How to get cited by ChatGPT

OpenAI publishes exactly one thing about this clearly, and it is the part most guides skip. Everything else in circulation is inference, some of it good. This page separates the two.

The short answer

Getting cited by ChatGPT has one documented rule, plus a set of well-evidenced habits on top. Mixing those two is why so much advice here is confidently wrong.

Start with what OpenAI actually states.

• One crawler decides whether you can appear at all. OpenAI documents that “OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features.” Block it and you are out of ChatGPT’s search answers.
• A different crawler trains the models. GPTBot “is used to crawl content that may be used in training our generative AI foundation models.” Blocking it affects training, not search visibility.
• A third fetches pages live for users. ChatGPT-User visits a page when somebody asks a question, and OpenAI notes it is not used for automatic crawling, and that robots.txt rules may not apply to it.
• A fourth exists now for ads. OAI-AdsBot “is used to validate the safety of web pages submitted as ads on ChatGPT”, which has no bearing on organic citation.

The practical consequence. Many sites block GPTBot to stay out of training, which is a fair choice. Then they block OAI-SearchBot alongside it by accident. That one line removes you from ChatGPT search answers entirely. It is the first thing we check.

Pepper is an agentic organic growth engine, not an SEO agency. This page comes out of what we run: organic for more than 250 enterprises across eight years, more than 10 million tracked prompts, and the questions buyers put to us in client reviews. Customers log in and run the platform themselves, with a Pepper growth team attached. Book a growth audit and we will check your crawler configuration before we talk about content.

Key takeaways

Key takeaways, split into what is documented and what is inferred.

• Documented: four OpenAI agents, with different jobs. OAI-SearchBot for search, GPTBot for training, ChatGPT-User for live fetches, OAI-AdsBot for ad safety.
• Documented: blocking search and blocking training are separate decisions. You can refuse training and still appear in search answers.
• Inferred: what makes a page quotable. The best evidence is academic rather than from OpenAI, and it is good evidence.
• Evidenced: ChatGPT cites fewer sources, with more weight each. Which raises the bar compared with engines that cite broadly.
• Not documented: any ranking factor. OpenAI publishes nothing resembling a ranking guide, and anyone presenting one is guessing.
• Measurable: your own citation rate. Twenty prompts and an afternoon gives you a baseline nobody can sell you.

What is a ChatGPT citation?

A citation means ChatGPT used a page from your domain as a source and showed the link. A mention is different: your brand name in the text, with none of your pages behind it.

This matters more on ChatGPT than elsewhere, for two reasons. One is documented, one is evidenced.

• Documented. Only OAI-SearchBot decides whether your site can surface in ChatGPT’s search features. Mentions can still happen without it, because the model picked up your brand elsewhere.
• Evidenced. One controlled study ran the same prompts across ChatGPT, Google AI Overviews and Perplexity. ChatGPT cited fewer sources, and leaned on each one more. The other two cited widely and leaned on each one less.

Where it falls short. Neither of those tells you how ChatGPT chooses between two eligible pages. Nobody outside OpenAI knows. So we are working from patterns we can see, not from a published rule. We separate the two outcomes properly in what is brand visibility in AI.

How to get cited by ChatGPT, step 1: fix access

This is free, it takes an afternoon, and it is the only step with documented consequences.

Check your robots.txt for all four agents by name. The one that matters for citations is OAI-SearchBot. OpenAI is explicit here: block it and your site will not appear in ChatGPT search answers, though navigational links may still show.

Three configurations are worth knowing.

• Allow OAI-SearchBot, block GPTBot. The common position for publishers who want visibility without contributing to training. Entirely supported.
• Block both. A legitimate choice with a known cost. If your business does not sell to people who research in ChatGPT, you do not need to open it, and we would say so.
• Block GPTBot with a wildcard. The accident. A broad rule intended to stop training can catch the search agent too, and the symptom is a site that is invisible in ChatGPT for no apparent reason.

Then check what robots.txt cannot tell you. Does the page return a real status code? Is the text in the HTML, rather than built by scripts after load? And is anything at the CDN layer blocking the agent? The full sweep is in how to audit your site for AI search.

One more documented detail. ChatGPT-User, the agent that fetches a page when a user asks about it, is a separate case: OpenAI states it is not used for automatic crawling and that robots.txt rules may not apply to it. So a page can still be fetched for a user where crawling is banned. Your robots.txt is not a privacy control.

Step 2: Build pages worth lifting from

Here the docs run out and the research begins. We will say which is which, rather than blur them.

The best evidence we have is one controlled study. It covered 602 prompts, 21,143 search-layer citations and 18,151 fetched pages across ChatGPT, Google AI Overviews and Perplexity. It found that the pages with real influence run longer, carry more structure, sit closer to the question, and hold more usable evidence: definitions, numbers, comparisons and steps.

That gives four practical moves. None of them is a trick.

• Answer the question in the first two sentences under each heading. An engine lifting a passage takes whatever sits at the top of it.
• Put extractable evidence on the page. A number with a source, a definition, a comparison table, a numbered procedure. A paragraph of positioning gives an engine nothing to use. The planning side of that sits in how to build an SEO content plan.
• Structure for passages, not for pages. Each section should stand alone, because engines retrieve passages rather than whole documents.
• Write the comparison you are avoiding. Engines reach most often for pricing, alternatives and “X vs Y” pages. Those hold the clean comparison the question asks for.

The honest caveat. That study measured what cited pages look like. It did not establish what causes a citation. The link is strong and the reason is plausible. It is still a link, not a cause, and we would rather say so than sell it as a formula. The page-level technique is in how to structure content for answer engines.

Step 3: Be present where ChatGPT looks

Your own pages are necessary and not sufficient. Here the evidence gets thinner.

One thing is clear from the research: AI answers lean heavily on sources you do not own. Independent research across 11,000 real queries and five systems found identical questions produce “structurally different information realities across systems”, with some source types systematically over-represented. Which sources those are varies by category. So the answer has to come from your own measurement rather than a list in an article.

So find out, rather than assume.

• Run twenty category prompts and list every domain cited. Not brand names. Domains, plus the third-party sites next to them.
• Sort by frequency. The top five or six are the real authority layer in your category. They are rarely the ones a PR plan would have picked.
• Decide what is reachable. Some are review sites you can join. Others are publications you can pitch. A few are communities where showing up has to be genuine, or it backfires.

We go through the full method in how to find which sources AI cites in your category.

What we will not tell you. A percentage of ChatGPT citations that come from earned media. Figures circulate, they vary enormously between studies, and the companies publishing most of them sell the service the figure implies you need.

Step 4: Measure it, because nobody will give you the number

There is no Search Console for ChatGPT. The nearest thing is Microsoft’s report, and it does not cover ChatGPT.

Bing Webmaster Tools added an AI Performance report in public preview on 10 February 2026. It shows total citations, citation counts per page, and the queries behind them. Microsoft states the data covers “Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations”. It does not name ChatGPT, so we would not present it as a ChatGPT measurement.

So you have to run it yourself.

• Fix twenty prompts. Spread them across awareness, comparison and purchase questions in your category.
• Run them in ChatGPT with search on, in a fresh session. Record two things: did it name your brand, and did one of your pages show up in the sources.
• Repeat monthly with the same prompts. Watch the citation column rather than the mention column, because that is the one your content work moves.

Two readings give you a rate of change for your own category. That beats any published benchmark, and the comparison discipline is in how to benchmark AI visibility against competitors. The wider routine is in how to track brand mentions in AI search.

What the evidence does not support

Four claims get repeated widely. We would not repeat any of them.

• That there is a ChatGPT ranking algorithm you can optimise against. OpenAI publishes crawler docs and no ranking guide at all. So anyone who describes the algorithm is guessing from samples.
• That Bing rankings determine ChatGPT citations. ChatGPT drew on Bing’s index at points, the relationship has shifted over time, and OpenAI documents no current dependency. Treat the Bing-first argument as a hypothesis worth testing on your own data rather than a mechanism.
• That llms.txt helps. Google has publicly mythbusted it, no major engine documents support for it, and we have seen no evidence that it changes citation behaviour.
• That volume drives citations. The research points the other way. Cited pages run longer, carry more structure and hold more evidence. They are not more numerous.

Why we are this cautious. Twice we have tried to read agency-published studies behind widely quoted citation-share figures in this field, and twice the study would not load. A number you cannot trace is a rumour with a decimal point, and we would rather leave a gap than fill it.

What nobody should promise you

• A guaranteed ChatGPT citation. Nobody outside OpenAI controls the output, and Google’s own guidance warns against providers who guarantee rankings for the same structural reason.
• A ranking factor list. None is published. Lists in circulation are inferred from samples, which is fine as a hypothesis and dishonest as a fact.
• A fixed timeline. Citation behaviour shifts as models and retrieval shift. Anyone quoting a precise number of weeks is guessing.
• That blocking GPTBot costs you visibility. It does not. Blocking OAI-SearchBot does, and mixing up the two is the most costly error in this area.

Frequently asked questions

How do you get cited by ChatGPT?
Allow OAI-SearchBot, which OpenAI documents as the agent that surfaces websites in ChatGPT’s search features, then publish pages with extractable evidence: direct answers, definitions, numbers with sources, comparisons and procedures.

Does blocking GPTBot stop me appearing in ChatGPT?
No. OpenAI documents GPTBot as the crawler used for training foundation models, and OAI-SearchBot as the one used to surface sites in ChatGPT’s search features. You can block training and still appear in search answers.

The crawlers

What is OAI-SearchBot?
OpenAI’s search crawler. Its documentation states it “is used to surface websites in search results in ChatGPT’s search features”, and that blocking it prevents your site appearing in those answers, though navigational links may still appear.

What is ChatGPT-User?
The agent that fetches a page when a user asks ChatGPT a question about it. OpenAI states it is not used for automatic crawling or to determine search results, and that robots.txt rules may not apply to it.

Is there a new OpenAI crawler for ads?
Yes. OAI-AdsBot is documented as validating the safety of web pages submitted as ads on ChatGPT. It has no bearing on organic citation, and blocking it does not affect search visibility.

Content and measurement

What kind of page gets cited most?
Research across 21,143 search-layer citations found high-influence pages are longer, more structured, more semantically aligned and richer in extractable evidence such as definitions, numerical facts, comparisons and procedural steps. That is a correlation rather than a published rule.

Can I see ChatGPT citations in any reporting tool?
Not from a first-party source. Bing Webmaster Tools’ AI Performance report covers Copilot, Bing’s AI summaries and select partner integrations, and does not name ChatGPT. Measuring your own prompt set is the reliable method.

How is ChatGPT different from Perplexity for citations?
A controlled study found ChatGPT cites fewer sources with higher influence each, while Perplexity and Google cite more sources with less weight per source. Earning a ChatGPT citation is harder and worth more when it happens.

Sources

Every claim traces to a primary source, read directly on the date shown.

OpenAI documentation, read 8 October 2026

• OpenAI, Bots and crawlers. Documents OAI-SearchBot: “OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features”, with blocking preventing appearance in those answers. Documents GPTBot: “GPTBot is used to make our generative AI foundation models more useful and safe. It is used to crawl content that may be used in training our generative AI foundation models.” Documents ChatGPT-User as visiting pages in response to user questions, not used for automatic crawling, with robots.txt rules that may not apply. Documents OAI-AdsBot: “OAI-AdsBot is used to validate the safety of web pages submitted as ads on ChatGPT.”

Microsoft documentation

• Bing Webmaster Blog, Introducing AI Performance in Bing Webmaster Tools, public preview, 10 February 2026. States the data covers “Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations”. Metrics include total citations, page-level citation activity and grounding queries. ChatGPT is not named.

Independent research

• Zhang Kai, He Xinyue and Yao Jingang, From Citation Selection to Citation Absorption, arXiv, 28 April 2026. 602 controlled prompts, 21,143 search-layer citations, 18,151 fetched pages across ChatGPT, Google AI Overviews and Perplexity. Finds ChatGPT cites fewer sources with higher influence each, and that high-influence pages are longer, more structured and richer in extractable evidence. Academic preprint, not peer-reviewed at the time of writing.
• Huang, Goyal, Saha and Chandrasekharan, Answer Bubbles: Information Exposure in AI-Mediated Search, arXiv, 17 March 2026, revised 28 August 2026. 11,000 real search queries across five systems, finding identical queries produce “structurally different information realities across systems”. Also an academic preprint.

On what this page is not

This is not a ranking-factor guide, because no such thing is published for ChatGPT. We have separated what OpenAI documents from what independent research evidences from what we infer, and labelled each. Where we could not verify a widely circulated figure, including citation-share percentages attributed to earned media, we have left it out rather than repeat it. If OpenAI publishes retrieval guidance, this page will need rewriting, and we would rather say that than imply the current position is permanent.