content quality, structured data, and production state

Model the information that answer systems need

AEM automation works best when important information has an explicit field instead of being buried in one large rich-text block. Depending on your content model, useful fields can include a concise answer, supporting explanation, named entity, approved evidence URL, audience or locale, review status, owner, and review date. These are design recommendations, not fields that Profound creates for you.

Keep factual generation constrained to approved evidence. An AI visibility finding can identify a missing answer or weak topic representation, but it does not establish the underlying product, legal, pricing, or policy facts. The workflow should stop when the supplied evidence cannot support the proposed claim.

Do not turn the fragment into a bag of repeated search phrases. Write the direct answer a person needs, use consistent entity names, preserve necessary qualifications, and add supporting detail only where it improves understanding. The goal is a clearer canonical answer, not a visible record of every query variant that triggered the workflow.

Structured content and structured data are related, but they are not interchangeable. Updating a Content Fragment does not automatically update the JSON-LD emitted by the rendered page unless your delivery layer maps those fragment fields into the markup. Verify the visible HTML and the resulting JSON-LD separately. If they describe the same entity or claim, they should remain aligned after the update.

Put operational controls around every write

Treat generated content as untrusted input until it passes your rules. The AEM nodes provide the content operations; your surrounding workflow still needs access, validation, review, recovery, and publication controls.
- Use an AEM identity with the least access needed for the approved path and model.
- Separate development or test targets from production targets, and prove the workflow against representative non-production fragments first.
- Allowlist paths, models, locales, and writable fields. Do not rely on prompt wording as the only permission boundary.
- Prefer field-level patches to full-object replacement.
- Re-fetch the fragment before Update and stop if the current state no longer matches the reviewed state.
- Preserve a recoverable prior version or snapshot before changing production content.
- Keep content writing separate from activation or publication so each can have its own approval rule.
- Log the evidence, retrieved target, proposed diff, validation outcome, write result, and final delivery state.
The announced AEM action set covers List, Search, Get, Create, and Update; it does not name Delete. That reduces one obvious failure path, but Update can still remove or replace valuable field content. Recovery and diff review remain necessary.

Measure delivery separately from visibility

A successful node execution means the requested AEM operation completed. It does not prove that the correct experience rendered, that a search system discovered the change, or that an AI answer will use it.

Track the workflow in three layers. First, confirm operational correctness: one target, the intended action, valid fields, and an approved diff. Second, confirm delivery: the stored fragment, activation state, rendered page or endpoint, links, metadata, and JSON-LD. Third, observe discovery outcomes through your normal crawling, indexing, search, and AI visibility monitoring. Keep those layers separate so a rendering failure is not mistaken for a content-strategy failure.

Changes in AI answers are especially difficult to attribute to one edit. Record what changed and where, but do not treat a later answer difference as proof that the fragment update caused it. The defensible result is a verified content improvement and a traceable delivery path; visibility remains an outcome to monitor.

Key takeaways
- Profound Agents can List, Search, Get, Create, and Update AEM Content Fragments, which removes a manual CMS handoff from an approved optimization workflow.
- Get before Update, and require one unambiguous target. No match or multiple matches should stop the write.
- Use Update for an existing canonical object and Create only for a defined new content unit with a known consumer and owner.
- Limit every run by path, model, locale, operation, and writable field. Review the exact diff rather than the new copy in isolation.
- Verify AEM storage, publication, rendering, and JSON-LD separately. A completed content operation is not the same as a live or discoverable change.
Start with one low-risk fragment family and one field-level optimization pattern. Write the contract, test the stop conditions, require diff approval, and trace the result through rendering and structured data. Expand the scope only when repeated runs select the right object, preserve untouched fields, and produce a recoverable audit trail.

References
- Try Profound Blog – Introducing Adobe Experience Manager Nodes in Profound Agents
August 28, 2026

PPC Automation for Better Leads: A Practical Framework

Your PPC account can hit its cost-per-lead target and still leave sales with little usable pipeline. When the bidding system is rewarded for a form fill, it will find people who are likely to fill forms. It cannot prefer future customers unless you return that distinction as data.

The fix does not begin with another bid adjustment or a tighter keyword list. You need to identify the business constraint, choose a conversion event that represents progress toward revenue, and then give automation enough room to find more of that outcome. This framework shows you how to do that without treating every unusual query or expensive lead as a failure.

Key takeaways

Decide whether the immediate constraint is insufficient lead volume or insufficient lead quality. They require different optimization signals and campaign levers.
Use the deepest conversion event that occurs often and consistently enough to guide bidding. That may be a qualified lead or opportunity rather than a closed customer.
Connect CRM outcomes to your advertising platforms. Form submissions alone do not tell an algorithm which people became valuable.
Broad match, automated audiences, and Smart Bidding need reliable conversion data, explicit exclusions, and clear landing pages.
Judge performance with cost per qualified lead, cost per opportunity, customer acquisition cost, and revenue. CPL is only an early-funnel diagnostic.

Pick the business constraint before the campaign metric

The useful question is not whether you want more leads or better leads. Every business wants both. The question is which constraint is preventing growth right now. Lead quantity and lead quality are different growth objectives with different inputs, not opposing philosophies.

Business condition	Primary objective	First PPC lever	Main risk
Sales has unused capacity and too few leads	Volume	Expand eligible demand and remove unnecessary conversion friction	Cheap form fills can crowd out valuable prospects if every submission is treated equally
Sales is overwhelmed by poor-fit inquiries	Quality	Optimize toward a qualified lead or opportunity	Lead count may fall and CPL may rise even while pipeline economics improve
A new market or offer has little outcome data	Volume and learning	Broaden reach while building consistent CRM classifications	A sparse customer signal may give automation too little information
Lead volume is healthy but revenue is weak	Quality and value	Return deeper outcomes and, where defensible, their business values	The problem may sit in qualification, the offer, or the sales handoff rather than targeting

CPL should not make this decision for you. A $30 lead that never becomes a customer is not inherently better than a $100 lead that regularly closes. The useful denominator is the business outcome you are trying to produce.

Cost per qualified lead equals media spend divided by qualified leads.
Cost per opportunity equals media spend divided by accepted opportunities.
Customer acquisition cost becomes useful when customer records can be matched reliably to acquisition.
ROAS is meaningful only when the revenue or conversion values sent back to the platform reflect real economics.

Write the objective as an operating sentence: “Paid media will optimize for [lifecycle event] because [business constraint], while [downstream metric] remains the guardrail.” That forces marketing, sales, and finance to agree on the event and the trade-off before the algorithm starts making it for them.

Also separate a media-quality problem from a sales-process problem. If leads meet documented fit criteria but fail to become opportunities, inspect routing, follow-up, sales acceptance, and the offer before narrowing targeting. Automation cannot correct a broken handoff by finding fewer people.

Feed CRM outcomes back into the bidding system

A circular flow connects an advertising engine, a qualification funnel, and a customer database, with glowing outcome signals returning to the advertising system.

Imagine that an ad platform records 1,000 form submissions while the CRM shows 300 qualified leads, 75 opportunities, and 20 customers. If only the form event returns to the ad platform, the system cannot distinguish those 20 customers from everyone else. It learns to reproduce the easiest visible action instead.

Your feedback loop should give each important lifecycle stage an unambiguous meaning:

Conversion event	What it proves	When it can guide bidding
Form submission	A person completed the initial action	When volume is the immediate goal or deeper outcomes are not yet recorded consistently
Qualified lead	The record meets written fit or eligibility rules	When opportunities and customers are too sparse but lead quality can be classified reliably
Opportunity	Sales accepted the lead into an active commercial process	When opportunity creation occurs often enough and follows a consistent definition
Customer or revenue	The acquisition produced a closed outcome and, where available, economic value	When the event is frequent, timely, and matched accurately enough for optimization

Build the connection in this order:

Define the stages. A qualified lead cannot mean “sales liked it.” Write the fit and eligibility rules, who owns the classification, and what causes a record to leave that stage.
Preserve the acquisition link. Carry the identifiers needed to connect the ad interaction, form submission, and CRM record under your consent and privacy requirements. A lifecycle event that cannot be tied back to acquisition is useful for reporting but not for campaign learning.
Clean the event stream. Deduplicate records, keep test submissions and spam out of optimization, and distinguish hard disqualification from an unsuccessful contact attempt.
Return downstream events. Send the selected lifecycle milestones to the relevant advertising platform with consistent names, timestamps, and values where those values are economically defensible.
Choose one primary optimization event. Keep shallower stages available for diagnosis, but do not reward every stage as though it represents the same result.
Reconcile platform and CRM reporting. Investigate missing matches, duplicate events, status reversals, and unexplained shifts before changing bids or targeting.

Google Ads supports qualified-lead and converted-lead goals, while Meta can receive down-funnel CRM outcomes through the Conversions API. These mechanisms close the visibility gap, but neither can repair a vague qualification rule. If sales changes the meaning of “qualified” from person to person, the machine receives inconsistent training data.

Choose the deepest event that still supplies a recurring, timely signal. If you generate only a handful of customers in a typical month, customer-only optimization may not provide enough learning data. Move one meaningful stage higher, such as opportunity or qualified lead. Do not retreat all the way to form submissions unless that is the only dependable event.

Conversion values deserve the same discipline. Use value-based bidding only when the values reflect expected revenue, margin, or another agreed business measure. Arbitrary points can look sophisticated while teaching the system to favor the wrong outcome.

Give automation room, but keep business guardrails

Keyword precision is no longer the control system it once was. Google required close variants for exact match in 2014, and automated products such as Performance Max and AI Max can expose advertisers to auctions they did not deliberately choose one by one. Trying to recreate perfect query-level control leaves you fighting the platform instead of shaping its objective.

Modern broad match can use context beyond the literal keyword, including previous searches and landing-page context. That makes it more capable of finding intent, but also more dependent on the accuracy of your conversion data and the clarity of your site.

Use an expansion sequence that protects the signal:

Confirm that the chosen conversion event reaches the platform accurately and excludes invalid records.
Expand keyword coverage or test broad match with automated bidding while maintaining negatives for clearly irrelevant or impossible intent.
Broaden geography or paid-social audiences only where the business can actually serve the resulting demand.
Add inventory such as Display, Demand Gen, YouTube, or other video placements when incremental reach is part of the objective.
Evaluate each expansion through qualified leads, opportunities, and customers rather than form volume alone.

The guardrails should encode business facts, not personal discomfort with an unusual search term:

Negative keywords and exclusions: Block structurally irrelevant demand, prohibited locations, services you do not sell, and patterns that repeatedly produce invalid records. Do not exclude a query solely because its wording looks odd if it contributes profitable downstream outcomes.
Clear conversion configuration: Make sure the bidding strategy is optimizing for the intended lifecycle event rather than an easier secondary action.
Landing-page specificity: Give people and matching systems a precise description of the offer, audience, service area, and next step.
Separate brand reporting: Keep branded demand distinct from prospecting. Automated campaign types and competitive bidding can blur that boundary, and revenue attributed to your own brand searches does not by itself show how much new demand the campaign created.
Downstream segmentation: Compare campaign, network, geography, audience, and query themes using qualified and opportunity outcomes. A segment with a low CPL can still be your most expensive source of pipeline.

Smart Bidding replaces thousands of manual bid decisions with auction-level choices guided by a target such as CPA or ROAS. That is useful operational leverage, not strategic judgment. A system can efficiently minimize the cost of the wrong conversion just as easily as the right one.

Review strange queries as patterns, not isolated screenshots. One unconventional search term that produces qualified opportunities may reflect context you cannot see in the term itself. A recurring cluster of irrelevant searches with no downstream value is evidence for a negative, a message change, or a tighter business boundary.

Make your ads, forms, and landing pages qualify together

Three connected panels representing an ad, a landing page, and a form progressively filter prospect tokens before they reach a sales representative.

When lead quality falls, adding form fields is an easy reaction. It also confuses friction with qualification. A longer form can reduce submissions without making the remaining people a better fit.

Your ad should help the right person recognize the offer and the wrong person opt out. A generic message such as “Get started today” does almost no filtering. Stronger qualification comes from saying what the offer is, who it serves, which real boundaries apply, and what happens after the click.

Name the use case. Do not make a buyer infer whether the offer concerns a product demo, a quote, an application, a consultation, or an informational download.
State genuine boundaries. If location, business type, eligibility, or service scope determines fit, make that information visible before the form.
Explain the next step. A person expecting instant access behaves differently from someone knowingly requesting contact from sales.
Reflect rejection data. If a recurring poor-fit group responds to the ad, revise the message that is inviting it rather than relying on sales to filter it later.

Apply the same standard to the form. Every question should support routing, qualification, follow-up, or measurement. If nobody uses an answer, remove the question. Keep discovery questions that sales can ask later out of the acquisition gate unless the answer is genuinely required to determine fit.

Do not label every unreachable lead as low quality. “Could not contact,” “not eligible,” “wrong service,” “outside service area,” “duplicate,” and “spam” describe different failures. Combining them into one bad-lead bucket hides the corrective action and corrupts the optimization signal.

Map each rejection reason to the lever that can plausibly fix it:

Wrong service or product: Clarify the ad and landing page, separate offers, and exclude consistently irrelevant search themes.
Outside the service area: Correct location settings and state the coverage area plainly.
Wrong buyer type: Use audience-specific language and route distinct buyer groups through appropriate paths.
Spam or duplicates: Repair validation and deduplication. Narrower audience targeting is not a substitute for data hygiene.
Qualified but never accepted as an opportunity: Inspect the qualification definition, sales handoff, offer, and follow-up process before blaming media.

The landing page completes the loop. It must confirm the promise in the ad, describe the intended customer, and make the conversion’s meaning unmistakable. This improves human self-selection and supplies the contextual information that modern matching can use.

For a volume objective, shorter forms, broader audiences, more creative variations, and additional conversion opportunities can remove unnecessary barriers. For a quality objective, start with better outcome data and clearer positioning. Making the form harder to complete should not be your proxy for teaching the platform what a valuable lead looks like.

Judge automation with mature, downstream cohorts

The funnel does not end at the thank-you page. Track the full progression from impression to click, lead, qualified lead, opportunity, and customer. Each transition tells you where performance changed and which team can act on it.

Your working dashboard should include:

Spend, clicks, form submissions, and CPL for acquisition diagnostics.
Qualified leads, lead-to-qualified rate, and cost per qualified lead.
Opportunities, qualified-to-opportunity rate, and cost per opportunity.
Customers, opportunity-to-customer rate, and customer acquisition cost.
Revenue or another defensible value measure, plus ROAS where attribution is reliable.
Rejection reasons by campaign, audience, location, query theme, creative, and landing page.

Read these metrics by acquisition cohort after that cohort has had enough time to move through your normal sales cycle. Recent leads will naturally have fewer opportunities and customers than mature leads. Comparing them without accounting for that delay can make a healthy campaign look weak or a deteriorating campaign look temporarily efficient.

Use the pattern in the funnel to choose the next action:

Lead volume rises, qualification rate falls, and cost per qualified lead worsens: Automation is probably scaling the easy signal. Move the optimization event deeper, correct exclusions, or strengthen qualification messaging.
CPL rises while qualification rate improves and cost per opportunity falls: The campaign may be working better. Do not reverse it merely to restore a cheaper form fill.
Qualified-lead volume holds but opportunity creation falls: Revisit the qualification definition and sales-acceptance process. The label may no longer predict commercial value.
Opportunities remain healthy but customer or revenue performance weakens: Inspect value assumptions, offer fit, close rates, and the sales process. Targeting may not be the root cause.
The deepest event appears only sporadically: Step up to a more frequent meaningful stage while keeping the final outcome in reporting.
Platform metrics look strong while sales reports poor quality: Require structured rejection reasons and reconcile the records. Anecdotes can flag a problem, but they cannot train an algorithm or locate the failure.

Your next move should be concrete: take a mature group of paid leads, assign consistent lifecycle stages and rejection reasons, then calculate cost per qualified lead and cost per opportunity. Select the deepest dependable event as the bidding goal before expanding match types, audiences, or inventory. Once the platform can see the same definition of success as the business, automation has something useful to optimize.

References

August 28, 2026

YouTube Ad Creative and DV360 Changes to Make by October

Your YouTube campaign can have a sound bid strategy and still underperform because the ad was built for another surface. At the same time, a promising creative refresh can stall before delivery if the Display & Video 360 integration behind it is not ready for October’s unversioned platform changes.

Treat this as one operating problem with two workstreams. Improve the message people see and hear, then verify that your API and Structured Data File workflows can still create, update and protect the campaign. Here is the sequence we would use.

Key takeaways

For Demand Gen in-stream skippable ads in the United States, July 2026 data associated human voice with 12% higher conversions on average, text overlays with 3% higher conversions and visible branding in the first five seconds with 4% higher conversions. These are test priorities, not guaranteed lifts.
Build image ads for the YouTube feed: use high-resolution, full-bleed imagery, include people when appropriate and remove black bars, excessive empty space and oversized logos.
Do not bake fake buttons or arrows into an image. They compete with YouTube’s functional call to action and can leave the viewer unsure about what is actually clickable.
On Oct. 1, DV360 API and Line Item Structured Data File workflows lose specified digital-content-label exclusions and most sensitive-category exclusion options.
On Oct. 12, YouTube responsive ad creation and updates require a business name and logo when those defaults are not already assigned to the parent advertiser.

Start with voice, early branding and useful on-screen text

Creative is not the decorative layer that you address after bidding and targeting. Nielsen attributed 49% of campaign ROI to creative, while Ekimetrics found that improving creative could more than double YouTube ROI. Those aggregate findings do not forecast what your account will gain, but they do justify giving creative testing the same operational attention as media settings.

The most actionable benchmarks are narrower. They apply to Demand Gen in-stream skippable ads, use U.S. data from July 2026 and describe associations rather than proof that an isolated element caused the result. That scope matters when you decide what to test and how confidently to interpret it.

Creative element	Observed conversion association	First controlled test
Human voice	12% higher on average	Compare a voiced cut with a closely matched cut that has no human voice.
Supers or text overlay	3% higher on average	Add concise on-screen wording to the same core edit and keep the offer and call to action unchanged.
Brand visible in the first five seconds	4% higher on average	Compare immediate visual brand identification with a later brand reveal.

Do not add those percentages together and turn the result into a forecast. The elements can interact, and campaigns that use them may differ in other important ways. Use the figures to determine test order: if you have enough traffic for only one new comparison, human voice is the most defensible place to start because it had the largest reported association.

Give the voice a real job. It can state the viewer’s problem, establish the offer or make the next action clear. A voice that merely reads every word on screen adds sound without improving the message. Keep supers equally disciplined: reinforce the key point rather than turning the frame into a transcript.

Early branding also needs restraint. The goal is to make the advertiser identifiable within five seconds, not to cover the opening with a logo that delays the reason to keep watching. Put the brand into the story while the viewer is still deciding whether to skip.

For a useful test, hold the audience, offer, bid strategy, call to action and landing page as steady as your campaign setup permits. Change one creative factor at a time. If your conversion volume cannot support several cells at once, run the comparisons sequentially instead of launching a test that never produces a clear decision.

Judge the result against the conversion action that matters to the campaign. A click-through improvement is not automatically a conversion improvement. Also, do not assume that benchmarks from U.S. Demand Gen in-stream skippable inventory transfer unchanged to Shorts, other formats or other markets. Those are separate questions for your account to answer.

Make image assets belong in the YouTube feed

An image can be polished in a design file and still look broken when placed in a YouTube feed. The common failure is not low production value. It is a layout that carries the visual habits of a banner, presentation slide or another ad platform into a surface where people expect immersive imagery.

For YouTube image ads, visible people and people interacting with products tend to outperform assets without human presence in Google’s platform observations. Human presence should still make sense for the product and message; inserting an unrelated face is not a substitute for a coherent concept.

Fill the available frame. Start with high-resolution, full-bleed photography or lifestyle imagery rather than an image floating inside a large solid canvas.
Show use, not just inventory. When appropriate, let a person hold, wear, operate or otherwise interact with the product so the viewer can understand its role quickly.
Keep the logo proportional. The brand should be identifiable without making an oversized logo the main visual event.
Remove structural clutter. Black bars and large empty solid areas can make the asset feel fragmented or incorrectly formatted.
Delete fake interface elements. A button, play control or arrow drawn into the image is not functional. Let YouTube’s actual call-to-action control handle the interaction.

Fake controls create two competing instruction systems. The platform presents a real action, while the picture implies another one that does nothing. That forces the viewer to determine which visual element is interactive instead of understanding the offer. If an arrow is necessary to make the call to action discoverable, the composition or message probably needs another pass.

Review the rendered asset in its intended placement, not only at full size on a designer’s canvas. Ask whether it reads as one complete image, whether the important person or product survives the crop, whether the brand remains recognizable and whether there is exactly one obvious functional path forward. This preview is also where black bars, oversized marks and deceptive button shapes become easiest to catch.

Native fit does not mean disguising an advertisement. It means using the visual language of the surface while keeping the advertiser and offer clear. A feed-compatible image earns attention through relevance and composition, not through an imitation of YouTube’s controls.

Prepare DV360 automation for the October deadlines

A better asset cannot improve results if the integration that manages it stops working. The October rollout contains three unversioned changes across the Display & Video 360 API and Structured Data Files. Do not assume an older client or a delayed API-version migration will preserve the previous behavior.

Oct. 1: specified exclusion controls are removed

Starting Oct. 1, advertisers will no longer be able to use API targeting to exclude specific digital content labels. The change affects available TARGETING_TYPE_DIGITAL_CONTENT_LABEL_EXCLUSION options and valid values in the Digital Content Labels - Exclude column of Line Item Structured Data Files.

Most sensitive-category exclusions are also being removed from targeting on that date. Audit any workflow that uses TARGETING_TYPE_SENSITIVE_CATEGORY_EXCLUSION, as well as the Brand Safety Sensitivity Setting and Brand Safety Custom Settings columns in Line Item Structured Data Files.

This is a change to available controls, not a reason to quietly weaken your brand-safety policy. Do not simply delete fields until an error disappears. First identify which business rule each field was implementing, who owns that rule and what the approved workflow should be once that targeting option is unavailable.

Do not guess how every existing line item will display or behave after the change. Inventory the affected line items and validate the actual transition in a controlled workflow. The important distinction is between authoring a new setting, updating an existing line item and observing a previously configured value; each path deserves an explicit check.

Oct. 12: responsive ads need business identity assets

Starting Oct. 12, developers creating or updating YouTube responsive ads must provide a business name and logo when default values are not already assigned to the parent advertiser. The requirement also applies to ads uploaded through Ad Structured Data Files.

That parent-advertiser condition gives you a clean preflight decision. If approved defaults exist, verify that every relevant workflow can use them. If they do not, make the business name and logo required inputs before an ad reaches the create, update or upload step. Do not wait for a production job to discover that the identity assets are missing.

Your technical audit should cover these exact paths:

Search code, configuration files and Line Item Structured Data File templates for TARGETING_TYPE_DIGITAL_CONTENT_LABEL_EXCLUSION, TARGETING_TYPE_SENSITIVE_CATEGORY_EXCLUSION and the affected column names.
List every scheduled job, internal tool and third-party workflow that creates or updates YouTube responsive ads through the API.
List every process that uploads Line Item or Ad Structured Data Files. Treat the two file types separately because the exclusion and identity changes affect different operations.
Inspect each parent advertiser used by those workflows and record whether an approved default business name and logo are already assigned.
Add a preflight check that blocks responsive-ad submission when neither advertiser defaults nor required identity inputs are available.
Have the brand-safety owner approve any operational change caused by the lost exclusion options, then test create, update and file-upload paths before their respective deadlines.

Record which test covers which deadline. A successful responsive-ad creation test does not prove that an exclusion workflow is ready, and a clean Line Item Structured Data File does not prove that an Ad Structured Data File contains the required identity. Separating those assertions will make a failure much easier to locate.

Run one joined creative-and-delivery sprint

Creative production and delivery engineering often sit in different queues, but the campaign depends on both. A new ad trapped behind a failed update request creates no learning. A perfectly updated integration serving weak recycled assets only automates the wrong input.

Use this order to turn the work into a test you can trust:

Clear the deadline risk. Open technical tickets for the Oct. 1 exclusion changes and the Oct. 12 identity requirement. Assign owners before asking the creative team to produce a large new batch.
Freeze a useful control. Preserve the current offer, landing page, audience and conversion action so the next result can be interpreted as a creative comparison.
Create focused video variants. Build a human-voice version, an early-brand version and a concise-text-overlay version. Keep the underlying proposition as consistent as possible.
Rebuild image assets for the feed. Use full-bleed imagery, meaningful human presence and one clear composition. Remove fake buttons, arrows, black bars and excess empty space.
Validate delivery before launch. Exercise the API create and update paths, the relevant Structured Data File uploads and the business-identity fallback. Do not mix a delivery defect into a creative performance test.
Label the change in reporting. Use variant names that identify the factor being tested. When conversions move, you should be able to connect the result to voice, branding, text or image treatment without reopening the design files.

If capacity is tight, prioritize the integration work first because its dates are fixed. Then test human voice, which had the largest reported conversion association, followed by early branding and text overlays. Feed-image cleanup can run alongside those video edits because it addresses a different asset type.

Open your highest-spend YouTube ad and its parent advertiser record side by side. Check whether the ad uses a human voice, identifies the brand within five seconds and gives on-screen text a clear purpose. Then confirm the advertiser’s default business name and logo and search your automation for the affected exclusion identifiers. You will leave that session with one defined creative experiment and one concrete technical readiness list, both in time for October.

References

August 28, 2026

Human-Led AI Workflows for SEO: A Practical System

You don’t need to choose between banning AI from SEO and letting an agent run your site. The useful middle is a workflow in which AI accelerates analysis and production while a person remains accountable for the decisions that can affect rankings, crawlability, brand trust, and measurement.

Your goal is not to put a human approval step at the end of an automated content factory. It is to place human judgment at the few points where a plausible answer can become an expensive mistake: choosing the page, defining its unique contribution, validating its evidence, approving the technical change, and interpreting the result.

Human-led means retaining decision authority, not doing everything manually

AI is genuinely useful for clustering keywords by intent, identifying content gaps, analysing pages, and producing first-pass outlines. Those tasks compress a large amount of reading and organisation. They do not require the model to decide what your site should publish or change.

The boundary should be based on authority. Let AI transform information, expose patterns, draft options, and run checks. Keep a person responsible for choosing the objective, accepting the evidence, resolving conflicts, approving live changes, and deciding whether an experiment worked.

That distinction matters because fluency is not reliability. A model can produce a tidy keyword map, persuasive rationale, polished page, and confident recommendation even when the underlying choice is wrong. It may not know that a proposed URL conflicts with an existing page, that a claim lacks support, or that a template renders essential content only after client-side JavaScript runs.

Google’s stated position is that using AI to produce content is not inherently against its guidelines when the result is helpful and made for people. The operational risk is therefore not the presence of AI. It is publishing low-value or technically unsound work because nobody tested whether the output deserved to exist.

Key takeaways

Use AI to analyse evidence and generate options; do not let it define success or approve its own work.
Separate opportunity selection, research, briefing, drafting, technical validation, publication, and measurement into distinct gates.
Require a unique contribution before drafting. A new keyword target is not, by itself, a reason to create a new URL.
Route every live change through a reviewable diff, a validation checklist, and a rollback plan.
Measure one declared hypothesis against the pages and metric the change could actually affect.

Turn the workflow into gates with visible pass conditions

A human reviewer inspects five abstract SEO workflow stages separated by approval gates on a studio table.

A single prompt that asks for research, strategy, a draft, optimisation, and publication collapses several different decisions into one answer. By the time you see the finished page, the model has already assumed the search intent, selected the format, decided whether to create or update a URL, filled evidence gaps, and judged its own quality.

Break that chain apart. Each stage should produce an artifact that the next reviewer can inspect. A pass condition should be observable rather than subjective: not good quality, but target intent is named, competing URLs were checked, every factual claim has support, and the proposed contribution is absent from the comparison set.

Stage	AI contribution	Human decision	Required artifact
1. Opportunity	Summarise query, page, conversion, and competitive data; surface patterns and anomalies.	Choose the business and user problem worth solving.	A work order with the target audience, objective, metric, scope, and exclusions.
2. Intent and URL mapping	Cluster queries, describe likely intents, and identify potentially competing pages.	Decide whether to create, consolidate, refresh, redirect, or stop.	A query-to-URL map that names the current owner and proposed owner of each intent.
3. Evidence	Organise supplied data, first-hand notes, examples, and references; flag unsupported claims.	Confirm provenance and decide what may be published.	An evidence pack in which every input has an owner or traceable origin.
4. Information gain	Compare the planned coverage with ranking pages and identify repetition or gaps.	Determine whether the page adds a useful fact, method, example, tool, dataset, or point of view.	A one-sentence unique-contribution statement plus the evidence needed to deliver it.
5. Brief and draft	Build an outline, draft sections, suggest internal links, and mark open questions.	Correct the framing, verify claims, remove filler, and protect the brand’s position.	A draft with unresolved questions clearly marked rather than silently completed.
6. Technical preflight	Run repeatable checks on metadata, links, structured data, indexation directives, and rendered content.	Inspect the actual change and resolve conflicts or failures.	A pass-or-fail report tied to the exact URL, build, or commit being reviewed.
7. Release	Prepare a diff, change log, test instructions, and rollback steps.	Approve the specific version that will go live.	A recorded sign-off and a recoverable previous state.
8. Measurement	Collect the declared metric and summarise what changed.	Judge causality, retain or reverse the change, and select the next test.	An append-only experiment record, including inconclusive results.

The information-gain gate belongs before the draft. If the only proposed difference is a longer word count, a new title, or rearranged coverage, stop. Ask for first-hand evidence, proprietary data, a concrete workflow, a useful tool, or a sharper answer to a neglected part of the intent. A gated system prevents average ideas from becoming finished pages merely because drafting is cheap.

A useful gate prompt is narrow: Review this opportunity as an SEO decision, not as a writing task. Using the target query, existing URL map, ranking-page notes, and evidence pack, return the dominant intent, the URL that should own it, any cannibalisation risk, the unique contribution, missing evidence, and one verdict: pass, revise, or stop. Do not fill evidence gaps with assumptions.

The verdict remains advice. The human reviewer should be able to explain why the page should exist without repeating the model’s wording. If you cannot state the intended reader, unmet need, unique contribution, and correct URL in plain language, the opportunity has not cleared the gate.

Keep AI away from unreviewed changes to the live site

A human operator reviews abstract page and code modules in a staging area before allowing them into a protected live website environment.

The most important permission boundary sits between proposing a change and applying it. Read access to analytics, crawls, keyword sets, page inventories, and content repositories can create enormous leverage. Unrestricted write access to a CMS, routing configuration, templates, redirects, canonical tags, robots directives, structured data, or measurement code creates a different risk class.

A live-site failure shows why. An AI system asked to recommend keywords and build the necessary pages produced two new URLs that largely copied the homepage while changing the title tag and H1. After six months, the two dedicated pages had zero impressions and zero clicks in Google Search Console, while the homepage continued to receive the relevant queries. This is one site’s result, not a universal performance benchmark. The reusable lesson is the failure mode: the system satisfied the surface instruction to create targeted pages without giving either page a distinct purpose.

The same cloning pattern appeared on a separate project, where a batch of keyword-targeted pages copied the homepage and changed little beyond their titles. That is what a human URL-mapping gate should catch before a draft exists. Microsoft has also confirmed that Bing’s models can group near-duplicate URLs and select an unintended representative, so duplication can obscure which page should appear in conventional search and AI-generated answers.

Use a change packet whenever AI proposes work that could reach production. The packet should contain:

Exact scope: every URL, template, file, rule, and structured-data type affected.
Before-and-after diff: the actual text or configuration change, not a prose summary.
Purpose: the user problem, target intent, and expected mechanism of improvement.
Evidence: the data and approved claims used to justify the change.
Conflict check: existing URLs, keywords, canonicals, redirects, and templates that could overlap.
Validation plan: what will be checked in staging and again after release.
Rollback: how to restore the previous state without reconstructing it from memory.
Measurement: the page-specific metric and the condition that would count as a valid result.

Then perform the preflight against the built page, not the intended page. Confirm that the title, H1, main content, internal links, canonical URL, indexation directives, and structured data are present in the delivered output. Check that structured data describes visible content and approved claims. Inspect server-returned HTML as well as the browser-rendered page when essential content depends on JavaScript.

That last check matters beyond Google. One practitioner’s measurement found ClaudeBot downloaded a JavaScript bundle in 24% of its requests but did not execute it. Treat that as one observed implementation behaviour, not a guaranteed rate for every site or bot. The practical response is still sound: do not assume a page is machine-readable because it looks complete in your browser.

For routine work, let the system create a CMS draft, branch, pull request, or staging build. Require a named person to approve URL creation or deletion, redirects, canonical changes, indexation controls, template-wide edits, bulk internal links, measurement code, and publication. AI can produce the checklist and flag deviations; it should not be the sole reviewer of its own output.

Measure a declared hypothesis instead of rewarding activity

Human control is also necessary after publication. An automated report can find a favourable movement and attach it to the latest task, even when the changed pages could not have caused that movement. That creates a learning system that rewards coincidence.

Define the experiment before the change. Use one sentence: If we make this change to these pages, we expect this metric to move because this user or crawler problem will be reduced. Name the affected URLs, the baseline, the primary metric, any guardrail metric, the review window, and the evidence that would make the outcome valid. Choose the review window based on the site’s crawl patterns, traffic, and decision cycle rather than inventing a universal deadline.

Keep each run narrow enough to interpret. A bounded agent can read the roadmap, state file, and prior log, then recommend one justified action. It can also recommend no change when the evidence is weak. If you permit execution, constrain it to a reviewable draft or branch unless the action has already been proven safe, is reversible, and falls inside an explicitly approved class.

The experiment log should record:

the hypothesis and why the action should affect the selected metric;
the exact pages and elements changed;
the baseline and date range used;
the model, instructions, evidence pack, and workflow version involved;
the human reviewer and approval decision;
the release date and any confounding changes;
the observed result, including negative and inconclusive outcomes;
the decision to retain, revise, reverse, or run a follow-up test.

Use a strict causal rule: a metric movement does not count if the shipped change did not touch the pages or mechanism that metric represents. In one autonomous run, average position improved from 48 to 39, but the result was logged as inconclusive because the change affected pages outside the measured target set. That is the behaviour you want from an AI-assisted testing system. Its job is to preserve the truth of the experiment, not to manufacture wins.

Do not hide rejected recommendations or failed tests. They reveal which inputs are missing, which instructions are ambiguous, and which permission boundaries need tightening. An append-only log turns human review from an approval ritual into operational memory.

Install a minimum viable workflow before expanding automation

You do not need to redesign the whole SEO operation at once. Start with one recurring unit of work, such as content briefs, refresh recommendations, internal-link opportunities, or schema proposals. Pick a task that happens often enough to expose patterns but can still be reviewed carefully.

Write the work order. Name the user problem, business objective, primary metric, allowed inputs, prohibited actions, and person accountable for approval.
Disable direct publication. Route output to a draft, ticket, branch, or staging environment. Preserve the original state.
Create three reusable templates. Use an evidence pack for inputs, an acceptance checklist for review, and an experiment log for outcomes.
Pilot a small batch. Ten items can be enough to expose recurring rejection reasons without turning the pilot into a production commitment. This is a practical batch size, not a performance threshold.
Classify every intervention. Record whether the reviewer corrected intent, URL choice, evidence, factual accuracy, duplication, brand framing, technical implementation, or measurement.
Improve the system at the earliest failed gate. If reviewers repeatedly catch duplicate intent at final QA, move the URL-map check ahead of drafting. Do not solve an upstream decision problem with more downstream editing.
Expand one permission at a time. Grant a new capability only when its inputs, output, reviewer, validation, and rollback path are explicit.

Before any item goes live, ask the reviewer five questions: Why should this page or change exist? What evidence supports it? What exactly will change? What could it conflict with or break? How will we know whether it worked? A missing answer is a stop signal, not an invitation for the model to improvise.

The next time your team asks to automate more SEO, automate the collection, comparison, drafting, checking, and documentation first. Keep the decision rights visible. Once the workflow can show its evidence, its diff, its reviewer, and its result, you can increase speed without surrendering control of what your site becomes.

References

August 28, 2026

2026 AI Search Optimization Agencies by Sector: Buyer’s Guide

If your shortlist looks identical for a medical network, a cybersecurity vendor, and a roofing franchise, your brief is too generic. AI search may appear as one channel in a dashboard, but the work behind a recommendation changes with the evidence, entities, regulations, locations, and buying decisions in your sector.

Use this guide to narrow the 2026 agency market by sector and operating model, then pressure-test each candidate at the prompt, citation, governance, and pipeline levels. You are not looking for the agency with the loudest GEO label. You are looking for one that understands what your buyers ask, what an AI system must trust, and what your organization can responsibly publish.

The short answer: sector fit beats a universal ranking

The recurring cross-sector candidates are First Page Sage, Focus Digital, and Driven Metrics. Genevate also appears prominently in finance, medical, and general B2B. That recurrence makes them reasonable starting points, but it does not make them interchangeable. Their operating models range from full-service content and lead generation to external authority building, lean execution, and analytics-heavy performance management.

Key takeaways

For finance and medical organizations, make domain review, claims governance, and compliance-sensitive writing pass-or-fail requirements. Content volume cannot compensate for an approval process that does not work.
For cybersecurity, test whether the agency can explain products, requirements, integrations, and technical tradeoffs at the depth buyers use to form a shortlist.
For B2B, insist on a measurement path from AI visibility to qualified opportunities or pipeline. Mentions without commercial context are not enough.
For local businesses, require service-and-location coverage, consistent business facts, and reporting segmented by market. A national content playbook is not a local GEO strategy.
If an autonomous agent may compare providers or take an action for the user, add agentic search optimization to the brief. GEO visibility alone does not prove that an agent will select you.
Use published rankings for discovery, then validate sector work, live AI outputs, client scope, capacity, and attribution yourself.

The following table is a market map, not a substitute for due diligence. It shows which agencies deserve inspection for each sector and the operating differences that should drive your first round of questions.

Sector	Agencies to inspect	What should decide the fit
Financial services and fintech	First Page Sage, Genevate, Driven Metrics, Focus Digital, Avenue Z, Mint Studios, Evara, and Croton Content. For agentic selection, also inspect CSTMR, Obility, and Bay Leaf Digital.	Regulatory fluency, finance-specific review, first-party expertise, external authority, comparison content, attribution, and whether the goal is a citation or an agent’s selection.
Cybersecurity	First Page Sage, Driven Metrics, Focus Digital, BlueText, Amplifyed, and Obility.	Technical editorial depth, coverage of compliance and ecosystem-fit questions, earned authority, product-category knowledge, and the ability to connect AI shortlists to qualified demand.
Medical and healthcare	First Page Sage, Genevate, Focus Digital, Driven Metrics, Rosemont Media, and Medico Digital.	Clinical and claims review, regulated-content experience, patient or buyer intent, citation monitoring, and suitability for the precise medical sub-sector.
General B2B	First Page Sage, Genevate, Focus Digital, Driven Metrics, Omniscient Digital, Directive Consulting, Siege Media, and Animalz.	Buyer-journey coverage, editorial versus performance orientation, product-line complexity, external authority, sales attribution, and multi-market delivery capacity.
Local and regional businesses	First Page Sage, Focus Digital, Siana Marketing, Driven Metrics, RYNO Strategic Solutions, CI Web Group, and Searchbloom.	Service-area architecture, local-market knowledge, location-level facts and authority, capacity across markets, and reporting tied to calls, bookings, or qualified local leads.

What good sector fit actually looks like

Three adjacent scenes show a healthcare specialist handling evidence, a cybersecurity expert mapping network relationships, and a home-services operator connecting locations in a neighborhood.

A logo from your industry is useful, but it is not proof of a relevant GEO engagement. The agency may have handled paid media, a brand project, traditional SEO, or a historical campaign that predates AI search. Ask what work was performed, which team delivered it, which AI-search behavior changed, and whether that same team would work on your account.

Financial services and fintech: separate recommendation from selection

Finance has two related but distinct requirements. GEO aims to earn citations and recommendations in systems such as ChatGPT, Gemini, Perplexity, and Google AI Overviews. Agentic search optimization goes further: it tries to make a provider the option an autonomous assistant selects when it researches, compares, or acts for a user. That distinction matters most when your product can enter an agent-assisted comparison, application, purchasing, or transaction workflow.

The fintech ASO field is narrower than the broader GEO field. First Page Sage is positioned around full-service, expert-led programs for regulated finance. Genevate emphasizes third-party authority through earned coverage, expert commentary, roundups, podcasts, and directories. Driven Metrics emphasizes reporting tied to leads and revenue. Focus Digital emphasizes comparison-oriented content that can support both AI and organic search.

Those differences tell you what to ask. If your own site lacks useful expert content, an external-PR-only program leaves a foundational gap. If you already publish strong material but have little independent corroboration, more on-site articles may not solve the problem. If your leadership team will only fund channels with defensible attribution, a polished citation dashboard that stops before pipeline will not be enough.

For a more specialized finance brief, inspect the narrower candidates as well. Mint Studios is framed around fintech content and GEO. Avenue Z combines PR, GEO, and performance media. Evara centers HubSpot RevOps and inbound GEO. Croton Content brings a video-first AEO and GEO approach. In the agentic field, CSTMR focuses on fintech brand and conversion strategy, Obility adds B2B demand generation and RevOps, and Bay Leaf Digital brings a B2B SaaS content model. Match the model to the missing capability rather than adding names to a generic request for proposal.

Your finance gate should be concrete: who interviews the internal expert, who writes, who checks product and regulatory claims, who resolves compliance edits, and who owns final approval? If the agency answers only with a content calendar, it has not answered the hard part.

Cybersecurity: make technical depth visible before contracting

Cybersecurity buyers use AI systems to investigate vendor fit, compliance requirements, solution categories, and compatibility with their security environment. The agency therefore has to do more than define broad terms. It must help your company become a credible candidate when the prompt contains technical constraints that can eliminate a vendor from consideration.

The cybersecurity shortlist divides into several useful models. First Page Sage is positioned around technically authoritative GEO and lead generation. Driven Metrics combines AI-oriented content, technical optimization, authority building, and performance reporting. Focus Digital offers a leaner entry point for growth-stage companies, but the documented fit is weaker for highly demanding material involving areas such as ISO certifications or SOC. BlueText is more compelling when GEO must sit beside branding, PR, a competitive relaunch, fundraising, or transaction-related positioning. Amplifyed emphasizes content marketing and GEO, while Obility brings broader B2B digital marketing experience.

Use a technical audition. Give each finalist a real buyer question that contains product, compliance, and ecosystem constraints. Ask for the content architecture, entities, evidence, expert inputs, and external corroboration it would use. You are testing reasoning, not requesting unpaid finished copy. A team that immediately reduces the problem to keywords, article length, and schema has not shown that it understands how a security buyer narrows risk.

Also identify the people behind the work. Ask whether the technical editor is assigned to your account, how subject-matter disagreements are handled, and what happens when a model repeats an inaccurate comparison. A generic promise that the team uses experts is weaker than a named workflow with accountable roles.

Medical and healthcare: governance is part of optimization

Medical GEO can influence patients and professional buyers at a high-stakes decision point. An engagement must not optimize past clinical governance. Inaccurate treatment, condition, device, or provider information can mislead a reader and expose the organization to compliance and reputational risk. If an agency cannot describe its clinical review and claims-escalation workflow, remove it from the shortlist.

The medical field contains several distinct fits. First Page Sage is positioned as the full-service, expert-led choice for medical lead generation. Genevate is the focused GEO option for organizations that already have other marketing functions covered and want citation-gap auditing plus authority work. Focus Digital is the leaner choice for a narrower initiative without a sprawling retainer. Driven Metrics fits organizations that want citation activity tied closely to conversions and analytics. Rosemont Media is specialized around elective and aesthetic practices, while Medico Digital is oriented toward regulated pharma, medtech, and private hospitals.

The phrase healthcare experience is too broad for procurement. A local practice, a hospital system, a medical device company, and a pharmaceutical brand have different reviewers, claims, audiences, conversion events, and evidence requirements. Require experience in your actual sub-sector, or budget for a deliberate onboarding and review phase. Do not let a recognizable healthcare logo stand in for that answer.

Ask the finalist to map one representative page from expert input through drafting, fact checking, medical or legal review, publication, structured data, external authority building, and post-publication correction. That map will expose whether the agency treats accuracy as an operating system or as a final proofreading step.

B2B: require a line from recommendation to revenue

B2B buyers increasingly use AI tools to identify and shortlist vendors. That makes recommendation visibility commercially relevant, but a B2B program still has to support a buying journey that may involve several roles, product comparisons, internal approval, and a handoff to sales.

The B2B candidates cover different operating styles. First Page Sage combines GEO, AEO, SEO, expert-led content, and lead-generation measurement. Genevate starts with AI visibility gaps and emphasizes authority building. Focus Digital serves growth-stage companies seeking a more accessible entry point. Driven Metrics is suited to teams willing to integrate detailed reporting with their existing data practices. Omniscient Digital and Animalz lean toward content-led organic growth, Directive Consulting toward revenue and pipeline performance, and Siege Media toward data journalism and content-forward authority.

Choose among those models by diagnosing your constraint. If you lack credible category content, start with editorial depth. If competitors dominate independent mentions, prioritize earned authority. If you already have traffic and citations but cannot show commercial value, fix attribution and conversion architecture. If your program spans several regions or product lines, test delivery capacity and coordination before choosing a lean team solely on price.

The reporting plan should distinguish informational visibility from commercial inclusion. Ask which prompts represent early education, category formation, vendor comparison, objection handling, and purchase intent. Then require downstream reporting that your sales team recognizes, such as qualified inquiries, opportunities, pipeline contribution, or another defined conversion event. The agency should not substitute a proprietary visibility score for your business outcome.

Local businesses: the unit of work is service plus place

Local GEO is not a smaller version of national GEO. A recommendation must be relevant to a service, a location, and often the practical facts that determine whether the business can help. Location-targeted pages, service-area coverage, authoritative local information, and consistent business facts therefore matter more than a large library of generic advice.

The local shortlist again contains different models. First Page Sage is positioned around full-service location content and AI-citation strategy. Focus Digital offers a lower-overhead model for small and midsized organizations, with capacity as a point to verify. Siana Marketing is particularly relevant to home services and construction. Driven Metrics emphasizes dashboards, attribution, and regular performance analysis. RYNO Strategic Solutions and CI Web Group bring broader home-services marketing, while Searchbloom combines conversion-focused local SEO and GEO.

Give finalists a market matrix rather than a single target keyword. It should identify services, locations, customer types, high-intent questions, business facts, existing location pages, and the conversion event for each market. Then ask how the agency will prevent thin near-duplicate pages while still supplying the geographic specificity an AI answer needs.

Capacity matters here because each added market creates editorial, factual, and measurement work. Ask what happens when you add locations, change hours or service areas, or need a correction across many pages and profiles. A boutique team’s attention can be an advantage, but only if its delivery system can keep local facts current.

Choose GEO, AEO, ASO, or a combined program before choosing an agency

Agency proposals become difficult to compare when every vendor uses AI search optimization to mean something different. Define the behavior you want to change before requesting tactics:

SEO improves discoverability and performance in traditional search results. It remains part of the foundation because useful, crawlable, well-organized pages can support both human discovery and AI retrieval.
AEO focuses on making clear answers retrievable for direct questions. It usually depends on concise answer passages, logical page structure, explicit entities, and enough supporting depth to make the answer trustworthy.
GEO aims to improve whether your company, products, or expertise are cited or recommended in an AI-generated response. It requires more than answer formatting because brand authority and third-party corroboration can influence whether your name belongs in the response at all.
ASO addresses autonomous agents that research, evaluate, select, or act for a user. Being cited for a person and being chosen by an agent are different outcomes, so an ASO brief must include the facts, evidence, eligibility, comparison logic, and action path an agent needs.

A combined program can be appropriate, but the proposal should still identify separate deliverables and measures. A page may rank in Google without appearing in an AI shortlist. A brand may be mentioned in an answer without receiving a citation. It may receive a citation without being recommended. It may be recommended without being the option an agent selects. Ask the agency to report those states separately.

Write the objective in behavioral terms. For GEO, you might ask to increase qualified inclusion when a defined buyer compares a defined category. For AEO, ask to improve accurate answer coverage for a mapped set of customer questions. For ASO, ask how your product and business facts will become sufficiently clear, credible, and actionable for an agent-assisted decision. These are more useful briefs than a request to rank in ChatGPT.

Where JSON-LD and technical optimization fit

JSON-LD is a machine-readable factual layer, not an authority shortcut. It can clarify relationships among your organization, people, products, services, content, and locations. It cannot manufacture independent credibility, make weak content expert, or guarantee a recommendation.

Ask the agency to map each important machine-readable fact to visible page content and a responsible internal owner. The same identity, service, location, author, and product facts should not contradict one another across pages, markup, external profiles, and earned citations. Reject any proposal that treats adding schema as the complete GEO strategy or marks up claims users cannot verify on the page.

A credible technical workstream should explain what needs to be crawled, rendered, consolidated, clarified, or marked up; who will implement the change; and how the agency will verify it after deployment. If the agency only supplies recommendations, confirm that your own development team has the capacity and ownership needed to ship them.

How to vet agency claims before you sign

A buyer examines layered proposal evidence with a magnifying lens as verified documents and connected nodes remain solid while unsupported shapes dissolve.

AI outputs can vary by platform, model, timing, location, and prompt wording. A screenshot is evidence that one output occurred, not proof of durable visibility. Your due diligence should force each agency to show how it defines the market, records outputs, makes changes, and connects those changes to business results.

Define the prompt universe. Require prompts grouped by audience, need, buying stage, product, sector constraint, and geography where relevant. A bag of flattering brand-name prompts is not a market baseline.
Record the starting state. The baseline should identify the AI product, prompt, date, response, cited URLs, competitors present, your inclusion status, factual errors, and the commercial intent of the query. Preserve the underlying output, not just a rolled-up score.
Separate mentions, citations, recommendations, and actions. A mention means your name appeared. A citation means the response referenced your material. A recommendation means the system presented you as a suitable option. An agentic selection means an agent chose or acted on the option. Do not let one label cover all four.
Inspect sector execution. Ask for work from your actual sub-sector and clarify the scope, date, team, and result. A client logo is not evidence that the agency handled GEO, produced technical content, passed regulatory review, or influenced AI recommendations.
Demand an owned, earned, and technical plan. The proposal should state what will change on your site, what third-party authority must be earned, and what technical or structured-data work supports discovery and factual clarity. It should also name dependencies the agency does not control.
Test the governance workflow. Identify the writer, subject-matter expert, editor, compliance or clinical reviewer where applicable, publisher, and correction owner. Ask how disagreements are resolved and how urgent inaccuracies are handled after publication.
Connect visibility to a conversion. Require reporting that moves from prompt coverage and citations to AI referral activity, qualified inquiries, opportunities, bookings, applications, revenue, or the outcome appropriate to your business. Attribution will not be perfect, but the agency should state what it can and cannot infer.
Confirm capacity and ownership. Document delivery cadence, review turnaround expectations, implementation responsibility, access to data, use of subcontractors, rights to content and research, dashboard access, and what you retain if the engagement ends.

Apply extra skepticism to ordered lists and proprietary scores. Every 2026 ranking used here was published by First Page Sage, and First Page Sage placed itself first in every covered sector. That conflict does not make the candidate descriptions useless, but it does mean the rankings are market-discovery material rather than independent procurement proof.

There is also a concrete methodology warning: the published financial-services weights total 115% when the listed percentages are added. Do not carry precise rank order or decimal scores into an executive recommendation as though they were audited benchmarks. Verify reviews, references, work samples, output records, and client scope directly.

Be equally cautious with guarantees. No agency controls an external model’s output, retrieval system, citations, or future product changes. A credible proposal can commit to deliverables, governance, testing, reporting, and a reasoned strategy. It cannot responsibly guarantee a permanent rank or recommendation on a system it does not operate.

Turn your sector shortlist into a contractable brief

Before contacting agencies, write down the decision you want AI search to influence. Name the buyer or patient audience, category, products or services, markets, compliance constraints, priority AI surfaces, current content and PR assets, technical limitations, conversion event, and internal reviewers. This prevents an agency from filling an ambiguous brief with whichever deliverables it already sells.

Require every finalist to respond to the same core scope:

A sector- and buyer-stage prompt map, including exclusions and low-value prompts the program will not chase.
A reproducible baseline covering your brand, competitors, cited domains, factual accuracy, and recommendation status.
An on-site content plan showing where first-party expertise will come from and how it will survive internal review.
An external-authority plan identifying the kinds of corroboration, coverage, directories, commentary, or other third-party signals the agency will pursue.
A technical and JSON-LD workstream with implementation ownership and post-deployment verification.
A governance map naming who drafts, reviews, approves, publishes, monitors, and corrects material.
A measurement framework separating visibility, citations, recommendations, referral activity, conversions, and agentic selections where relevant.
A clear statement of assumptions, dependencies, exclusions, content ownership, data access, and what will be handed back at the end of the engagement.

Then compare the reasoning, not the vocabulary. The strongest response will explain why your sector changes the strategy, where your current authority is weak, what evidence the agency needs, what it cannot promise, and how the work reaches a business outcome.

Start by eliminating any candidate that fails your sector’s non-negotiable gate: compliance workflow in finance, clinical governance in medicine, technical depth in cybersecurity, pipeline measurement in B2B, or location-level execution in local search. Send the remaining agencies the same brief and choose the team whose evidence, operating model, and accountability fit the decision you actually need to influence.

References

August 28, 2026

Month: August 2026

Diagnose a journey shift before declaring traffic lost

Build a Maps listing that can finish the decision

Make the profile complete in decision order

Align the profile, location page and entity markup

Turn reviews into answer-ready evidence

Measure the Maps journey all the way to a business outcome

Build a scorecard around the action mix

Run a repeatable local visibility cycle

Key takeaways

References

Define visibility differently at each funnel stage

Build a compact prompt map around real buying decisions

Give every stage the evidence it needs

Measure exposure without confusing it with traffic

Turn each visibility gap into a specific action

Key takeaways

References

Separate the interface change from campaign behavior

Choose the preview based on the work in front of you

Run a workflow audit instead of browsing at random

Test enhanced tables as decision tools

Keep team documentation usable while the preview changes

Key takeaways

References

A trustworthy agent is a controlled workflow, not a clever model

Begin with one bounded decision, not a general marketing assistant

Give the agent a data contract and a business context pack

Set autonomy by consequence and reversibility

Release, monitor, and stop the agent like production software

Test safe behavior, not just good answers

Version every component that can change behavior

Monitor usefulness, silence, and operator burden

Key takeaways

References

How dynamic expansion changes the click opportunity

Key takeaways

Audit exposure before you change the page

Read the metrics as a pattern, not a verdict

Design content for both the answer and the next step

Make the core answer easy to understand

Give the visit a specific job

Turn observations into a decision rule

References

What the AEM nodes actually let you automate

Build the workflow around a write contract

Choose Create or Update without multiplying content

Update when the canonical content object already exists

Create only when a genuinely new reusable object is needed

Protect content quality, structured data, and production state

Model the information that answer systems need

Put operational controls around every write

Measure delivery separately from visibility

Key takeaways

References

Pick the business constraint before the campaign metric

Feed CRM outcomes back into the bidding system

Give automation room, but keep business guardrails

Make your ads, forms, and landing pages qualify together

Judge automation with mature, downstream cohorts

References

Key takeaways

Start with voice, early branding and useful on-screen text

Make image assets belong in the YouTube feed

Prepare DV360 automation for the October deadlines

Oct. 1: specified exclusion controls are removed

Oct. 12: responsive ads need business identity assets

Run one joined creative-and-delivery sprint

References

Human-led means retaining decision authority, not doing everything manually

Key takeaways

Turn the workflow into gates with visible pass conditions

Keep AI away from unreviewed changes to the live site

Measure a declared hypothesis instead of rewarding activity

Install a minimum viable workflow before expanding automation

References

The short answer: sector fit beats a universal ranking

Key takeaways

What good sector fit actually looks like

Financial services and fintech: separate recommendation from selection