Your best-fit buyer just asked ChatGPT who to hire. Your name wasn’t in the answer.

Here’s the analogy I use with founders on the first call. Imagine your category is a dinner party, and the AI is the host introducing guests to each other. The host doesn’t read your website. The host reads a name tag. Whatever is on that name tag — two lines, maybe three — is what everyone else in the room hears about you.

Most B2B sites do not have a name tag. They have a mood board. Clever taglines, vague headers, product pages padded with adjectives that were written to make a human feel something. Language models don’t feel anything. They parse. They look for clean, structured statements they can lift into an answer without having to guess.

The first time we ran this test on a client — a Series B SaaS company with a very good product — we asked ChatGPT to describe them in one sentence. It came back with something like “an all-in-one platform that helps businesses grow.” That was their tagline. Verbatim. The model had nothing else to work with, so it just handed the tagline back. Meanwhile, their two closest competitors got described in specific, verb-driven sentences that actually explained the job they did.

So the first thing we do is boring and technical, and it is the thing almost nobody wants to pay for: we rewrite the parts of your presence that a model reads first. Schema markup that says exactly what you are. Entity definitions that tie your brand to the specific problem, category, and buyer you serve. An llms.txt at the root of your domain that acts like a briefing document for every crawler: this is who we are, this is what we sell, this is who it’s for, these are the questions we’re a real answer to.

Then we go through your top product and category pages and quietly rewrite them into a structure the model can actually quote — short, factual Q&A blocks, comparison paragraphs with clear subjects and objects, definitions that read like an entry in a reference book rather than a headline in an ad. This is not more content. It is content the model can pull cleanly into an answer without having to invent the words itself.

The result, three to six weeks in, is that you stop being described as “a platform” and start being described as the specific answer to the specific question your buyer is typing. That single shift — from adjective to answer — is what puts your name back into the running.

If step one is your name tag, step two is your reputation in the room. The model doesn’t evaluate you in isolation. It evaluates you inside a graph — a web of who talks about you, in what context, next to which competitors, using what language, and with what sentiment. When that graph is thin, out of date, or dominated by the wrong signals, no amount of on-site work will move you into the answer.

Think of it the way an executive recruiter thinks. Before they put your name on a shortlist, they don’t read your CV. They call three people who’ve worked near you and ask what you’re actually like. The model is doing the same thing, at scale, in milliseconds. It’s asking: which sources already have opinions on this brand, and do those opinions agree?

This is where most “AI SEO” services quietly fall apart. They audit your website, hand you a checklist, and stop. But your website is maybe 20% of the signal. The other 80% lives on third-party surfaces — category pages, review sites, comparison articles, industry roundups, forum threads, podcast transcripts, YouTube descriptions — the sources retrieval systems weight most heavily when they piece an answer together.

So we start with a map. We plot your brand, your competitors, and every source the model is currently reaching for when it answers questions in your category. Then we score each node: is it there, is it missing, is it saying the wrong thing, is it saying nothing at all? That map becomes the plan for the quarter — a specific, prioritized list of citations we need to earn, refresh or reframe.

Then we go do the work. Not spam. Not paid placements dressed up as editorial. We pitch the analysts who write the category pages. We contribute to the comparison articles that already rank. We get you into the round-ups your buyers screenshot and send to their boss. Everything auditable, on real domains, in a way that stands up to a general counsel review.

After a few months of this, something clicks. The model starts describing you inside the same “neighborhood” as the category leaders, because in its training and retrieval signals, you now are. When a buyer asks “top providers of X,” you’re inside the small set of names the model is choosing from — not fighting to break in from outside.

The last signal is the one most teams underrate, and it’s the one that compounds hardest. Independent human conversation about your brand. Honest reviews. Comparison threads on Reddit. Unprompted recommendations in Slack communities, in Substack posts, in podcast rundowns, in the “what do you use for X” question that gets asked in every industry group every week.

These are exactly the sources retrieval systems trust most, because to the model they look unpaid and unedited. A Reddit thread where three strangers agree your product is the right pick for a 20-person team is worth more, to a language model constructing an answer, than a page of glossy testimonials on your own website. It’s the same reason you trust a friend’s offhand recommendation more than an ad — the model was trained on the same instinct.

Here’s the important part, and I want to be blunt about it. We do not run fake accounts. We do not astroturf. We do not have an intern paid $6 an hour to post “have you tried [client]?” under every question in r/SaaS. That work is easy to spot, it gets caught, and when it gets caught it poisons the exact graph you’re trying to build. It also, quietly, makes the founder’s life worse, which is not a trade we’re willing to make.

What we do instead is much less glamorous. We find the conversations already happening in your category. We equip your customers, your operators, and your genuine advocates with the language and context they need to participate honestly. We help the analysts and creators who cover your space discover that you exist and are worth writing about. We turn the on-record wins you already have — a great review, a case study, a product launch — into artifacts that other people actually want to cite.

That’s the loop. Each new mention strengthens your entity graph, which makes the on-page work land harder, which earns you more citations, which seeds more conversation, which produces more mentions. The reason the third client for a good GEO program always looks easier than the first isn’t luck. It’s that by then, the model is already halfway convinced.