The question every founder asks me first is some version of “how does the AI decide who to mention?” They want the one rule. The headline factor. The thing they can go fix on Monday.
There isn’t one rule. There is not even one AI. The engines barely agree with each other, and the way they pick sources looks almost nothing like the SEO game everyone spent fifteen years learning. So before the tactics, here is what is actually happening when an engine decides whether your name shows up.
First it decides whether to look at all
The single biggest factor is one most people never think about: whether the engine even retrieves live sources for your question.
ChatGPT answers from memory by default. It was trained on a snapshot of the web, and for a lot of questions it just answers from that training, with no live search and no citation of anyone. It only goes and fetches sources when the question clearly needs something recent or specific, or when the user turns search on.
Perplexity is the opposite. It was built as a citation-first engine. It retrieves live sources for almost everything and shows them every time.
This matters more than any on-page tactic. If the engine answers your category question from memory, the only way to be “cited” is to have been prominent enough, often enough, in the training data that the model recalls you unprompted. If it retrieves live, you are in a fast, winnable race for the sources it pulls right now. Same brand, same page, completely different game, decided before any ranking factor comes into play.
When it does retrieve, it is not counting backlinks
Here is where people get it most wrong. They assume AI citation is just SEO with extra steps, so they reach for domain authority and backlinks. Those help at the edges, but they are not what the engine is optimizing for.
When an engine assembles an answer, it is looking for text it can lift cleanly and trust. In practice that means a few things, roughly in order of how much they move the result:
- Extractability. Can the engine pull a clean, self-contained answer off your page without rewriting it? A sentence that directly answers the question, stated plainly, beats three paragraphs that circle it. The engine is quoting, and it quotes what is quotable.
- Specific, checkable claims. A named number, a dated result, a concrete comparison. Engines lean toward text that asserts something verifiable, because a specific claim is safer to repeat than a vague one. “Improves retention” is skippable. “Cut churn from 4.1 to 2.8 percent over two quarters” is citable.
- Corroboration. If several independent sources say the same thing, the engine trusts it more and is likelier to surface whoever said it clearly. Being the only page making a claim is weaker than being the clearest page making a claim others also support.
- Topical fit and freshness. A page that is tightly about the exact question, and recently updated, reads as more relevant than a broad page that mentions the topic in passing.
Notice what is not at the top of that list: how many backlinks you have, how much traffic you get, how big your domain is. Those were the currency of Google’s blue links. AI engines are optimizing for answer-readiness, not authority-as-popularity. That is genuinely good news if you are small and clear, and bad news if you are big and vague.
The part the explainers skip: the engines disagree
Most write-ups stop at “here are the factors,” as if there were one set of them. The thing you actually have to internalize is that the engines disagree with each other, a lot.
Independent studies that compare citations for the same query keep landing in the same place: the overlap between what ChatGPT cites and what Perplexity cites is small, often only around one in ten sources. The rest are unique to one engine. And they lean on different kinds of sources by temperament, ChatGPT toward established, encyclopedic references, Perplexity toward live and community sources.
I see this on real brands constantly. I ran the same buying question about issue trackers through two engines recently. One named a strong, well-known product first. The other filed that exact same product under “lightweight” and led with a competitor. Same product, same week, two engines, opposite verdicts. Neither was wrong from its own logic. They were reading different sources.
So “getting cited” is not one achievement you unlock. It is five separate races, on five engines that weight things differently and trust different corners of the web. A brand can own ChatGPT for its category and be completely absent from Perplexity, and the fix for each is different.
What this means you actually do
Strip it back and the implications are simple, even if the work is not:
- Be the clearest extractable answer to the specific question. Lead the page with the plain statement the engine can lift. This is the highest-leverage thing on the list and the most ignored.
- Put a checkable fact in reach. Your own number, your own result, stated honestly. Specific beats authoritative.
- Earn presence in the sources each engine trusts. For the encyclopedic-leaning engines, that means established references and your own well-structured pages. For the community-leaning ones, it means the forums and reviews real users write. You cannot fake either, which is the point. The community side has its own playbook, since the engines lean on Reddit more than anywhere else: see why AI search keeps citing Reddit, and what you can actually do about it.
- Measure per engine, not in aggregate. Because the engines disagree, an average hides the truth. You have to check each one separately, which is the whole discipline of measuring AI search visibility.
The mechanics of writing the page itself, the extractable answer and the checkable claim, are laid out in how to write content AI engines will cite. The question of which buyer queries to even target comes first, in the buyer query map.
Why this matters
If you take one thing from this, let it be that AI citation is not Google with a new coat of paint. The currency changed. It moved from authority-as-popularity, which big incumbents accumulate over years, to clarity and verifiable specificity, which anyone can produce this quarter. That is why a two-year-old company with one honest benchmark can get cited over a competitor with a decade of backlinks.
But it only pays off if you stop looking for the one rule and start playing the actual game: five engines, different logic each, all rewarding the same underlying thing, the clearest and most specific answer to the question your buyer is really asking.
If you want to see exactly how the engines currently treat your category, that is what a paid audit measures, and the full method is on the methodology page.