Are you looking to elevate your affiliate marketing efforts? With Profound and PartnerStack, I’ve been able to efficiently activate the right affiliate publications on a larger scale than ever before.
Through this powerful collaboration, I’ve discovered new ways to enhance my campaigns and drive significant growth by engaging with the right audiences at the right time.
Inspired by this post on Try Profound Blog.
I recently discovered that OpenAI is set to introduce conversion-optimized ad campaigns starting in early June. This marks a significant step towards creating a performance advertising ecosystem within ChatGPT.
Why does this matter to us? This move by OpenAI, as reported by The Information, confirms the development of conversion-focused ads along with necessary tracking infrastructure and performance measurement tools for advertisers like us.
What’s the current update? OpenAI has communicated with advertisers, stating that those who set up the OpenAI Pixel or Conversions API in advance will get early access to these campaigns in June.
According to the company:
This system enables advertisers to measure actions triggered by ads, enhancing campaign effectiveness.
A deeper look. OpenAI is setting up an infrastructure akin to performance platforms like Google and Meta. With the OpenAI Pixel, advertisers can track website activity post-ad interaction, while the Conversions API allows them to send first-party conversion data back into OpenAI’s systems directly.
This capability allows OpenAI to optimize campaigns for measurable business outcomes, beyond just engagement metrics.
What’s at stake? The future of OpenAI’s advertising strategy largely hinges on measurement accuracy and gaining advertisers’ trust.
With browser restrictions and privacy changes eroding traditional tracking methods, OpenAI’s Conversions API could play a crucial role in demonstrating campaign performance and attribution within AI-driven ad experiences.
Inspired by this post on Search Engine Land.
I find it fascinating that users interact differently when faced with AI Overviews compared to AI Mode. New clickstream data reveals that AI Overviews significantly alter user behavior—from reverse scrolling to extended evaluation of search results across various intents.
Take Netflix, for example. The average user spends about 18 minutes just browsing. They skim through tiles, watch trailers, and often circle back. It turns out, searching isn’t much different these days, thanks to new insights.
This week, I’m diving into:
Eric Van Buskirk from Clickstream Solutions mined anonymized clickstream data supplied by Surfer SEO. The study analyzed around 846,000 U.S.-based Google searches from February and March of 2026.
This marks the fifth study on user behavior with Google’s AI features over the past year. Earlier, a UX study on 70 users in May 2025 utilized think-aloud and screen recording methods, while a study from October 2025 examined AI Mode specifically. This research trades depth for scale, uncovering patterns too subtle for smaller studies.
For a bit of context, previous SERP mouse-tracking studies involved only a handful of people—this one, however, evaluates queries from tens of thousands of users.
A fascinating contrast surfaces: User behavior in AI Overviews starkly opposes that in AI Mode, where AI Mode is akin to autoplay, while AI Overviews replicate the browsing experience.
This article outlines four major findings from this recent study and how they might influence your title tags and meta descriptions in 2026. Full methodology available here.
With groundbreaking insights, like how nearly half of AI Overview interactions involve reverse scrolling and how search types no longer reliably predict behavior, this data is invaluable. It challenges traditional assumptions and has meaningful implications for e-commerce and decision-heavy categories.
Surprising findings include brand searches losing their shortcut advantage, implying even users searching specifically for brands might pause to consider adjacent content on the SERP.
Read more intriguing insights on how the AI landscape shifts user engagement and strategy in SEO.
Inspired by this post on Search Engine Land.
You do not need another advertising dashboard that promises smarter automation. You need to know whether an AI-powered platform can reach the right people, optimize for a business result, and prove that it contributed to that result.
The safest way to evaluate these platforms is to separate reach, decision-making, and measurement. When those three layers are clear, you can use automation without surrendering control of your budget or accepting a platform’s preferred version of success.
Start with the moment you want to influence. A visual discovery campaign and a conversational recommendation may both use AI, but they address different behaviors.
Google is consolidating visual discovery inventory inside Demand Gen. A campaign can reach people across YouTube, Discover, Gmail, Maps, and Google Display Network sites. Advertisers can manage Display placements through Demand Gen and, when needed, keep delivery limited to the Display Network.
That setup is useful when your job is to create or reinforce demand across visual environments. It can support product discovery, introduce a service, or bring a previous visitor back with a stronger message.
Conversational advertising is developing around a different moment. OpenAI is preparing ads intended to generate purchases, appointment bookings, and contact-form submissions. The reported direction includes paying for completed outcomes rather than impressions, with an initial emphasis on smaller and local businesses. These capabilities are still emerging, so they belong on a readiness plan rather than in a forecast as guaranteed inventory.
Write one sentence before opening any platform: “We need this campaign to move a person from ___ to ___.” If the first blank is awareness and the second is consideration, broad visual distribution may fit. If the person is already discussing a need and the second blank is a booking or purchase, a conversational placement may eventually fit better. If you cannot complete the sentence, the platform will end up defining the campaign for you.
Calling a product “AI-powered” tells you very little. Ask what the system controls at each layer and what you can still inspect.
| Layer | Question to ask | Evidence you should require |
|---|---|---|
| Distribution | Where can the platform place the ad? | A channel list, placement controls, exclusions, and a delivery breakdown |
| Decision-making | What signals determine who sees it and when? | Optimization settings, audience inputs, creative combinations, and change history |
| Measurement | What event counts as success? | A written conversion definition, deduplication rules, attribution settings, and reconciliation with your own records |
This separation prevents a common mistake: treating more inventory as proof of better performance. Wider reach gives an algorithm more opportunities to serve ads. It does not automatically mean those opportunities are equally valuable.
Google has reported an average ROI increase of 9.5% among advertisers that added Display Network inventory to Demand Gen. Treat that as a reason to test the inventory, not as the return your account will receive. Your audience, creative, margins, conversion definition, and channel mix determine whether expansion produces incremental value.
For every automated expansion option, ask for a channel-level answer to three questions: How much did we spend? What did we receive? Would those conversions have happened through another channel anyway? If reporting cannot help you investigate those questions, do not increase the budget merely because the blended result looks efficient.
An optimization system can only pursue the signal you give it. If a low-value form submission and a completed sale are recorded as equivalent conversions, AI will optimize toward whichever event is easier to generate.
This work matters even more for conversational ads. OpenAI’s reported performance-advertising plans include a website pixel and API connections for conversion data. Pixel-only tracking can lose visibility because of browser restrictions and ad blockers. An API connection can provide a stronger path for confirmed customer actions, but only if your systems use stable identifiers and consistent event definitions.
Do not wait for a new platform to launch before cleaning up this layer. A reliable conversion specification can be reused across Google, Meta, a future ChatGPT campaign, and your internal reporting. It also gives finance, sales, and marketing one shared definition of a result.
Automation needs room to find patterns, but a useful test still needs boundaries. The goal is to learn whether the AI-controlled change produces incremental business value.
Channel controls are especially important as Google moves more Display management into Demand Gen. The ability to use broad cross-channel delivery or remain on the Display Network gives you a practical testing sequence: establish how the narrower setup behaves, expand deliberately, and then inspect where the additional spend went.
Use the same discipline when conversational ads become available to your business. A pay-for-success model sounds low-risk, but the definition and verification of “success” determine what you actually buy. Confirm whether the billable action is a submitted form, a qualified lead, a kept appointment, or a completed transaction. Those events are not interchangeable.
Your next step is not to move the whole budget into an AI-led campaign. Write the conversion specification, audit the tracking path, and select one contained inventory or optimization decision to test. That gives the platform enough freedom to help while keeping the business outcome under your control.
Your AI pilot probably does not need a smarter demo. It needs an accountable owner, a credible baseline, reliable data, permission boundaries, an escalation path, and a clear reason to exist after the demonstration ends.
That is where many enterprise programs stall. In adoption data compiled through May 14, 2026, enterprises led at 25% adoption, but adoption covered everything from an initial trial to full-scale implementation. Among enterprise adopters, 62% remained in experimentation and only 13% had reached full deployment. If you are responsible for moving AI automation into production, the job is not to collect more use cases. It is to turn a carefully chosen workflow into a controlled, measurable operating process.
Start below the level of a department. Customer service transformation is too broad. Qualifying an after-hours inquiry, answering approved questions, and offering an available appointment is a workflow. Supply chain optimization is too broad. Detecting a delayed shipment, checking an approved set of alternatives, and preparing a resolution for review is a workflow.
This distinction matters because ordinary automation and agentic AI solve different parts of the process. A conventional automation follows predefined rules. Generative AI produces an output such as a summary or draft. An agentic system can plan, decide, and execute a multi-step task from beginning to end. More autonomy creates more ways to complete useful work, but it also expands the number of decisions, integrations, and failure modes you must control.
A strong initial candidate has the following properties:
Score candidate workflows as high, medium, or low on each property. Do not average away a fatal weakness. Low data access, an undefined finish, or an unbounded consequence should block the candidate until the underlying process is redesigned.
Structured processes tend to move first. Customer service and supply chain coordination show stronger agentic AI adoption, while finance faces more regulatory scrutiny. The practical lesson is not that every enterprise should begin in customer service. It is that repeatable inputs, explicit policies, and observable outcomes make automation easier to validate.
A useful workflow can also be unglamorous. One documented PR automation locates a completed Zoom recording, creates a transcript, and prepares an email containing both for the journalist. It saves about 30 minutes per interview while shortening the handoff. The value comes from removing a specific delay, not from inventing a new communications platform.
Apply the same discipline to the build-versus-buy decision. Existing software should handle commodity functions such as scheduling, transcription, telephony, CRM records, and routine orchestration when it meets your requirements. Custom development is easier to justify when the workflow depends on a proprietary process, distinctive formula, or exclusive data that is central to the business. Otherwise, concentrate engineering effort on integration, policy, evaluation, and observability rather than recreating a mature product category.
Before selecting a model or vendor, write a testable operating hypothesis:
By automating these defined steps for these eligible cases, we expect this business metric to move from its recorded baseline to an approved target, without worsening these guardrails, as measured in this system over this evaluation window.
If the team cannot fill in each part, it is not ready to approve the pilot. A goal such as improve productivity leaves too much room to declare success after the fact. Reduce median handling time for eligible requests while maintaining resolution quality and escalation compliance can be measured.
The measurement plan should separate five kinds of evidence:
Measure the baseline before the team sees pilot results. Otherwise, definitions tend to drift toward whatever the system can demonstrate. Specify which cases qualify, which are excluded, where each metric comes from, and who resolves disputed labels. When feasible, compare pilot cases with equivalent manually handled cases rather than assuming every change came from the automation.
Do not count outputs as outcomes. Drafts generated, conversations handled, or tasks attempted are activity measures. They matter only when the workflow reaches a valid completion or produces verified capacity that the business can use. Time saved is not automatically a cash saving, either. State whether the capacity will absorb growth, reduce a queue, improve service, avoid new hiring, or be reassigned to higher-value work.
Revenue automations need an additional capacity check. AI can help build targeted prospect lists, accelerate qualification, recover missed calls, and respond outside staffed hours, but increased demand can damage the customer experience when the business cannot fulfill it reliably. Map the next handoff before accelerating the top of the funnel. A faster response is not valuable if it creates an unstaffed queue downstream.
Finally, define the stop rule while expectations are still neutral. Stop, narrow, or redesign the pilot if it cannot move the primary outcome, breaches an approved guardrail, depends on unsustainable review labor, or lacks a credible path to production economics. Unclear success criteria and weak data are recurring reasons AI projects fail to progress, while cost pressure is particularly important for smaller organizations. An enterprise budget may delay that reckoning, but it does not remove it.
Map the workflow from trigger to completion before deciding what the model should do. For every step, record the input, rule or judgment, system of record, permitted action, expected output, exception path, and owner.
Use ordinary code or workflow rules where the answer is deterministic. Required fields, account permissions, arithmetic, approved status transitions, duplicate checks, and routing tables should not become probabilistic merely because a language model is available. Use AI where interpretation is genuinely required, such as extracting intent from a message, summarizing an interaction, comparing unstructured evidence, or preparing a response under policy constraints.
This separation makes failures easier to locate. It also reduces the chance that a persuasive output will bypass a rule the business intended to enforce.
Autonomy should be an explicit permission level, not an accidental property of an integration. A practical authority ladder is:
Start at the lowest level that can test the business hypothesis. Advance only when the prior level meets predeclared quality and guardrail requirements. Full deployment does not require maximum autonomy. A stable draft-and-approval system can be the right production design when the action carries legal, financial, employment, security, reputational, or regulatory consequences.
Use least-privilege credentials and separate test access from production access. Restrict the agent to the systems, records, fields, and actions required for the approved workflow. Payments, deletions, contractual commitments, price changes, sensitive employee decisions, and regulated communications should not become autonomous merely to remove a review step. If the business later approves that authority, it needs risk-specific testing, monitoring, and recovery controls.
A production trace should let an operator reconstruct what happened without relying on the model to explain itself. Capture the case identifier, input snapshot, relevant data version, workflow and prompt version, model and tool calls, retrieved evidence, proposed action, approval or override, external write, error, retry, elapsed time, unit cost, and final business outcome.
Design retries so they do not duplicate a booking, order, message, refund, or record. Provide a clear shutdown control, queue failed work for recovery, and document how the operating team restores the last valid state. Alerts should identify an actionable condition and its owner; a dashboard that merely shows activity will not shorten an incident.
Data readiness should be scoped to the workflow. You do not need to repair every enterprise dataset before beginning, but you do need a reliable contract for the fields this automation uses: canonical definitions, stable identifiers, permitted sources, freshness expectations, missing-value behavior, conflict resolution, and write-back ownership. Poor-quality and inconsistent data are common barriers to successful agent deployment. Giving an agent access to more systems does not solve disagreement between those systems.
Build an evaluation set from representative normal cases, boundary cases, known exceptions, and costly failure modes. For each case, define an acceptable result, required escalation, and prohibited action. Run it before live access, compare the system with the existing process in shadow mode, and retain it as a regression suite whenever the prompt, model, tools, policy, or data mapping changes. Production monitoring then checks whether real traffic is drifting beyond what the evaluation set covered.
The large gap between experimentation and full deployment is a governance problem as much as a technical one. Teams can keep improving a demonstration indefinitely when nobody has defined the evidence required for the next decision. Gartner has projected that around 40% of agentic AI projects could be canceled by 2027. Cancellation is not necessarily the wrong outcome; discovering weak value or uncontrolled risk early is cheaper than scaling it.
| Gate | Evidence required | Decision |
|---|---|---|
| Workflow approval | Named owner, process map, baseline, eligible cases, business hypothesis, risks, and stop rule | Approve a bounded test, redesign the workflow, or reject the use case |
| Offline validation | Data contract, representative evaluation set, expected results, prohibited actions, permission design, and cost model | Move to shadow operation only if declared quality and safety requirements are met |
| Shadow operation | Comparison with the existing process, exception analysis, reviewer feedback, diagnostic logs, and revised operating procedures | Enter limited production, narrow the scope, or return to offline work |
| Limited production | Verified business outcome, guardrail performance, coverage, review burden, incident response, rollback, and actual unit cost | Scale, maintain the bounded scope, redesign, or stop |
| Operational scale | Accountable service owner, support model, change control, recurring evaluation, capacity plan, security review, and portfolio funding | Expand only while value and controls remain intact |
Set the thresholds for these gates according to the consequence of failure, and approve them before results arrive. A drafting assistant and a payment agent should not share the same tolerance. The important discipline is that the team cannot redefine success after seeing the output.
At portfolio level, centralize the controls that should be consistent and decentralize ownership of the business outcome. A central AI function can provide identity, approved integrations, logging, evaluation tooling, security patterns, vendor review, and incident standards. The operating team should still own the process, metric, exceptions, staffing impact, and customer consequence. If ownership remains with an innovation lab after launch, the automation has not truly entered the business.
Maintain a register of active automations showing the workflow owner, systems touched, data classification, permitted actions, risk level, deployment stage, model and vendor dependencies, current economics, and next gate. Use it to find duplicate experiments, unsupported integrations, and pilots that consume resources without approaching a decision.
Before the next platform purchase, choose a specific queue or handoff that is already causing measurable loss. Name its owner, baseline, eligible cases, prohibited actions, escalation path, and stop rule. If those items cannot be written clearly, more AI will not make the process ready. If they can, you have the beginning of an automation that can earn its way into production.
You may still be managing AI search, paid media, product data, and ecommerce as separate workstreams. That separation is becoming the risk. AI platforms are starting to answer a question, present a promotion, select a call to action, and support a shopping task inside the same environment.
You don’t need to rush into every beta. You need a commerce system in which your product facts, content, ads, landing experience, checkout, and measurement agree. Build that foundation now, and you can test new platform inventory without handing the platform control of your customer truth.
The familiar funnel hasn’t disappeared. Its stages are being compressed. A shopper can ask for a recommendation, compare options, encounter an ad, and begin a transaction without moving through the sequence of search result, publisher page, product page, and checkout that your reporting was designed to measure.
Two developments make that shift concrete. Google has introduced Universal Cart as a cross-platform shopping protocol. OpenAI is testing ChatGPT ads with automatically selected calls to action such as Shop Now, Book Now, Sign Up, and Learn More, based on the creative and destination experience. The platform is no longer limited to referring demand. It can shape how that demand moves toward an action.
| Commerce layer | What the customer is doing | What your brand must control | What to measure separately |
|---|---|---|---|
| Answer and discovery | Asking, comparing, or narrowing a choice | Clear claims, product facts, evidence, and current availability | Visibility, mentions, referrals, and assisted discovery |
| Paid placement | Considering a promoted option or call to action | Creative, targeting, budget, offer, and destination alignment | Impressions, clicks, spend, and qualified arrivals |
| Transaction | Starting a cart, booking, signup, lead, or purchase | Price, inventory, eligibility, checkout rules, and customer support | Completed actions, order value, margin, cancellations, and refunds |
| Owned customer system | Receiving the product or continuing the relationship | Order records, consent, service, retention, and first-party history | Fulfilment, repeat business, support cost, and customer value |
A single customer interaction may cross all four layers. That doesn’t mean one platform deserves credit for the entire outcome. Keep discovery, paid exposure, transactional handoff, and the final owned record distinct whenever the available data allows it. If you collapse them into one conversion number, you won’t know whether you improved demand, bought more traffic, reduced checkout friction, or merely changed which system claimed the sale.
This distinction also protects your SEO, AEO, and GEO work. An organic recommendation, an ad beside an answer, and a platform-assisted purchase are different events. Report them separately even when they happen in the same interface.
OpenAI’s Ads Manager beta is gaining the controls expected of a more established media platform. New campaigns can use a daily or lifetime budget, while daily budgets currently apply only to newly launched campaigns. U.S. targeting can be set by state, designated market area, or ZIP code and adjusted later in campaign settings. Reporting tables now show aggregate impressions, clicks, and spend across campaign, ad group, and ad views. These changes make the channel easier to operate, but they don’t settle attribution, customer ownership, or transaction governance.
Google’s Universal Cart raises the stakes further because a shared shopping protocol can move the platform closer to the transaction itself. That may reduce steps for a shopper. It can also increase a merchant’s dependence on platform rules, identifiers, interfaces, and reporting. The right response is neither automatic adoption nor blanket refusal. It is a staged implementation with an exit path.
That caution matters because AI products are shipping quickly. At Google I/O 2026, overlapping Search and Gemini functions were explicitly framed around velocity and reduced managerial overhead. Information agents in Search and Spark or Daily Brief functions in Gemini already point toward overlapping ways to monitor the web. Some lifecycle questions, including how aging alerts and accumulated information should be managed, were still unresolved in the demonstrations.
Use four operating rules for any AI commerce or advertising integration:
Before enabling a transactional protocol, get written answers to a short set of questions: Which system wins when price or inventory conflicts? Where is the cart created? How is a platform cart mapped to an owned order? What data can you export? What happens when a product becomes unavailable during the handoff? Who handles cancellations, returns, and customer contact? If a provider can’t answer those questions yet, limit the scope until it can.
AI commerce readiness begins before the campaign setup screen. An agent, answer engine, ad system, and checkout can only coordinate reliably when the same offer is described consistently across your visible page, product feed, structured data, ad creative, and transactional system.
The apparent conflict between human-focused publishing and agent-readable commerce is avoidable. Google’s Search quality guidance told publishers to write for humans rather than AI, while Google’s own agent demonstrations showed systems browsing, interpreting, transacting, and creating web content. You shouldn’t respond by producing bot-only pages. Give the person a useful answer and make the underlying facts explicit enough for a machine to interpret without guessing.
Use this sequence for each important product, service, offer, or location:
Do not assume that adding schema automatically enrolls you in a commerce protocol or guarantees inclusion in an AI answer. Platform eligibility, integrations, and advertising access are separate from good structured data. The purpose of your content and JSON-LD layer is to reduce ambiguity and keep your own representation coherent, whether the next consumer is a crawler, an agent, an ad system, or a customer.
Avoid four shortcuts: pages written only for bots, duplicated doorway content for every conversational query, markup that overstates what the visible page offers, and platform-specific product records with no owned master. Each shortcut may make an initial integration look faster. Each also increases the chance that your answer, ad, cart, and fulfilment system disagree later.
AI-native advertising should begin as an acquisition experiment with one decision attached to it. Don’t launch merely to learn whether the interface can spend money. Decide whether you are testing qualified traffic, completed purchases, bookings, leads, incremental demand, or a particular geographic market.
A practical first test looks like this:
The aggregate totals now available for impressions, clicks, and spend make pacing checks faster at campaign, ad group, and ad level. They do not replace the rest of the commercial record. A reporting table can confirm that delivery occurred and money was spent. Your analytics, CRM, commerce system, and finance records still have to confirm what happened after the click.
Keep four evidence classes separate in your analysis:
Label inferred results as inferred. Do not mix them into directly reconciled conversions and present the sum as one observed total. That distinction will matter more as discovery and transactions happen inside interfaces where your analytics may see only part of the journey.
Set your scaling conditions before the campaign starts. At minimum, confirm that product data remains correct, the automated or displayed call to action reaches a matching experience, the final action is validated in an owned system, platform spend reconciles, and the resulting customer or order quality meets the target you already use for other channels. If one of those conditions fails, repair that layer before increasing the budget.
Start with one product family or service, one valid geography, one destination, and one business outcome. Audit the product record and structured data, test the complete action path, and instrument the handoff before you launch. Expand only when an order or lead can travel from platform exposure to your owned system without the facts changing along the way.
You probably don’t need another dashboard. You need a dependable way to turn one campaign brief into coordinated channel work, bring the results back into one operating view, and move from a useful signal to an approved action without reopening every platform.
AI can shorten that loop, but only when it sits inside a clear operating system. Give it shared definitions, bounded permissions, review gates, and a record of every decision. Without those controls, AI simply produces inconsistent work faster.
When a campaign spans 12 channels, weekly reporting can become a chain of exports, spreadsheet repairs, naming lookups, metric reconciliation, screenshots, and explanations. The obvious cost is staff time. The more damaging cost is latency: a performance problem can continue consuming budget while the team is still assembling the evidence needed to discuss it.
Start by tracking a full working week before choosing an AI tool. Record the work as it happens, including small tasks that disappear inside a reporting block. Use one row per task and capture:
Then classify each task by the kind of work it contains. Retrieval moves information out of a channel. Reconciliation makes names and totals line up. Interpretation decides what the evidence means. Execution changes a live campaign. Explanation turns the decision into something another person can understand.
This classification reveals where AI belongs. Repeated retrieval, formatting, matching, and first-draft explanation are strong candidates for assistance. Budget choices, attribution judgments, brand claims, audience exclusions, and live publishing require tighter human control. A task can contain both kinds of work, so automate the bounded transformation rather than handing over the entire task.
Prioritize bottlenecks by their effect on the data-to-action cycle, not just by the hours they consume. Map the path as signal → review → decision → platform change → verification. A repetitive task near the beginning of that path can delay every decision downstream. Removing that delay is usually more valuable than automating a polished deliverable that nobody uses to make a decision.
Cross-channel automation needs a control plane: a small set of shared objects and rules that exist independently of any network. The central object should be a campaign contract. This is the approved record of what the campaign is trying to do and which elements must remain consistent when work moves between channels.
A practical campaign contract should identify the business objective, intended audience, offer, message, conversion event, budget guardrails, geographic scope, active period, creative concept, required claims or disclaimers, asset identifiers, owner, approval state, and canonical campaign ID. It should also distinguish fixed elements from adaptable ones. The offer may be fixed while format, length, crop, placement, and channel-specific wording remain adaptable.
The canonical campaign ID matters because network names are presentation labels, not reliable identity. Adopt a consistent naming convention across accounts, but keep a separate registry that maps every network campaign, ad group, creative, and tracking asset back to the shared campaign. This lets a shortened or platform-constrained name change without breaking the relationship.
Build a metric dictionary beside that registry. For every metric used in a cross-channel view, record its business meaning, originating system, calculation, attribution basis, refresh expectation, exclusions, and owner. Networks can use different campaign structures and attribution logic, so identical labels do not guarantee identical measurements. Keep platform-reported conversions, analytics conversions, and modeled business outcomes visibly distinct unless you have an explicit reconciliation rule.
| Operating layer | Authoritative record | What AI may do | What must be controlled |
|---|---|---|---|
| Intent | Approved campaign contract | Draft channel adaptations and identify missing fields | Objective, offer, audience, claims, and approval state |
| Identity | Canonical campaign registry | Suggest matches between network objects and shared IDs | Ambiguous matches and changes to existing mappings |
| Evidence | Raw channel data plus metric dictionary | Normalize formats, flag gaps, and prepare summaries | Definitions, attribution differences, and reconciliation rules |
| Decision | Recommendation and approval ledger | Generate hypotheses, summarize evidence, and draft actions | Final judgment, accountable owner, and authorization |
| Execution | Platform change history | Prepare or queue permitted changes | Spend, publishing, targeting, deletion, and rollback |
This design prevents a common failure: forcing every channel into one flattened schema and calling the result unified. Unification should make relationships visible while preserving meaningful differences. Normalize identity, ownership, dates, currencies, and approved definitions. Do not erase attribution differences or channel-specific context merely to make the spreadsheet look tidy.
An AI assistant performs better when each job has a defined input, transformation, output, and permission boundary. Telling it to optimize the campaign mixes analysis, judgment, execution, and accountability into one instruction. That makes errors harder to detect and leaves nobody certain about what the system changed.
Write an AI work order for every automated workflow. Include:
For example, a creative adaptation job could receive an approved campaign contract and master asset. It may adjust length, format, placement language, and crop guidance for each channel. It must preserve the offer, approved claims, audience, and call to action. Its output should contain draft variants, assumptions, missing assets, and a review status. It should have no publishing permission.
Use deterministic rules where the answer must be exact. IDs, currencies, required fields, date formats, budget caps, and approval states should be validated by explicit logic. AI is useful when language or context is ambiguous: matching imperfect names, classifying creative themes, finding possible explanations, adapting a brief, and turning structured evidence into a readable draft. It should not quietly invent a value when an exact field is missing.
A sensible permission ladder moves from read to draft, then recommendation, approval queue, and finally limited execution. Advance a workflow only after you can reconcile its inputs, inspect its logs, identify an accountable owner, detect failures, and reverse an incorrect change. For paid campaigns, unreviewed budget or targeting changes can waste money. For owned channels, an unreviewed publishing action can expose inaccurate claims. Keep those actions behind explicit approval until the controls have proved dependable.
The goal is not to keep humans clicking every button forever. It is to reserve human attention for decisions that involve trade-offs, accountability, or material risk. The system can handle preparation and coordination while the owner approves the action and remains able to explain why it happened.
A unified dashboard still leaves someone hunting for the important row. An effective operating view should instead tell you what changed, what needs attention, what decision is blocked, and whether an approved action reached the platform correctly.
Organize the working queue around four kinds of exception:
Check data health before discussing performance. A persuasive summary built from a stale connector or broken campaign mapping is still wrong. Surface the affected channels, the last successful refresh, the missing entities, and the decisions that should be paused until the evidence is repaired.
Turn every recommendation into a decision record. Capture the campaign ID, evidence considered, attribution basis, proposed action, expected effect, uncertainty, reviewer, approval status, execution status, platform confirmation, and rollback instruction. If the recommendation changes during review, preserve both the original and approved versions. This gives you a traceable chain from evidence to action instead of a collection of chat messages and overwritten spreadsheet cells.
Reporting should follow the same logic. Lead with business outcomes and material changes. Show what moved across channels, but label differences in attribution and data freshness. List actions completed, decisions required, owners, and unresolved data-quality issues. Put diagnostic detail in an appendix rather than forcing a stakeholder to infer the decision from a wall of metrics.
Agencies can also automate branded reports assembled from multiple networks. The narrative still needs controls. Generate it from the approved metric dictionary and decision ledger, require links back to the underlying evidence, and prevent the report from presenting a hypothesis as a confirmed cause. Automation should remove assembly work without hiding uncertainty.
Evaluate AI-native tools against your workflow, not their most polished demo. The useful promise is a shared brief that can coordinate work across channels and a unified view that shortens the route from evidence to action. Whether a product can support that promise depends on its connectors, identity model, controls, and failure behavior.
Ask each vendor or internal team to demonstrate the following with a representative campaign:
Begin the pilot with a frequent, reversible workflow such as weekly data assembly, exception detection, recommendation drafting, and report generation. Connect data in read-only mode first. Establish the campaign mappings and metric definitions, reconcile the output, and then allow the system to draft recommendations. Keep execution behind approval while you test whether the evidence, reasoning, and logs are good enough to support a real decision.
Measure the pilot against your own baseline. Track hands-on reporting time, waiting time, manual transfers, corrections, unmatched entities, stale-data incidents, recommendations accepted or materially changed, and elapsed time from signal to verified action. Do not substitute a vendor’s productivity claim for the bottleneck you observed in your own audit.
Pause expansion if the system cannot reproduce agreed totals, preserve attribution context, identify the evidence behind a recommendation, enforce approval boundaries, or reveal what it changed. Those are operating requirements, not optional refinements. Adding more channels before they work will multiply ambiguity.
On your next reporting cycle, start the task log before opening the first platform. Use what it reveals to write the campaign contract and select one approval-gated workflow. Once that workflow can move from clean evidence to a verified action with a complete record, you have something worth extending to the next channel.
If AI seems unavoidable in your professional feed, it is easy to assume your customers have already moved their discovery and buying journeys into ChatGPT, Claude, or Gemini. That assumption can send budget toward the loudest channel rather than the audience you actually serve.
The useful question is not whether AI is popular. It is which audience uses which assistant for which job, and whether that behavior affects discovery, evaluation, or purchase. Once you separate those questions, you can make a defensible AI search plan instead of reacting to general enthusiasm.
Broad reach and segment-level growth can move in opposite directions. At its measured high point, OpenAI or ChatGPT reached 37% of U.S. desktop users in September 2025, then slipped to 34% by March. That is a reach signal within a specific geography and device class. It does not mean 34% used the tool daily, preferred it over every alternative, or relied on it during a purchase.
The professional pattern looks different. Claude usage among B2B professionals was 373% higher than the U.S. average, while Claude and Gemini continued to gain users as ChatGPT’s desktop growth slowed. The 373% figure describes relative overrepresentation. It is not a market-share percentage, and it does not prove that most professionals use Claude.
Retail-shopping audiences provide the counterweight. People in that audience were 15% less likely to use ChatGPT than a typical U.S. consumer, and Claude did not rank among their top four AI tools. An AI-heavy professional network can therefore give you a distorted baseline for consumer behavior.
This is not a clean split between people who use AI and people who do not. The same person can be a heavy assistant user at work and follow a conventional search, marketplace, or retailer journey when shopping. Adoption depends on context, task, and perceived value, not just demographics.
A market-wide AI number cannot tell you where to publish, what to optimize, or which assistant deserves attention. Build an audience-by-task map instead. It should distinguish what has been observed from what still needs to be tested.
| Audience | Observed signal | What it does not establish | Planning response |
|---|---|---|---|
| Broad U.S. desktop users | OpenAI or ChatGPT moved from 37% reach in September 2025 to 34% by March | Frequency, task, loyalty, mobile behavior, or purchase influence | Maintain a baseline presence, but do not forecast automatic growth from general awareness |
| B2B professionals | Claude usage was 373% higher than the U.S. average | Which roles, industries, or work tasks produced the difference | Add Claude to role-specific discovery and evaluation tests |
| Retail-shopping consumers | ChatGPT usage was 15% lower than among typical U.S. consumers; Claude was outside the top four AI tools | Whether AI influences an earlier research step or a later purchase decision | Preserve conventional shopping journeys and test assistants selectively |
The most common planning error is to start with a platform and look for reasons to fund it. Start with the audience and task instead. The platform should be the last column you fill in, not the first.
Adoption signals should change your priorities, not your technical standards. Pages still need to be crawlable, indexable, internally linked, consistent about named entities, and clear enough for a person to verify. Structured data must describe visible content accurately; it cannot compensate for a vague, unsupported, or inaccessible page.
Where your audience resembles the measured B2B cohort, Claude belongs in the test set. That does not justify abandoning ChatGPT or Gemini. It means a ChatGPT-only visibility report can miss an assistant that is unusually prominent among professional users.
Lower ChatGPT incidence among retail shoppers and Claude’s absence from that audience’s top four do not make AI irrelevant. They do make an assistant-only discovery plan hard to defend. Keep the complete shopping journey usable without requiring an AI intermediary.
The useful distinction is not B2B equals AI and B2C equals conventional search. It is that professional adoption currently provides a stronger reason to test multiple assistants aggressively, while consumer planning needs more segment-specific proof before AI becomes the primary route.
A single AI traffic chart cannot tell you whether customers are adopting assistants, whether assistants know your brand, or whether visibility changes business results. Track those questions in separate layers.
Use a simple decision rule. Scale when a defined audience uses an assistant for a relevant task, your visibility has a fixable gap, and improvement is associated with a qualified outcome. Run a contained test when audience and task are supported but commercial impact remains uncertain. Keep monitoring lightweight when the only evidence is broad market enthusiasm.
For your next planning cycle, choose one high-value professional segment and one important consumer segment. Build separate audience-task maps, test the assistants indicated for each, and move the next content investment only where audience, task, and outcome align.
I’ve just delved into Goodie’s enlightening AI search traffic report for early 2026, covering the period from January to April, and I’m excited to share my insights with you. This report dives into trends in usership, referral traffic, and marketing considerations, offering a comprehensive view of the shifting landscape.
You’ll want to pay particular attention to how ChatGPT’s dominance is starting to wane, with some surprising contenders like Claude and Gemini making waves. This shift could significantly impact how marketers strategize their efforts in AI-driven search optimization.
The data reveals fascinating patterns in user habits and referral traffic, which could inform future marketing strategies and the allocation of resources. For a full dive into these emerging trends and what they might mean for businesses, I encourage you to explore the detailed findings of the report.
Inspired by this post on HiGoodie Blog.
Entity optimization might sound like a complex term, but trust me, it’s incredibly powerful when you’re trying to make AI understand your brand better. Essentially, my goal is to help AI see exactly who I am and what I’m about. Let me share more about how you can do the same.
When I optimize entities related to my brand, I start by clarifying what my brand represents. This means ensuring that all my online content clearly reflects my brand’s identity and core values. By creating a strong, consistent message, AI can better understand and categorize my content.
Next, I focus on strengthening associations. This involves connecting my brand with relevant entities and concepts within my industry. When AI detects these connections, it increases my brand’s relevance in related searches.
Finally, driving accurate AI citations is crucial. I make sure that any references to my brand on different platforms are correct and consistent. This helps in building trust with AI, ensuring that it can reliably reference my brand in the right contexts.
Inspired by this post on HiGoodie Blog.