Be the Source, or Be the Guess

Somewhere in the last two years, the most valuable piece of real estate on the internet stopped being a page and became a sentence. A user types a question into ChatGPT, and the model answers in a paragraph, and inside that paragraph there is one clause — "tools like X" — and either your startup is the X or it is not. There is no page two. There is no scrolling. There is the clause, and the clause has room for roughly one name.

This is the part founders find genuinely disorienting, so it is worth stating the mechanism plainly. When someone asks an AI which project-management tool fits a five-person team, the model does not hand back ten blue links and wish you luck. It decides. It retrieves a handful of sources it trusts, synthesizes them, and produces an answer that reads like a recommendation from a knowledgeable friend. The user, more often than not, does not click anything. They just act — they sign up, they buy, they carry the fact forward as if they'd always known it. The click, that sacred unit of the old web, has quietly become optional.

That gap — five times the conversion rate — is the whole ballgame. It means a citation from ChatGPT is not a vanity metric adjacent to your marketing. It is the marketing, delivered by a narrator the user has decided to trust. Which raises the obvious question, the one this piece exists to answer: how do you get the narrator to say your name?

01 / The definitionWhat GEO Actually Is

The practice has a name now, which is how you know it's real: Generative Engine Optimization, or GEO. It is the work of structuring your content, your schema markup, and your earned-media footprint so that large language models cite your site when they answer a question. GEO is to AI search what SEO was to Google search — with one unforgiving difference. SEO gave you a ranking, a position in a list, a chance to be the third result someone begrudgingly clicks. GEO gives you a binary. You are in the paragraph or you are not.

Everyone in the space is now writing the same playbook, and out-citing the incumbents — the Averis and Coinbounds of the world — is less about a secret tactic than about actually running the boring parts consistently. So let's do the boring parts properly.

02 / The premiseCitations Equal Retrieval Trust

Here is the load-bearing idea. A model choosing whom to quote is casting a trust vote, and trust votes compound. Every time ChatGPT retrieves you on a topic and cites you, it reinforces a pattern in the systems that decide these things: this source is authoritative here. The reverse is also true, and quieter, and worse. Every time it cites your competitor instead, it learns that they are the authority and you are noise. Retrieval trust behaves like interest. The startups that start compounding it now will own the answer for years; the ones that wait will spend those years being the guess.

There is no page two in AI search. There is one paragraph — and either you are in it or your competitor is.

03 / The mechanicsThree Engines, Three Diets

The great oversimplification of GEO is treating "AI search" as one thing. It is at least three things, and each one eats differently. ChatGPT is a snob for the encyclopedic — Wikipedia alone is roughly 47.9% of its top citations — and its web search rides Bing's index, which means the single most overlooked GEO move is submitting your sitemap to Bing Webmaster Tools, not Google's. Perplexity, by contrast, is a creature of freshness and community; 46.7% of its sourced content comes from Reddit, of all places. Gemini pulls from Google's index and, tellingly, YouTube — so video schema and a real channel are levers there that do nothing for the other two.

The point is not to memorize percentages that will drift by next quarter. The point is that one footprint does not feed all three engines. A strategy that makes you delicious to ChatGPT may leave Perplexity hungry, and Gemini wants a food group — video — the other two ignore.

04 / The playbookThe Step-by-Step

Enough theory. Here is the sequence that actually moves citations, in the order that matters.

05 / The mistakePublishing Only When You Raise

And here is where nearly everyone goes wrong, in the same way, for the same understandable reason. A startup treats its newsroom like a confetti cannon: loaded once, fired at the Series A announcement, then holstered for a year. The founder feels the coverage, sees the LinkedIn likes, and mistakes a spike for a strategy.

But models retrieve the fresh and the frequent, and a single announcement in January is a very old fact by June. What the confetti-cannon startup has actually done is hand the model twelve months of silence — twelve months in which the question "what's the best tool for X?" gets asked thousands of times and answered, every time, with a competitor's name, because the competitor kept talking. The raise-day spike doesn't compound. It decays. GEO is a cadence, not an event, and the founders who internalize that one sentence are already ahead of most of their category.

Publishing only when you raise is like introducing yourself once and expecting to be remembered forever.

There's a tidy irony here. The big companies that seem to be everywhere in AI answers are not there because the models love them. They are there because they never shut up — a steady, unglamorous drip of announcements, profiles, product notes, and customer stories, structured and dated and shipped week after week. As YesPress likes to put it: the big companies aren't lucky, they're loud. And loudness, it turns out, is not a personality trait. It's a schedule. Which means it can be copied.