How to Get Your Business Cited by ChatGPT

If you want to get cited by ChatGPT, it helps to understand that this is a retrieval problem before it is a content problem. ChatGPT does not rank the web the way a search results page does. When it answers a question that needs current information, it fetches a small set of documents, reads them, writes an answer in its own words, and attaches links to the sources it actually leaned on. That is normally a short list rather than a page of results. Your job is to be in that fetch, and then to be the easiest thing in it to quote.

How does ChatGPT decide which sources to cite?

ChatGPT picks sources in two stages: a retrieval stage that gathers candidate pages for the rewritten query, and a synthesis stage where the model quotes whichever passages most directly answer the user. A page can win the first stage and still lose the second, which is why sites with good rankings sometimes get no citations at all.

The retrieval stage behaves like a search engine — it depends on your page being indexed, crawlable and topically matched to the query. The synthesis stage behaves like an editor with a deadline. It scans what came back and lifts the sentence that stands on its own. Vague, throat-clearing prose is skipped even when it is technically the better article, because there is nothing in it that survives being pulled out of context.

OpenAI has said ChatGPT's search experience draws on its own crawler-built index alongside third-party search providers, so classic technical health still underpins everything. If Google is struggling to index you, an AI retrieval layer will struggle too — the diagnosis in our guide on why Google is not indexing your pages is the right starting point before any AEO work.

Does ChatGPT crawl my website directly?

Yes, and it does so under more than one identity. OpenAI documents separate user agents for separate jobs, and your robots.txt file controls each of them independently:

• GPTBot — the crawler associated with collecting data that may be used for model training.
• OAI-SearchBot — the crawler that builds the index ChatGPT's search feature retrieves from.
• ChatGPT-User — the fetch that happens live, in the moment, when a user or a tool asks ChatGPT to open a specific page.

This distinction catches a lot of businesses out. A developer blocks GPTBot because leadership does not want the content used for training, and everyone assumes the site is now invisible to ChatGPT. It is not — but if someone blocks OAI-SearchBot as part of the same tidy-up, the site really does drop out of the retrievable set. Check your robots.txt line by line and decide on each agent deliberately rather than blocking anything that sounds like AI.

The other silent killer is rendering. If your main content only exists after JavaScript executes, you are relying on every one of these crawlers rendering it. Server-side rendering or static generation removes the gamble entirely, which is one of the reasons we push it hard on every build we take on.

What makes a page quotable rather than merely readable?

A quotable page answers the question in the first sentence after the heading, in about 40 to 60 words, with no dependency on the paragraph before it. That is the whole trick, and almost nobody does it, because it is the opposite of how people are taught to write.

Human readers tolerate a build-up. A model assembling an answer does not have one. It is looking for a passage where the subject, the claim and the qualifier all sit inside the same block of text. Compare these two openings to a section headed "How long does an SEO audit take?":

• Weak: "This is one of the most common questions we get, and the honest answer is that it depends on a number of factors, which we will unpack below."
• Liftable: "A technical SEO audit on a site under 500 pages typically takes one to two weeks: a few days of crawling and log analysis, then the rest writing up prioritised fixes. Larger sites take longer because sampling stops being reliable."

The second one can be pasted into an answer with a citation attached and it still makes sense. Write every section that way and you have at least given a model something it can use. The same principle drives FAQ schema that actually gets quoted — structure is only useful when the sentence inside it is self-contained.

The page-level checklist

1. Phrase most section headings as the question a person would actually type.
2. Answer in the very next sentence. Elaborate afterwards, not before.
3. Put specifics in the answer — a number, a name, a step, a constraint. Generic sentences never get chosen.
4. Keep one idea per paragraph so a passage can be extracted cleanly.
5. Date the page visibly and update it when the facts change. Stale pages get deprioritised on anything time-sensitive.
6. State who wrote it and why they are credible, on the page, not just in schema.
7. Use tables for comparisons. Structured rows survive summarisation far better than prose.

Do I need schema markup to get cited by ChatGPT?

Schema is helpful but not the deciding factor. It makes your page easier to parse and disambiguates what your business is, which supports retrieval — but no amount of JSON-LD rescues a page whose answers are buried. Treat markup as reinforcement for content that is already clear, and see schema markup for AI search visibility for the types worth implementing.

What does matter more than people expect is entity clarity. Your business name, what you do, where you operate and how to reach you should be stated consistently everywhere — your own site, your Google Business Profile, LinkedIn, industry directories. Models resolve entities from repetition across independent sources. Inconsistent naming splits your identity into two half-formed entities, and neither becomes citable.

What about being mentioned on sites you do not own?

Off-site mentions often carry as much weight as your own pages, because for commercial queries the retrieval layer commonly pulls listicles, forums, review platforms and industry roundups rather than vendor sites. Try it yourself: ask an assistant for a shortlist of agencies in your city and look at what it cites. Often none of the sources are agency websites.

The practical response is unglamorous: be present, accurately, where those documents get written. Answer questions in communities where your customers actually are. Get included in genuine roundups. Keep your directory listings identical to each other. None of this is a growth hack, and none of it works quickly, but in our experience it is the part of our SEO work that shows up most consistently in AI answers, and it compounds with the entity work described in entity SEO and knowledge graph presence.

How do you tell whether any of it is working?

You measure it by testing prompts on a schedule and by watching referral traffic, because there is no ranking report for AI citations and there probably never will be one. Build a list of 20 to 40 questions a real buyer would ask, run them monthly across ChatGPT, Perplexity and Google's AI mode, and record whether you were cited, mentioned without a link, or absent.

On the analytics side, referrals from chatgpt.com and perplexity.ai are visible in your traffic reports and should be segmented out. The volume will look small next to organic search, so judge it on the quality of what arrives rather than the count. Someone who clicks through from an assistant has already had a comparison summarised for them, which tends to make the conversation you have with them a different one. Measure that against your own baseline rather than trusting anyone's benchmark. If you want a second opinion on what to track and what to ignore, that is the kind of thing our AI consulting team does in a single working session.