How to Get Cited by ChatGPT: Seven Things That Actually Move It

You get cited by ChatGPT by being readable to the right crawler, quotable in the first sentence, and corroborated somewhere other than your own website. Everything else is detail. Most brands we audit fail on the first point and the last, and have never checked either.

This is the practical companion to our working definition of generative engine optimisation. If you want the concept first, start there. If you want the checklist, here is the order we work in.

1. Find out which crawlers you are blocking

This is where the single most expensive mistake lives, and it is almost always accidental.

In 2024 a great many sites added AI crawlers to their robots.txt to keep their content out of model training. Reasonable at the time. The problem is that the training crawler and the search crawler are different bots, and blocking one does nothing to the other.

• GPTBot — OpenAI's training crawler. Blocking it keeps your content out of foundation model training. It does not affect whether ChatGPT can cite you.
• OAI-SearchBot — the one that matters. This indexes pages so they can be surfaced in ChatGPT's search features. Block this and you are invisible to ChatGPT search.
• ChatGPT-User — fetches a page when a user asks about it directly or pastes a link.
• OAI-AdsBot — validates the safety of pages submitted as ads on ChatGPT. Nothing to do with organic citation.
• ClaudeBot, Claude-SearchBot, Claude-User — Anthropic's equivalents: training, search quality, and user-initiated fetches.
• PerplexityBot and Perplexity-User — Perplexity's index crawler and its live retrieval agent.
• Googlebot — still the only crawler behind Google's AI Overviews and AI Mode.
• Google-Extended — not a crawler at all. It is an opt-out token controlling whether your content is used for Gemini. It has no effect on AI Overviews.

The practical consequence: a brand that "blocked AI" two years ago may have kept itself out of training, which is a defensible choice, while also cutting off the search crawler that decides whether it can be recommended. Those are two entirely different commercial decisions and they should be made separately. Check your robots.txt before you do anything else on this list.

2. Tell the machines what each page is

Structured data is how a page states its own type, author, date and subject rather than leaving a parser to infer them. An article marked up as an Article, with a headline, a description, a named author and a publication date, is far easier to attribute than an identical page with no markup at all.

Do not overreach here. Marking up things that are not what you claim is a fast route to a manual action, and the schema types worth having are unglamorous: Article, Organization, Breadcrumb, Product, and FAQPage where the questions are real.

3. Put the answer in the first sentence

An assistant lifts a passage, not a page. This is the single biggest change to how content should be written and the hardest one for brand teams to accept, because it inverts the house style most of them have spent years building.

The scene-setting opening — the paragraph about how the industry is changing, before you get to the point — is the enemy. If someone asks what a term means and your definition is in paragraph four, paragraph one gets extracted instead and it says nothing. Lead with the claim. Put the evidence underneath it. You can still write beautifully; you just cannot warm up first.

4. Give every page a summary that stands alone

Most sites treat the meta description as a teaser designed to earn a click. In an answer engine that is exactly wrong: a teaser withholds the information the model is looking for, so the model goes elsewhere.

Write the description as a precis instead — a short, complete statement of what the piece argues, which would make sense read aloud with no surrounding page and your name attached. Assume it will be, because that is what happens.

5. Publish something only you know

Models reward specificity, and the most citable thing any brand can produce is a number nobody else has published. If you are the only source for a fact, you are the only thing to cite.

For most brands this is easier than it sounds. First-party data you already have — conversion patterns, category shifts you have observed, results you can disclose with a client's permission — is original research the moment you write it up. Our analysis of why ingredient-led beauty brands dominate AI citations is the shape of the thing: a specific finding, in a specific category, that nobody else had written down. A restatement of what everyone in the category already says has no reason to be quoted over anyone else's version of it.

6. Get corroborated somewhere that is not your website

An assistant weighs independent sources far more heavily than a brand's own claims about itself, for obvious reasons. If the only place a claim appears is your About page, it carries almost no weight.

What counts: trade press, retailer product copy, review platforms, category round-ups, podcast and interview transcripts, and specialist forums. What matters is consistency — the same claim, in compatible language, in several independent places. This is slower than every technical fix on this list and it is the one that moves the outcome most. It is also why public relations and search have stopped being separate disciplines.

7. Measure by sampling, and be honest that it is sampling

There is no rank tracker for answers. The working method is to fix a set of category prompts, run them across several assistants on a schedule, and record which brands are named and which sources are cited. Then watch referral traffic from assistant domains in analytics as a secondary signal.

It is repeated observation rather than measurement, and the numbers move for reasons you did not cause. Report it that way. A supplier presenting AI visibility as a precise metric is telling you something the tooling cannot support.

What does not work

Hidden instructions to the model in white text. Keyword-stuffed pages with an AI-flavoured vocabulary. Publishing a file of instructions to crawlers and treating it as a strategy — the evidence for these being read at all is thin, and the one that ships as a genuine standard is worth having only because it costs an hour, not because it will carry you.

And volume. Twenty thin posts about a topic will not outrank one piece that actually answers the question, because the competition is no longer for shelf space in a list. It is for a sentence.

Studio UNA runs generative engine optimisation for beauty, fashion and aesthetics brands. The audit starts by running this list over your site and showing you what the assistants say about you today.