How to get cited by AI (and show up in AI Overviews)

When ChatGPT, Gemini, or Google’s AI Overviews answer a question, they increasingly name their sources: little citation links, footnotes, “according to” mentions. Being one of those sources is the new version of ranking, and an entire cottage industry has already formed around promising it.

Most of what that industry says is guesswork dressed as strategy. Here is what’s actually known, where it’s known from, and the four moves that follow from evidence rather than vibes.

First, the honest map of how AI answers get built

There is no secret AI index you can submit your site to. Each engine builds answers from sources you can name:

Google’s AI Overviews and AI Mode are built on Google’s regular search index. Google’s own guidance, in its AI features documentation, says it plainly: there is no special optimization for AI Overviews beyond normal SEO. Google describes the links in Overviews as drawn from its regular ranking systems, for the question and the related questions an answer touches.

ChatGPT, when it searches the live web, draws on outside search providers, Bing being a known one, plus its own crawler. When it answers without searching, it draws on what it learned in training, which means it recommends brands it saw mentioned, repeatedly and credibly, across the web it read.

Gemini pulls from Google’s index. Perplexity runs its own crawler and, from what its citations show, leans toward pages that already rank.

Notice what this means: every AI answer engine that searches the live web sits on top of classic search. They didn’t replace the index, they built on it. Which is why the first move is the least glamorous one.

Move one: rank, because citation follows ranking

Most pages cited in AI Overviews already rank somewhere meaningful for the query or a closely related one. Overlap analyses from SEO data companies, Ahrefs and Semrush among them, disagree on how strong the effect is, with top-ten overlap estimates ranging from under half of citations to a large majority depending on the dataset and query mix. Ranking is not a guarantee of citation, but it is the strongest pattern every one of those analyses shows. All the fundamentals that earn rankings, real pages that answer real searches, clean technical health, trust signals, are also the price of admission to AI answers.

There’s a quiet bonus move inside this one: Bing matters now. Bing is a known provider behind ChatGPT’s live search, and almost nobody optimizes for Bing, because for years it barely mattered. Registering in Bing Webmaster Tools and getting indexed there is an afternoon of one-time work most competitors have never bothered with; our ChatGPT guide walks through the setup.

Move two: write answers a machine can lift

AI engines don’t cite pages, they cite passages. When an engine assembles an answer, it pulls the paragraph that answers the question cleanly, then links the page it came from.

This is the one area with real experimental evidence. A 2023 study from researchers at Princeton and collaborators, presented at KDD 2024 and the paper that coined “generative engine optimization,” tested what makes content more likely to be picked up in AI-generated answers. The methods that worked were adding statistics, quotations, and cited sources, with visibility lifts of up to roughly 40% in their best cases, and effectiveness varying by topic and query type.

In practice, that means structuring pages so the answer comes first:

• Ask the question as a heading, then answer it in the first two sentences, plainly. Elaborate after.
• Put numbers and specifics in liftable sentences, and name the source next to the number. A sentence carrying a concrete, sourced figure is a citation magnet. A paragraph of mood is not.
• Use real FAQ sections for real questions, marked up properly so machines can parse them.
• One page per question. An engine looking for the answer to X cites the page about X, not the page about X, Y, and Z.

If you’ve read our guide on getting ChatGPT to recommend your business, this is the same logic extended from your brand to your content: make the right answer easy to find and easy to quote.

Move three: be mentioned where the engines read

Here is the part that feels least like traditional SEO. ChatGPT’s training-data answers, the ones it gives without searching, recommend businesses it saw mentioned across the web: review platforms, directories, industry lists, forum threads, local press. Those mentions work even without links. The model isn’t counting votes the way Google’s link system does, it’s absorbing associations: this name keeps appearing next to this service and this place, credibly.

So the work is presence in the sources engines actually consume:

• The directories and review platforms that matter for your industry, complete and consistent.
• Real reviews, steadily. Review text is exactly the kind of source AI answers lean on for “best X in Y” questions.
• The lists and roundups your customers already find. If “best [your service] in [your city]” articles exist and you’re absent, you’re absent from the raw material of the answer.
• Consistent facts everywhere: same name, same locations, same claims. Contradictions read as noise, and engines skip noisy entities.

Move four: measure it, because this is a channel now

You’d never run SEO without checking rankings, but most businesses have never once asked the engines about themselves. The baseline takes an evening: ask ChatGPT, Gemini, and Perplexity the questions your customers would ask. Who do they recommend? Do they mention you? What do they say, and what sources do they cite when they say it?

Do it monthly and you have a scoreboard. The engines’ answers shift as your rankings, mentions, and content change, and watching the shift tells you what’s working, the same way Search Console does for classic search.

One technical check while you’re at it: make sure your site isn’t blocking the AI crawlers in its robots.txt file. For ChatGPT alone there are two different bots with two different jobs: OAI-SearchBot decides whether you can appear in its live search results, and GPTBot decides whether the model learns about you at training time. Sites accidentally block one or both, usually a leftover from a plugin or a template, and then wonder why they don’t exist in ChatGPT’s world. Checking takes a minute: open yourdomain.com/robots.txt and look for OAI-SearchBot, GPTBot, or a blanket block.

Three questions everyone asks about this

Does schema markup help you get cited by AI? Somewhat, and indirectly. Structured data helps machines parse what your pages say and helps you win the rich results that feed Google’s systems. It is worth doing properly. It is not a magic ticket into AI answers, and nobody serious claims otherwise.

Do llms.txt files matter? An llms.txt file is a plain-text site guide some sites now publish for AI crawlers. The honest status: it’s an emerging convention, cheap to add, unproven in effect. We publish one ourselves because the cost is nearly zero, and we’d describe the benefit the same way: plausible, unmeasured, cheap. File it under “why not” rather than “strategy.”

How long does this take? The mention layer moves slowest, because models retrain on the web over months. The ranking layer moves at normal SEO speed. The quotable-content layer can show up in live-search answers within weeks of a page ranking. Which is one more reason the measurement habit matters: without a monthly baseline, you can’t see any of it move.

What to ignore

Anyone guaranteeing placement in ChatGPT is selling something that cannot be guaranteed. Anyone offering a secret AI submission service is describing a thing that does not exist. And any strategy that skips the boring foundation, rankings, quotable pages, real mentions, is a roof with no house under it.

The honest summary in one line: be ranked, be quotable, be mentioned, and measure. Four moves, no mythology.

Seeing what the AI engines currently say about your business is exactly what our free AI visibility audit covers: the questions your customers ask, the answers the engines give, and where you stand in them.