How to get cited by ChatGPT, Perplexity and Google AI Overviews

A buyer used to search, scan ten links, and click two. Now they ask a question and read one answer with four or five citations under it. If you are not one of those citations, you are not in the consideration set — and unlike a page-two ranking, there is no partial credit.

The uncomfortable part is that the two outcomes are decided differently. Ranking is a judgement about a page relative to a query. Citation is a judgement about a passage: can this specific span of text be lifted out, stand on its own, and be attributed safely? Plenty of pages that rank well are terrible citation candidates, because everything useful in them is spread across six paragraphs of build-up.

What actually changes

Answer engines do roughly three things: retrieve a set of candidate documents, extract passages that address the question, then compose an answer and attribute the passages they leaned on. You cannot influence the composition step. You can influence the other two, and they reward different properties than a ranking algorithm does.

• Retrieval still looks a lot like search. Crawlable, server-rendered HTML, topical relevance, a domain the engine has reason to trust.
• Extraction is the new part. It favours text where the claim and its context sit in the same place, close to a heading that names the question.
• Attribution favours claims that are checkable — specific numbers, named entities, dates — over claims that are merely confident.

Write passages that survive being lifted

The single most useful edit you can make to an existing page: find every question it implicitly answers, put that question in a heading, and make the first two sentences underneath a complete answer that needs no prior context. Not a teaser. The actual answer.

This is why FAQ blocks punch so far above their weight, and it is not because of the schema markup. It is because the format forces a self-contained question-and-answer pair — exactly the shape an extraction step is looking for. The markup helps a parser find it; the structure is what makes it quotable.

The same logic explains what gets ignored. A sentence like “as we discussed above, this approach is generally preferable” cannot be cited, because lifted out of the page it asserts nothing. Anaphora is the enemy: “it”, “this”, “the former” all bind to context that will not travel with the quote.

Corroboration beats assertion

You can declare what your company is on your own site — and you should, in your structured data — but a self-claim carries little weight on its own. Engines resolve entities by looking for agreement across independent sources. If your site is the only place asserting that your brand is a company in category X, that assertion is fragile.

This is the part teams skip because it is not a content task. A company page that exists in more than one place, consistent naming across your profiles, and third-party references that describe you the same way you describe yourself do more for citation than another thousand words on your own domain.

Measure it, or you are guessing

Answer engines are non-deterministic. Ask the same question twice and you can get different citations, so a single check tells you almost nothing about whether an edit worked. What you need is the same set of buyer questions, asked repeatedly, with the cited domains recorded each time — then a change in citation share means something.

Two metrics are worth the tracking effort. Citation share is how often you appear across your question set; it tells you whether you are in the answer at all. Competitive displacement is which domains appear when you do not; it tells you what to write next, because those pages are a live list of what the engine currently considers a better source than you.

That second one is the useful one, and it is the reason a spreadsheet stops scaling quickly. Knowing you are absent is mildly interesting. Knowing exactly which competitor page took the citation, on which question, is an editorial brief.

What to do this week

1. Write down twenty questions your buyers actually ask. Not keywords — full natural-language questions, the kind someone types into ChatGPT when they are comparing options.
2. Ask them, and record who gets cited. This is tedious and it is also the only way to know your starting point. Keep the raw answers; you will want to compare wording later.
3. Pick the five questions where a rival is cited and you are not. Those are your highest-value pages, already validated by the engine itself.
4. Restructure before you write. On each target page, convert the implicit questions into headings and make the first two sentences under each one independently quotable. Often this alone changes the outcome.
5. Fix your entity footprint. Consistent naming, organisation structured data, and at least a few independent sources describing you the way you describe yourself.

None of this is exotic, and none of it is a trick. It is closer to technical writing than to marketing: say the thing plainly, near a heading that names the question, in a way that stays true when someone quotes only that part.

If you want the starting point without the manual pass, our free AI visibility audit runs a question set against the major engines for your domain and shows you which competitors are being cited instead of you. For how the scoring and tracking work under the hood, the platform guide goes through the pipeline stage by stage.