I see Google’s unveiling of Gemini Intelligence at the May 12 Android Show as a significant step toward an agent-powered future. Announced alongside a new laptop called the Googlebook, Gemini Intelligence is designed as an underlying layer that works across the Android operating system on laptops, phones, watches, and glasses.
The Googlebook makes that vision tangible to me. Built from the ground up around an AI agent, it can understand what is on the screen and act on it. I could point to a date in an email and have the agent schedule a meeting, or select furniture in an app and see how those pieces might look in my living room.
I believe this ability to complete tasks without requiring someone to open a webpage will fundamentally change how people search, discover information, and conduct commerce. Here is how I expect that shift to affect the search industry.
What the shift to an agentic operating system means
Until now, I have viewed search as a familiar sequence: someone has a question or intent, enters it into a search engine, receives a list of links, and chooses one. Earning a prominent position on that list was the prize, and much of the SEO industry was built around winning that click.
Gemini Intelligence starts from a very different assumption. Search intent still exists, but an AI agent can handle the steps between the request and the outcome. It can read pages, complete forms, and increasingly finish the entire task. Instead of visiting a website myself, I may have an agent visit and use it on my behalf.
When I look for an early example, Chrome Auto Browse stands out. Launched in January and built on Gemini 3, it can manage multistep tasks such as researching flights, filling out forms, scheduling appointments, and managing subscriptions. It then pauses for approval before making a purchase.
A 2025 preprint supports this view. Researchers evaluated the declared-tools approach across online shopping, authentication, and content management. They found that giving an agent pre-structured interaction data reduced processing requirements by 67.6% and lowered costs by 34% to 63% compared with parsing a complete HTML document. Task success declined only slightly, from 98.8% with the traditional method to 97.9%.
The architecture behind Gemini Intelligence
To me, the architecture is as important as the interface. AI agents naturally favor websites they can interact with cleanly and efficiently, and Gemini Intelligence can only deliver on its promise if those agents can perform tasks reliably.
I see two protocols as central to making that possible. WebMCP turns a website’s actions into callable tools, while the Universal Commerce Protocol (UCP) allows an agent to complete a sale. Together, they enable an agent to finish a task without requiring a person to load and navigate the underlying webpage.
As AI reshapes search, every glowing path to discovery carries commercial value—turning SEO investment into a conversation about pipeline, risk, and customer acquisition costs.
WebMCP
I think of WebMCP as a labeled menu for AI agents. The API allows a website to declare functions as structured tools an agent can call, including searching inventory, beginning checkout, or submitting a support request.
Google co-developed WebMCP with Microsoft. An origin trial is live in Chrome 149, Firefox has committed to the third quarter of 2026, and Safari is expected to follow in the fourth quarter.
Universal Commerce Protocol (UCP)
I see UCP as the transactional counterpart to WebMCP. It gives AI agents a shared language for discovering products, building a cart, completing checkout, and managing orders without requiring someone to visit the merchant’s website.
Google also offers a consumer-facing layer called Universal Cart. It can collect items as I move across Search, Gemini, YouTube, and Gmail, creating a more connected shopping experience across Google’s products.
The range of companies behind UCP shows me how seriously the industry is taking this shift. Google, Shopify, Walmart, Target, Etsy, Wayfair, PayPal, and Stripe co-developed the protocol, which launched in January.
How I would prepare for agentic AI
My main takeaway is that websites are rapidly evolving from destinations into backends—from places people actively visit into systems agents quietly use. As the operating system becomes a search and action layer, I no longer think ranking is the only question that matters. I also need to ask whether an agent can actually use the site.
To prepare, I would begin by auditing the site’s most valuable actions, whether that means submitting a lead form, completing a booking flow, or reaching checkout. I would determine whether an agent could complete each action reliably and check the site’s Lighthouse Agentic Browsing score much as I would review Core Web Vitals. The goal is to understand whether an agent can use the site, not merely read it.
If I ran an ecommerce business, I would confirm whether the checkout process is accessible through UCP or ACP. I would also continue investing in retrieval and visibility because an agent still needs to find and trust the business before it can act on anyone’s behalf.
I’m introducing support for GPT-5.6 in Profound, bringing OpenAI’s newest flagship model family directly into the workflows I rely on for advanced AI performance.
With GPT-5.6, I can work across the new Sol, Terra, and Luna tiers, giving me the flexibility to support everything from frontier reasoning to high-throughput production workloads.
I’m especially focused on what this means for agentic workflows, coding, research, and enterprise knowledge work. GPT-5.6 delivers meaningful improvements in capability, reliability, and efficiency, making it easier for me to apply AI across a wide range of business use cases with more confidence.
I see Profound’s MCP evolution as a meaningful shift for Marketing Engineers. It now connects agents to a knowledge graph and adds 15 new capabilities built around how marketing teams actually work.
For retailers, I believe this demands a serious reframe. Answer engines are already shortlisting products and shaping purchase decisions long before shoppers ever land on retail or ecommerce websites. That compresses the shopping funnel and makes traditional search less reliable as the primary channel for customer acquisition.
Instead of waiting for shoppers to arrive through search, I need to think about how retailers can be recommended throughout the entire shopping journey. That means understanding how people use answer engines for Christmas gifting, how brands earn mentions and citations in relevant AI responses, and how visibility can be maximized across AI search experiences.
I see this report as a practical edge for retailers preparing for the next holiday cycle. It uses real shopper behavior from Christmas 2025, analyzed through Profound’s AI visibility lens, to show how people are using AI to shop for the holidays.
Most importantly, it turns those insights into actionable takeaways. By understanding where answer engines influence discovery, comparison, and purchase decisions, I can see how ecommerce teams should optimize product visibility before the 2026 season ramps up and compete more effectively for the AI shelf this Christmas.
I am introducing support for Grok 4.5 in Profound, bringing SpaceXAI’s newest flagship model into workflows built for deeper, more capable AI analysis.
Grok 4.5 is designed for agentic workflows and knowledge work, which makes it a strong fit for teams and operators who need AI systems that can reason, assist, and move complex tasks forward with more context.
With this support now available in Profound, I can use Grok 4.5 as part of a broader AI workflow and explore how its capabilities help with research, strategy, automation, and day-to-day knowledge work.
I’m reading this Cornell Tech research as a clear warning: deep-research AI agents can be steered by surprisingly small edits on public, user-generated pages. In the study, a single injected Reddit-style comment could become a cited recommendation for fake products, services, or entities.
The researchers described these altered pages as “poisoned” because the added text was written to influence what an AI system cites and repeats. The weakness appears in systems that search the web, collect sources, and produce cited reports. The paper calls the attack WARP, short for Web Agent Retrieval Poisoning.
How I see injected text reaching reports. The attack does not require access to the model, prompts, search engine, or retrieval system. Instead, an attacker edits or appends text to a page the agent already tends to retrieve, such as a Reddit thread, Wikipedia page, or forum post.
When the agent later searches related topics, it may pull in that page, cite it, and repeat the attacker’s chosen message as part of an otherwise normal-looking answer.
That matters because deep-research tools often run many related searches for a single user request. The paper found that the same user-generated pages surfaced across related queries, giving poisoned content more chances to appear.
Reddit stood out as the biggest opening. Across STORM, Co-STORM, and OmniThink, 17% to 23% of retrieved URLs came from user-generated platforms, including Reddit, YouTube, Facebook, and Wikipedia.
Reddit made up the largest share of those pages. It accounted for 54% to 71% of the user-generated URLs retrieved by the three open-source systems.
The researchers did not alter live websites. Instead, they used a simulation framework called GeoStorm to insert manipulated text into retrieved content during testing.
A few words were enough. What stood out to me most is how little text the attack needed. The researchers found that snippets as short as about 13 words could influence what these systems recommended.
In one test, a 15-word sentence pushed a fake cryptocurrency, BananaCoin, into a Co-STORM report as an “emerging” long-term investment option. The report cited the altered source alongside legitimate crypto sources.
When the manipulated page was retrieved, the fake entity appeared in 38% to 51% of reports across systems. When the researchers targeted multiple pages, that range increased to 42% to 62%.
The attack still worked when systems retrieved full Reddit threads, although mention rates were lower. When injected text was added to complete Reddit threads and represented less than 4% of the retrieved content, the fake entity still appeared in 30% to 53% of reports when the page was retrieved.
The defenses struggled. Blocking user-generated domains stopped this attack path, but I see the tradeoff immediately: it also removes useful sources such as firsthand product experiences and local recommendations.
The tested text filters also failed to reliably separate injected passages from normal user content. Because the manipulated passages were fluent and written by an AI model, perplexity-based filters were more likely to flag normal user content than the injected text.
Report-level checks missed the manipulation too. The altered reports looked similar to clean reports because the agent itself folded the fake recommendation into an answer that otherwise appeared normal.
Why I care. A small edit to a public page can become part of a cited AI answer, even when the underlying source is user-generated. Misinformation planted on sites like Reddit or in forums can move from discussion threads into AI recommendations that look credible to users.
About the research. The paper, Deep-Research Agents Can Be Poisoned via User-Generated Content, was written by Tingwei Zhang, Harold Triedman, and Vitaly Shmatikov of Cornell Tech and posted to arXiv on May 22. The researchers tested the full attack on three open-source systems: STORM, Co-STORM, and OmniThink.
They also analyzed OpenAI Deep Research and Gemini Deep Research for user-generated citations, but they did not run live manipulation tests because doing so would require publishing altered content to the open web.
I started this series with a simple observation: AI systems do not always give the same answer to the same question. My argument was that this inconsistency is not just randomness. It is confidence loss across a pipeline we can measure, diagnose, and improve.
As I worked through the AI engine pipeline gate by gate, I eventually reached the won gate. That is where three kinds of clicks appear: the imperfect click of search, the perfect click of recommendations, and the agentic click of agents.
That is also where I realized this conversation could not stay inside marketing. When an agent makes the purchase, it becomes a client I have to satisfy directly.
The funnel now runs through machines that connect directly to the business itself. SEO therefore becomes part of something larger: assistive agent optimization, and ultimately AI-era business engineering.
To understand why, I need to connect the pieces. The framework explains why AI systems make the decisions they make and what shapes those decisions. When I apply those principles across the business, the goal becomes clear: organize the company so search engines, AI assistants, agents, and people can find it, understand it, recommend it, and buy from it.
Everything Builds On SEO
The process sits above the familiar disciplines I already work with: SEO, content, PR, paid media, and digital marketing. It helps me prioritize the actions that most affect recommendations and visibility.
Here is the part every SEO should value: assistive agent optimization is built on SEO. It does not replace it.
I think of it like a Russian doll. SEO sits at the center. It draws from the open web, the same crawled and indexed foundation search has always used.
At that core are two parts of the algorithmic trinity: the search engine, which indexes and ranks information, and the knowledge graph, which stores entities and the relationships between them.
The next layer is assistive engine optimization. It adds the third component: the large language model. The LLM provides reasoning, grounding, and conversation.
Instead of returning only a list of links, it evaluates corroborating evidence and answers the user directly. This layer builds on traditional SEO with entity corroboration, machine-readable proof, and signals that help AI systems understand what content actually means.
The outer layer is the agent. It introduces what the layers below it never had: direct access to business systems through protocols such as MCP. An agent can check inventory, compare prices, and complete transactions without visiting a page or clicking through a search result. This is where AI stops recommending and starts acting.
Each layer depends on the one beneath it. The stronger the SEO foundation, the more effectively I can build everything above it. That makes SEO more central to digital marketing, and to the business itself, than it has ever been.
If I understand how machines read the web, I hold the foundation every other AI-facing initiative depends on.
The Funnel Has Not Changed, But The Build Direction Has
The acquisition funnel has not fundamentally changed since marketers first drew it in the 1800s. Awareness still sits at the top, consideration in the middle, and decision at the bottom. The customer still moves downward while the brand tries to catch them. What has changed is where I have to stand to catch them.
Traditional marketing stood in front of people in the real world, on billboards, shelves, and stages. Digital marketing did the same online through SEO, paid search, social media, and content. AI-era marketing extends that logic again.
Now I have to stand where I always stood and also inside the AI engines. Those engines put brands in front of buyers, present the best solution, and increasingly make the purchase.
The modern buyer mixes all three modes in a single purchase, so I have to be present in all of them. The client still travels from the top of the funnel down, but the engines learn from the bottom up. That is how I need to build for them.
Marketers draw the funnel top-down because that is the customer path. But businesses have always had a reason to read it the other way. Winning the result for your own name is the cheapest and highest-converting move because it reaches the warmest traffic: people already at the door.
I have made that case since 2012, when I started working on brand SERPs. Your name is the one search result you can most completely own, yet the industry ignored it for years.
Comparison and consideration queries come next because they sit near the purchase, where buyers are most likely to convert. Awareness is the last thing I build, because those people often do not yet know what they want or what the solution might be.
The engines make this flip unavoidable. Search engines let users move between sites on the way down the funnel, so top-down building could still work. Assistive engines pull the funnel inside themselves. Now I build from the bottom up because that is how the machine learns who to trust.
Agents push this even further. The funnel goes dark, and the choice often goes with it. Each step takes more of the journey out of my hands, and each rewards the same brand: the one built from the bottom up.
The Agentic Spectrum Decides How Much Must Change
Two ideas tell me how much of a business has to change. The first is the delegation boundary. The second is the agentic spectrum.
The delegation boundary is the micro view. It tracks how much of one buyer journey, from searching to comparing to choosing to buying, a person hands to a machine.
The agentic spectrum is the macro view. It asks what share of the clientele has gone agentic and how quickly that share is growing.
The micro view tells me how to win one buyer in the moment. The macro view tells me how much of the business has to change to keep winning buyers over time. This is the number I would start measuring first.
Here is why it reorganizes the business, not just the marketing. When the agent makes the purchase, it becomes a client I have to satisfy directly, even as it acts for the person behind it. It answers to one priority: keeping its own user happy.
That means the sale turns on confidence. Can the machine trust the business to meet the need and keep its client satisfied?
That confidence has to clear a much higher bar than search or assistive engines required. It runs across the full funnel. If I earn it across the stack, I become the brand the agent buys from.
Preparing for that is what AI-era business engineering means. Pricing, qualification, product data, checkout, service, and retention all need to be built so an agent can transact as cleanly as a person can.
The agent navigates the whole funnel on its own. I have to convince it at every stage, from awareness to the final yes, while getting almost no visibility into the journey. What I do get is granular measurement at negotiation and transaction stages. The agent tells me what it wants, and I either satisfy it or I do not.
That is why I need to build the business to work cleanly with agents and people alike, from the top of the funnel to the moment the deal is struck.
Translating what a company does for humans into something machines can read and act on used to feel optional. Ignoring search engines and assistive engines was never wise, but many companies survived it. In the age of agents, ignoring the engines hands a growing share of the clientele to competitors.
Your Untrained Salesforce Is Already Selling
Every business now has a salesforce it never hired: Google, ChatGPT, Perplexity, Claude, Copilot, Siri, Alexa, and many more. The number keeps growing as major tech platforms add AI answers inside social media, video, search, operating systems, and workflow tools.
The apps people already use now embed assistants that recommend tools, vendors, and products. A buyer does not need to open a separate AI engine for this to happen.
Those engines reach prospects in explicit, implicit, and ambient ways. However they appear, the outcome is the same: they work around the clock, speak to prospects in rooms I will never see, and decide whether to recommend me or a competitor.
The default state of that salesforce is untrained. If someone asks about my category, it answers with the brands it happens to understand, and that may not be mine. It may hedge on basic facts, confuse the brand with a namesake, cite proof that does not exist, recommend the wrong use case, or name a competitor at the exact moment the user was looking for me.
The cost is real, but it often never appears on a dashboard. I cannot watch the AI research the brand, evaluate it, recommend it, or talk a buyer out of choosing it. It all happens inside the machine. That is why I pay attention to three taxes: invisibility, ghost, and doubt.
AI engines recommend the solution they are most confident in, and that is not always the best solution. It is often the one they understand best. The recommendation depends on what they grasp and how confident they are in it.
So if my solution is truly the best, I have to train them. I have to educate them and brief them. They answer to the user, and my client is their client. They retain that client by surfacing the strongest solution they can see.
The practical question is simple: have I made it unmistakably clear that I am the best answer to the specific problems I solve, for the ICP I serve?
Three Taxes Quietly Cost Recommendations
I pay a tax at every stage of the funnel for as long as this AI salesforce is not working explicitly in my favor.
Someone types the brand name directly into an engine, and instead of a clean answer, it hedges with phrases such as “claims to be,” “reportedly serves,” or “says on its website.” Worse, it may start offering alternatives.
Search engines usually do that only when a competitor pays heavily to appear on the brand SERP. Otherwise, the brand owns its own name.
AI can raise the alternative on its own, purely because it is uncertain. That is why brand SERP and AI résumé protection are no longer optional.
That hedge and nudge are the doubt tax. I pay it when the engine lacks enough independent corroboration to commit. It sits at the understandability layer, and the cost is every prospect who came looking for the brand by name and left with doubt.
The ghost tax appears when a prospect asks the engine to compare the category and name the best options. The engine lists several brands, but mine is missing. It knows I exist, yet it does not surface me because its confidence in my credibility is too low.
The invisibility tax appears at the top of the funnel. Someone asks a question I am well qualified to answer, and I am nowhere in the response because the engine never identified me as belonging in that conversation. I never see it because the conversation ends without me.
I need to track these taxes across every engine and every layer, and I should not use only my own account. It is biased toward me. The right approach is proper tracking, neutral testing, and better questions.
The funnel query pathway is the best way to read this over time and across the web. What I am measuring is leakage at each layer. Because the system is opaque, I read the macro trend rather than overreacting to one response.
Then I build from the bottom up and clear the taxes in revenue order.
I clear the doubt tax first because it affects the warmest traffic.
I clear the ghost tax next because it affects buyers comparing close options.
I clear the invisibility tax last because it sits furthest from the purchase.
That is the funnel flip again. AI engines have turned the old top-down playbook upside down.
The Algorithmic Trinity Is Where The Work Lands
I train the AI salesforce in three places, and I need to be present in all three for that training to hold.
Large language models do the reasoning at the moment of the query. This is the intelligence layer: ChatGPT, Claude, and Gemini.
Search engines index and rank fresh content. This is the information layer: Google and Bing.
Knowledge graphs store entities and verified relationships. This is the verification layer: Google’s Knowledge Graph, Wikidata, and Bing’s entity graph.
Those three layers are the algorithmic trinity.
I may be aiming at dozens of platforms and surfaces where this salesforce appears, but there are only a few machines at the root. At mass-market scale, the practical LLM list narrows quickly to ChatGPT and Gemini. There are two major web indexes, Google and Bing, and two major knowledge graph owners, Google and Bing again.
Everything I train reaches back to the same small set of underlying systems. The corroboration work I do for one engine often strengthens the foundation for all of them.
That is why the effort compounds. The knowledge graph confirms the entities the LLM reasons about. The search engine surfaces the fresh content the LLM grounds on. The AI salesforce becomes fully trained when all three converge on the same answer about the brand.
That convergence is where I win: independent systems reaching the same conclusion about who I am, what I do, who I serve, and why I am credible. When I give them that picture in detail, they can hold it with confidence.
At that point, the trinity can surface the brand at the bottom of the funnel, recommend it over competitors in the middle, and advocate for it at the top across search engines, assistive engines, and agents.
The results vary because each platform mixes technologies differently, but the direction starts to favor the trained brand.
Google owns all three layers and remains the dominant force across search and assistive engines, so it remains the main target.
I am not suggesting that I ignore smaller players such as Claude or DuckDuckGo. They matter to the audiences that use them. But for most brands, users, and SEOs, the major public engines are still the key to commercial success.
A tight digital footprint, cleaned up and optimized on-site and off-site, feeds the trinity. At mass-market scale, that means Gemini and ChatGPT, Google’s and Bing’s knowledge graphs, and Google’s and Bing’s search indexes.
The useful side effect is that this strategy also helps with smaller players.
Third-Party Proof Is What AI Believes
Knowing where the work is ingested is only half the job. I also need to know which evidence the AI salesforce believes. Not all evidence carries the same weight, and the gap between weak and strong proof is often the differentiator.
The weakest evidence is what a brand publishes about itself, in its own voice, on its own properties: homepage copy, about pages, and product descriptions. I call this first-party evidence. It is a claim and a baseline, but it proves little on its own because the engines know who wrote it.
If I surface a client outcome, case study, or customer review on my own off-site channel, I move up to second-party evidence. The substance is no longer entirely my assertion, even though I still control the publish button.
Then there is evidence I had no hand in publishing: clients and partners describing their own experiences, an independent journalist’s article, an analyst report, or coverage controlled entirely outside my reach. That is third-party evidence, and it is the strongest proof the salesforce can read because I could not directly shape it.
It is also the category many brands lack because it requires real-world activity, not just publishing. First-party claims, second-party corroborates, and third-party proves. Without proof, nothing stands.
Three Levels Of Effort Create Different Outcomes
Most brands sit at the bottom without consciously choosing to. The minimum-effort brand keeps a website, runs some content marketing, responds to occasional mentions, and otherwise lets the ecosystem do what it does. It appears in machine-readable form but does not shape that form.
Because minimum effort is treated as normal, many companies land here and never recognize it as a decision. Their AI salesforce is barely trained.
The next level appears when a brand notices specific problems and fixes them: an incorrect fact in an AI Overview, a competitor outranking it for a query, or a structured data gap. Those fixes help, and the brand becomes better positioned.
But the work is still symptom-driven. It patches what breaks loudly without building the discipline that prevents the next break. The salesforce is partially trained, but problems are driving the strategy.
The systematic brand runs an operational discipline against the pipeline every week: entity home maintenance, evidence harvested from service teams, machine-readable proof, distribution across publication tiers, and continuous monitoring of the brand SERP and AI résumé.
Most companies are not organized to make that happen naturally. But if I can harvest, codify, and distribute the evidence created by business operations, I can train the AI salesforce to work in my favor around the clock.
I would start from the entity home. I would organize the brand SERP and the AI résumé, then optimize the digital footprint wherever the brand appears. That is understandability, and it is the most important first move.
With the core entity locked, I can build credibility on top of it through engagement, reviews, client feedback, PR, and evidence that the business is genuinely good at what it does.
Deliverability follows because work on the brand SERP and AI résumé already strengthens credibility and reach. Then I can spread the same discipline across every entity the company owns: products, services, and people.
For each entity, I need the right content, presence where the audience is looking, a path down the funnel, and a clear connection back to the entity home. I need to walk the walk and apply the mirror principle.
The Salesforce Is Already Working
In 2026 and beyond, the AI salesforce operates inside the supply chain as well as the sales funnel. AI sits at the gates that decide whether to include a brand in what it knows, whether to deploy it in an answer, and whether to reselect it after every transaction.
Every outcome customers experience feeds back into the system for the next prospect who has never heard of the brand. That is the convergence this series has been pointing toward. The salesforce is selling 24 hours a day, for the brand or for a competitor. The difference is how well it has been trained.
This is why I see the discipline as AI-era business engineering, not just AI-era marketing. It is not a content tactic. It is a reorganization of how the business operates so pricing, qualification, product presentation, sales, retention, and customer success all create machine-readable evidence as a byproduct of doing the job.
SEOs Are In The Best Seat In The Room
When I speak with entrepreneurs and CEOs, I use nine questions to show where the company stands.
Tech, bottom to top: Is our entity home locked down so engines have one source of truth about who we are? Is our structured data complete enough for them to verify what we claim? Are we discoverable across every engine when topical questions appear?
Marketing, bottom to top: What does our brand SERP look like today, and what does the AI résumé say when engines are asked about us directly? Where is our third-party corroboration weakest, and what are we doing about it this quarter? Which topical territory do we own in the engines, and which territory do we want but not yet hold?
Branding, bottom to top: Does our brand story match what AI is currently saying about us, and where is the gap? Are our client outcomes being engineered into machine-readable evidence, or are they dying in CRMs and quarterly retrospectives? Are we placing proof now for the categories we want to own in three years?
All of those questions run from the bottom up, which is ironic because marketers usually work the funnel from the top down. The customer is the one moving from top to bottom, looking for a solution.
So I take a step back and read the funnel from the bottom up. Everyone is building the same thing: understandability, credibility, and deliverability. They are just approaching it from different ends.
The business builds from the foundation up: know who you are, know who you serve, become credible, then reach the right people.
The marketer wants the maximum audience and starts with reach, then works down to who the brand is and why it should be trusted.
AI starts at the bottom. Who are you? Are you credible? Only then will it put the brand in front of more people.
The SEO is the person who can see that it is all the same system. I understand that I must work from the foundation up, the way the machine does, and then meet the customer coming down from the top.
I should build for the customer, but work upward toward them. That has always been the stronger approach, and AI engines have now made it obvious.
The business now has two kinds of clients: the human and the agent. I need to speak to both. The agent is emulating a person and reflecting the world’s view of the brand, so pleasing the agent and pleasing the human are closely connected.
That is what makes SEO impossible to sideline. I am well positioned to tell the business and the marketers what must change to satisfy the agent without losing the human.
Whether agents represent 5% of the business today or nearly all of it, the agentic share will grow year after year. That means I have to step out of the SEO corner and look at the wider business. I am in a rare position to see business, marketing, and machines at the same time.
The audience used to be only human. Now it includes machines, too, and I am the one who can speak to both.
This is the 19th and final piece in my AI authority series, and it has been a long journey. My thanks to Danny Goodwin, Angel Niñofranco, and the Search Engine Land team for their immense support throughout.
When I started, the framework was a complete idea, but I had not fully worked through all the details. Week by week, I worked through each of the 15 gates, and every one turned out to be more intricate, more in-depth, and more thought-provoking than I expected.
What I have finished is a practical framework for SEO, marketing, and business in the AI age, one that search professionals, marketers, and business leaders can apply to real business problems.
Series Index
Parts 1 through 18 built this framework step by step: cascading confidence, assistive agent optimization, the AI engine pipeline, infrastructure gates, competitive gates, the entity home, the push layer, annotation, topical ownership, the funnel flip, the framing gap, pipeline repair, the delegation boundary, funnel query pathways, macro measurement, customer-success proof, AI opinion formation, and the collapse of paid and organic visibility across AI surfaces.
Have you ever wondered how AI search platforms have evolved from simple Retrieval-Augmented Generation (RAG) to sophisticated agentic systems? These days, AI search has advanced beyond mere RAG, transforming into something far more complex and dynamic. In this article, I’ll guide you through how today’s advanced AI retrieval systems determine if your content is showcased or left in the shadows.
About two and a half years ago, I penned an article for Search Engine Land on how RAG represents the future of search. It wasn’t just a reactionary measure from Google in response to ChatGPT, but rather an architecture in development since the REALM paper in August 2020. Observing developments since then, everything has aligned with what I speculated.
The RAG pipeline of the past, which I outlined as a query transforming to an answer with citations, is already outdated. Major AI search platforms like Google AI Mode and ChatGPT Search have transitioned to a more complex architecture. They now possess planning capabilities, tool-routing options, and iterative retrieval methods that continuously refine results until they reach a suitable conclusion. The one-retrieval-to-answer model is defunct.
This sophisticated approach is what we now refer to as agentic RAG, a framework that’s become the industry standard. If your content strategy still relies on single-shot retrieval, you’re optimizing for a non-existent system. What’s more, in agentic RAG, you can’t witness the gatekeeping process—only the final outcome shows if your content made it.
By the time you finish reading, you’ll have a functional understanding of agentic RAG, the patent evidence showing its application by companies like Google, insights into what each major platform is doing, and concrete tactics to enhance your content strategy. You’ll also gain my important takeaway of the year: the future hinges on model distillation.
The October 2023 perspective is still relevant. Passage-level retrieval remains essential to relevance, and knowledge graphs work in tandem with LLMs. Search systems aim to lower what are known as Delphic costs, minimizing the effort users expend to find answers. Google’s guiding principle has always seen traffic as a means rather than an end. This aspect of my argument needs no change.
What has evolved is the structure of the retrieval pipeline. Back in 2023, RAG was straightforward and linear. A query was encoded, top passages were retrieved, and an answer was generated. If your content was within the top set of results, you had visibility; if not, you were invisible. This framework served its purpose at the time.
Today’s pipelines boast abilities absent from linear models: planning, tool usage, multi-hop iteration, and reflection. Rather than being a single occurrence, retrieval now involves up to twenty sub-retrievals orchestrated by a central agent, which refines its foundation of evidence continuously before finalizing an answer.
My earlier writing hinted at these upgrades without naming them precisely.
The word “agentic” is used liberally, but its structural definition is specific. Understanding agentic RAG requires grasping four properties each system must embody to wear the label.
1. Planning involves restructuring the user query into a research plan, breaking it down into sub-queries, pre-selecting tools, and strategizing retrieval sequences. The system doesn’t just respond; it plans each step with precision.
2. Tool usage extends beyond basic retrieval to include inquiries through APIs, code execution, live web browsing, and more. The agent selects the best method for each task, weaving these tools into cohesive outputs.
3. Iteration or multi-hop retrieval is where the agent refines its findings by visiting the source multiple times, continually improving the evidence base.
4. Reflection involves the agent critiquing its own output, determining its sufficiency and quality, and retrieving more information if needed to resolve discrepancies or improve source diversity.
These are the qualities that set agentic RAG apart and what make it the new default for AI search platforms.
Drawing a contrast between the classic RAG and agentic RAG, imagine the former as a direct process and the latter as a comprehensive loop where steps can be revisited until the solution is optimal. This is what my content needs to withstand.
The six shifts required for effective content engineering in the realm of agentic RAG are clear. I need to optimize for a spectrum of sub-retrievals, present well-structured and cohesive passages, leverage bridge entities, offer tool-callable content, and ensure freshness within my content.
The path forward involves navigating measurement’s increasingly complex landscape with the aid of model distillation. By understanding the full lifecycle from internal query generation to external execution, I can effectively target content positioning and citation strategy.
Engaging with this agentic environment demands observation, adjustment, and perpetual calibration. The choice is simple: evolve to survive and thrive or remain static and risk obscurity.
From February to May 2026, I dove deep into the fascinating world of agentic AI adoption. I explored how it’s being embraced by enterprises, mid-market players, and SMBs across the U.S. and worldwide. By gathering insights from top consulting firms like McKinsey, Gartner, and IDC, as well as academic institutions and AI leaders, I pieced together a comprehensive overview of agentic AI’s current landscape.
This report fuses insights from over 30 research efforts and industry surveys, covering 15,000+ businesses. It provides a granular look into how businesses are integrating autonomous AI agents this year, breaking it down by company size, industry, deployment stage, primary use cases, and adoption and abandonment patterns.
*Statistics are based on data up to May 14, 2026, unless indicated otherwise.
While generative AI generates immediate outputs, agentic AI shifts the way systems function entirely. This piece zeroes in on agentic AI’s adoption, defined as follows:
Agentic AI revolves around AI systems autonomously planning, deciding, and executing complex tasks from beginning to end.
The term adoption signifies any case where an organization uses at least one agentic AI system at any stage, from initial trials to full-scale implementation.
Meanwhile, abandonment involves halting an agentic AI program or specific projects. This doesn’t always mean closing an organization’s entire AI operations, as they might continue other initiatives.
Agentic AI adoption significantly varies by organization size. A breakdown of recent adoption rates across different segments unveils fascinating trends.
As I dug into the data, I discovered enterprises are leading the way with 25% adoption, thanks to their resources and AI budgets. However, smaller sectors, like mid-market firms and SMBs, are catching up fast. Their year-on-year growth rates are even outpacing those of enterprises!
I predict that SMBs and mid-markets will continue adopting agentic AI faster than their larger counterparts. This trend is partly driven by accessible solutions such as Salesforce Agentforce and Microsoft Copilot Studio, which empower companies with tighter budgets. In contrast, enterprises face challenges due to their intricate systems and diverse data environments.
Agentic AI deployment spans various maturity stages, presenting unique challenges depending on available resources. For SMBs, scaling can be costly, making it particularly challenging.
The table showcases deployment stages among adopters, revealing that 62% of enterprises, despite higher resources, linger in the experimentation phase. Notably, only 13% achieve full deployment.
A few patterns stand out from the data:
Firstly, experimentation predominates across sizes, with a 56% average gap to partial deployment. This highlights caution across sectors in deploying agentic AI.
Despite enterprises’ resources, mid-market companies are seeing greater partial deployment rates, likely due to fewer approval bottlenecks and more budgetary leeway compared to SMBs.
Also, scaling correlates with resources. Enterprises, despite early-stage phases, manage full-scale deployment at rates double those of mid-markets.
These patterns reveal that most organizations are still exploring, with few transitioning to production deployment.
It’s not all smooth sailing. According to Gartner, around 40% of agentic AI projects might be canceled by 2027, due to challenges encountered during deployment.
Although abandonment rates generally decline over time, mid-markets still see higher rates due to their broader range of obstacles and fewer resources compared to large enterprises.
Summarizing the common reasons for project failures:
Data quality matters. Without quality data, agents struggle, highlighting a universal need for centralized and uniform data pre-deployment.
Clear expectations are vital. Projects without well-defined success criteria often fail to demonstrate value, risking cuts in resources when results are inconspicuous.
Costs weigh heavily on SMBs. For SMBs, financial constraints dominate abandonment reasons, overshadowing other factors. Mid-market firms display more varied primary drivers.
Such insights explain why full implementation is elusive for many, despite significant investments. Companies need to address multiple challenges concurrently to progress beyond experimentation.
On an industry level, exploring adoption across sectors shows where agentic AI thrives and lags. Regulatory factors, data readiness, and competitive dynamics result in differing adoption levels.
Industries like education, construction, and real estate lag, owing to budget constraints, less advanced data infrastructures, and fewer automation opportunities. Nonetheless, even these sectors demonstrate notable enterprise adoption, signaling a broader reach beyond tech and financial services.
Finally, examining use cases underscores where agentic AI is making headway. Customer service and supply chain coordination rank high due to their structured processes. On the other hand, finance sees lower adoption due to stringent regulatory scrutiny.
If you fancy obtaining a PDF copy of this insightful report or learning more about our work, feel free to reach out here.
For further exploration into agentic AI and its surrounding trends, consider delving into the following reads:
Have you ever wanted to customize your Google Search experience? Now you can build your own apps right within Google Search.
I discovered this amazing feature powered by Google Antigravity and Gemini 3.5, which lets me set up a search feature that delivers exactly the kind of information I need, formatted just how I like it, and sourced from where I trust.
During this year’s Google I/O, Liz Reid, head of Google Search, unveiled this innovation. She mentioned, “Search can build the ideal response, in the right format for your question – completely on the fly. You’ll get custom generative UI, including visual tools and simulations, tailored to your needs.”
Exciting Examples
Imagine creating custom layouts for understanding astrophysics or how your wristwatch works. Google assembles interactive visuals, tables, and real-time simulations to suit your learning style.
I’ve also been able to manage ongoing tasks like wedding planning or home moves with customized dashboards that act as helpful companions throughout the process.
Let’s not forget fitness! I asked Google Search to build me a custom fitness tracker. It taps into live data like weather and reviews to keep me on track, making my health goals more achievable.
Visualizing the Experience
These custom search experiences, including generative UI examples, will become widely available this summer. I’m particularly excited as they roll out first to Google AI Pro and Ultra subscribers in the U.S.
Why This Matters
It’s groundbreaking to have the ability to code mini apps within Google Search, answering questions in ways that are uniquely mine. It’s a level of personalization I’m thrilled about, achievable only through such advanced generative-AI tools.
I’m excited to share that Google has announced some transformative updates to its search capabilities. These updates include the introduction of information agents and enhanced agentic experiences that will elevate how we interact with search. Google’s AI will continuously scan the web, ensuring we receive the most current information, much like a personal assistant would.
In a recent announcement, Google revealed new search agents, focusing on information agents and additional agentic functionalities within Google Search. These information agents are designed to monitor the web for changes to our tasks, seamlessly supporting us on our journey through various challenges and questions.
Liz Reid, the head of Google Search, stated, “We’re entering the era of Search agents, where you can easily create, customize, and manage multiple AI agents for your many tasks, right in Search.” This new era provides a unique opportunity to tailor search experiences to our specific needs.
Information Agents. These agents are designed to keep us informed about our questions and tasks. Google’s agents will intelligently sift through the internet—exploring blogs, news sites, social posts, and accessing the freshest real-time data on finance, shopping, and sports, to ensure we receive the most relevant updates on our inquiries.
The information agents will then compile an “intelligent, synthesized update” that not only provides the necessary information but also enables us to take action.
The Example. Envision yourself apartment hunting. You can simply input all your specific requirements, and your agent will continuously scan listings, alerting you whenever a match surfaces. Similarly, if you’re keen on not missing any sneaker collaborations from your favorite athletes, your agent will notify you about new releases.
Availability. These exciting capabilities are set to roll out this summer, initially available to Google AI Pro & Ultra subscribers.
Agentic Experiences. Google is also extending its agentic booking capabilities within Google Search to encompass new tasks like finding local experiences and services. Imagine effortlessly booking a private karaoke room for an exact time and with specific food options, all handled by Google Search.
Google will provide the most current pricing and availability information, along with direct links for purchase, streamlining experiences across various services, including home, repair, beauty, and pet care. These features are expected in the U.S. this summer.
Personal Intelligence Expanding. In addition, Google has revealed plans to broaden its Personal Intelligence feature within AI Mode, now reaching around 200 countries and territories, supporting 98 languages.
