If you've ever asked ChatGPT or Perplexity to recommend a marketing agency, you've seen the whole game. Three or four names. No scrolling, no page two. As far as that model is concerned, in that moment, that's the entire market.

Most agencies never make that list. Not because the work isn't good enough, but because of how the list actually gets built. Here's what's happening behind the answer.

Retrieval, not ranking

Google ranks pages. ChatGPT and Perplexity work differently. When someone asks either of them a question, the model doesn't crawl the live web and rank everything in real time. It runs on retrieval-augmented generation: it breaks the question into a handful of search queries, pulls in a batch of candidate pages and mentions from a search index, then narrows that pool down to the few it's confident enough to cite.

Think of it less like a search results page and more like a fast, literal researcher skimming a stack of documents for the clearest, most confidently stated answer, then quoting three or four of them.

That's the whole mechanic. And it explains most of who gets found and who doesn't.

Domain authority still plays a part in that first pass, too - though less than the raw numbers suggest. 5W's AI Platform Citation Source Index found 68% of AI citations concentrate on roughly the top 15 publisher domains on the web. But that's mostly a story about mega-publishers like Wikipedia and Forbes, not individual brand sites: separate analysis from Clairon puts domain authority's actual predictive power over any single brand's citation odds at under 4% of the variance. For a site that isn't already a top-15 publisher, structure and specificity do more work than domain rating ever will.

What actually earns a place in that pool

A few things make a page more likely to survive the cut.

It's been said elsewhere first.

Earned mentions - press, other people's sites, directories, industry write-ups - carry real weight. Models apply a credibility filter before citing anything, and outside corroboration is one of the strongest signals in that filter. A line on your own homepage is a claim. The same line appearing on someone else's site too is evidence.

Your own site backs it up, clearly.

If your website says the same thing in plain, extractable language, that's the confirming signal. Vague "full-service creative partner" copy gives a model nothing to lift. A specific claim does.

It's consistent, not a one-off.

A single strong page rarely moves anything by itself. The same core claim, stated the same way across several pieces, is what turns a mention into something a model treats as fact rather than noise.

It's structured to be lifted.

Clear headings, a direct answer near the top, no PDF downloads, no JavaScript-only rendering a crawler can't see through. A model can't cite what it can't parse.

It's reasonably current.

Ahrefs' analysis of 17 million AI citations found AI-cited content runs about 26% fresher on average than traditionally ranked organic content - stale, undated content gets pushed down the list in favour of anything that reads as up to date.

None of this is about gaming a system. It's the same discipline as writing a good press release: say something specific, say it clearly, and make it easy to quote.

Why ChatGPT and Perplexity don't show you the same shortlist

Here's the part that surprises most people: getting cited by one doesn't mean you'll get cited by the other. Averi's analysis of 680 million AI citations found only 11% of domains earned citations from both ChatGPT and Perplexity for the same query set - a figure independently corroborated by Whitehat SEO's separate study of 118,000 responses. The two platforms lean on different retrieval sources and weight them differently: Perplexity leans harder on live web search, ChatGPT blends more of its trained knowledge with retrieval.

Practically, that means "visible in AI search" isn't one target. It's several, and they don't move together.

The scale of what's actually happening

Most B2B buyers are already doing this. Estimates vary depending on how a study defines "use" - a March 2026 multi-source analysis put the figure at 73% of B2B buyers using AI tools like ChatGPT and Perplexity during vendor research. Separately, Crackle PR's Q2 2026 AI Citation Benchmark found 51% of B2B tech brands have zero citations at all across ChatGPT, Perplexity and Gemini.

Most buyers are asking. Most companies aren't in the answer. For an agency, that's not an abstract marketing statistic. It's a new business pipeline forming without you in the room.

What actually changes the outcome

The fix isn't "publish more." It's narrower than that: a small, consistent set of claims about who you are and what you're genuinely good at, stated clearly, repeated in your own words across a handful of pieces, until a retrieval system has no reason to guess. Consistency, not volume and not a single clever page, is what turns "mentioned once" into "cited reliably."

Say something true and specific about yourself. Say it the same way in enough places. Structure it so a model can actually use it. That's the mechanism behind how agencies get found in ChatGPT and Perplexity.

A few common follow-up questions

Does this replace SEO?

No. Different mechanic entirely. SEO earns a ranking position against a query. This earns a citation inside a generated answer. Both are worth doing, and neither replaces the other.

Do I need to publish constantly to show up?

No. Consistency of a clear claim beats volume of content. A handful of well-structured pieces that all say the same true thing outperforms a large, scattered blog.

Will I show up the same way on every platform?

No. Because ChatGPT and Perplexity retrieve differently, visibility on one doesn't guarantee visibility on the other. Worth checking both, not just one.