The AI Buyer Is Already Here: How Agents Are Quietly Filtering Your Pipeline
Tracy Thayne6/2/2026·8 min read
A marketing leader I work with ran a clean experiment a few weeks ago. She asked ChatGPT, Perplexity, and Gemini the exact question her best customers ask before they buy: "What are the best tools for X, and how do they compare." Her company is a real category leader with strong reviews and a recognizable name. It showed up in one of the three answers, halfway down, described in language she would never have chosen.
Her competitor, a smaller and frankly weaker product, led all three.
Nothing was wrong with her marketing. Her problem was that her marketing was written for a reader who was no longer the first one in the room. Before any human on her buying committee saw her site, an AI had already read the category, formed an opinion, and built a shortlist. She was on it once out of three times, and not at the top.
This is the shift most B2B teams have not absorbed yet. You are no longer marketing only to people. You are marketing to the machine that screens you before the people ever arrive.
The First Reader Is Not Human Anymore
The numbers here moved faster than most teams updated their assumptions. G2's 2026 research found that 51% of B2B software buyers now begin their purchasing process in an AI chatbot rather than a search engine, and that AI chatbots are the single largest source influencing which vendors make a buyer's shortlist. Gartner has projected that by 2028, AI agents will intermediate roughly 90% of B2B buying, channeling more than $15 trillion in spend.
The part that should stop you is what happens after the AI gives its answer. The same G2 study found that 69% of buyers chose a different vendor than they had originally planned to, based on AI guidance, and 33% bought from a vendor they had never heard of before the AI surfaced it. That is not a tool helping a buyer execute a decision they already made. That is a tool making the decision and handing the buyer a conclusion.
So there is now a reader in your funnel who reads everything, never fills out a form, never shows up in your analytics, and quietly decides whether the humans behind it ever consider you at all. Most teams are not marketing to that reader. They are not even accounting for it.
Subscribe
Get notified by email when we publish a new post. No spam — unsubscribe anytime.
The Agent Is a Persona, and You Do Not Have One for It
In Why Your Buyer Personas Are Probably Wrong, I argued that static personas built from assumptions fail, and that a persona only matters if it is operational, tied to how a real buyer actually moves through a decision. That argument still holds. It is just incomplete now.
Every persona on your wall is a human. The economic buyer. The champion. The technical evaluator. The skeptic in procurement. What none of them captures is the new first-pass reader: an AI intermediary that evaluates you before any of those humans do, on criteria none of those humans use.
The agent does not skim your homepage for vibes. It does not respond to a clever headline or a tasteful gradient. It reads for structure, specificity, and consistency. It wants to know exactly what you do, exactly who you serve, exactly how you compare, and whether every source it can find about you agrees. It is, in effect, the most literal-minded analyst on the buying committee, and it goes first.
If your personas describe only the humans, you are designing for the second reader and ignoring the one that screens. The agent is a persona. It has needs, it has evaluation criteria, and right now most teams have built nothing for it.
What the Agent Actually Screens For
When I map how these intermediaries evaluate a category for an Expona customer, the same three filters show up every time. They will look familiar to anyone who read Generative Engine Optimization: When AI Becomes the Buyer's Front Door, because the agent in the funnel and the generative engine at the front door are the same kind of reader applying the same kind of test.
Specificity beats persuasion. The agent cannot be sold. It can only be informed. Vague positioning that asks a human to feel something gets averaged into nothing. A precise, structured statement of what you do, who it is for, and what it replaces gets lifted directly into the answer. Persuasion is for the human in the room. The agent that decides whether the human ever sees you rewards only clarity.
Comparison is the unit of evaluation. Buyers no longer ask AI "tell me about Vendor A." They ask "compare the options for X." The agent answers in a comparative frame whether or not you participate in it. If you have not clearly defined how you differ from the alternatives, the agent invents a comparison for you, usually an unflattering one built from whatever scraps it can assemble. The vendor that defines the comparison tends to win it.
Corroboration decides trust. The agent does not take your word for anything. It cross-references. G2's research found that 45% of buyers say citations from third-party review sites are the single most confidence-inspiring signal in an AI-generated recommendation. The agent trusts what multiple independent sources confirm, and it discounts what only you claim. If your site says one thing and every other surface says something vaguer, the agent resolves the conflict by trusting you less.
Notice that none of these is a messaging trick. Each is a property of how operationally sound your information actually is. The agent is not grading your copy. It is grading your context.
This Is the Same Problem, Pointed One Layer Earlier
Here is where this connects to everything I have been writing this spring.
In Context Is the Whole Game, I argued that context is the actual locus of value in AI-driven marketing, across three layers: content, structure, and execution. GEO was that same context architecture, pointed outward at the discovery layer. The AI buyer is that same architecture, evaluated one step further into the funnel, at the moment a shortlist gets built.
The agent screening your pipeline and the generative engine answering the category question are reading from the same corpus and asking the same questions. Is this clear. Is this specific. Does this hold up against every other source. A company with a coherent context layer passes both tests with the same investment. A company with a sloppy one fails both, and never sees the failure, because the buyer it lost was filtered out by a reader that leaves no trace in the CRM.
This is why I keep returning to context as the unit that matters. The work you do to make your category position clear and consistent is not a GEO project, or a persona project, or a content project. It is one project that pays off at every layer where an intelligence reads you. As I argued in RAG Is Only Half the Story, ingestion quality determines output quality. The agent in your funnel is just one more reader whose output depends entirely on what you gave it to ingest.
What to Do About It This Quarter
Start by adding the agent to your persona set. Write it down like any other buyer. What it reads, what it rewards, what makes it discount you. Treating it as a real participant in the buying committee is most of the shift, because it changes what you optimize for.
Then audit your category the way a buyer's agent would. Ask the major engines the comparison question your buyers ask, in their words, not yours. Read the answer as a verdict on your context, not as a vanity check. Where are you absent, where are you described wrong, where does a weaker competitor own the framing you should own. That gap is your actual pipeline leak, and it will not show up in any dashboard you currently run.
Finally, fix it at the source, not the surface. Do not write more content hoping to flood the agent. Write the clear, specific, comparative, corroborated version of your category position once, and make every surface agree with it. The agent rewards coherence, not volume. One consistent story beats ten inconsistent assets every time.
The Takeaway
The AI buyer is not a forecast. It is already in your funnel, reading everything, screening silently, and handing your human buyers a shortlist you had no idea was being assembled. Most teams are still optimizing for the second reader and losing to a first reader they have not named.
Your personas describe the humans. The agent goes first, and it does not evaluate you the way humans do. It does not want to be persuaded. It wants to be informed, precisely and consistently, and it filters out everyone who cannot manage that.
The teams that win the next two years will be the ones who market to the machine and the human as two distinct readers, and who understand that satisfying the machine is, underneath, just the discipline of having context clear enough to survive being read by something that cannot be charmed.
Tracy Thayne* is the founder of Expona, an AI-powered operational intelligence platform for B2B marketing. Read the Expona founder story or subscribe to the blog (below) for weekly insights on context, AI, and the future of marketing operations.*
