ChatGPT cites web pages by searching for them, fetching a handful, and generating an answer grounded in what it found. Which means being cited is mostly a question of being findable and readable, with one platform-specific detail that quietly disqualifies a lot of sites.
That detail is your robots.txt, and it is worth checking before anything else.
The crawler distinction that matters
OpenAI runs more than one user agent, and they do different jobs. One gathers web data that may be used to improve models. Another fetches pages in real time so ChatGPT can answer a question a user just asked, and it is this one that produces the citations beside an answer.
The consequence is easy to miss. A site that blocked “AI crawlers” wholesale, often during the wave of concern about training data, may have also blocked the retrieval fetcher, and therefore removed itself from every answer it might have been cited in. Those are two different decisions with two different trade-offs, and they deserve to be made separately.
The training question is about content licensing: do you want your material contributing to future models? Reasonable people answer differently, and publishers with valuable archives often say no.
The retrieval question is about distribution: do you want to be quoted and linked when someone asks a question you can answer? For almost any business selling a service, the answer is yes, because that citation is a recommendation in front of someone with intent.
What decides whether you are a candidate
Once retrieval can reach you, selection looks a lot like search. ChatGPT searches, gets results, and works from them. So everything that determines whether you surface in search determines whether you are in the candidate pool at all: relevance to the query, technical health, and enough authority to be worth drawing on. The same argument we made about how AI Overviews choose their sources applies here with different plumbing.
Three things then influence whether you get used rather than merely fetched.
A direct answer, early. Generated answers quote passages. State the answer plainly near the top, under a heading phrased the way a person would ask, and you have supplied an obvious candidate.
Specifics over generalities. When several sources say roughly the same thing, the one with concrete numbers, named examples or a clear procedure is the more useful citation. Vagueness is not just weak writing here; it is a reason to be passed over.
Unambiguous identity. Structured data, a real author, consistent business details and an About page that states plainly what you do all reduce the work of establishing who you are. Google’s helpful content guidance frames this as demonstrable experience, and it travels across systems.
What nobody can sell you
There is no submission form, no paid inclusion in organic answers, no verified ranking control and no dashboard reporting your position. If a vendor offers guaranteed placement in ChatGPT answers, they are describing a product OpenAI does not offer.
The same scepticism applies to precise “AI visibility scores.” Answers vary between users, sessions and model versions, so any single number is a sample presented as a measurement.
Measuring it honestly
Two signals, neither complete.
Citation clicks arrive as referral traffic, so you can see some of them in analytics, with the usual caveats about attribution across apps and devices. Treat that as a floor rather than a total.
The prompt audit is the real instrument. List twenty questions a prospect would ask before hiring in your category. Run them in ChatGPT monthly. Record whether you are named, whether you are cited with a link, and which competitors appear. The trend over months tells you something; any single run does not, because the same question can produce different sources on different days.
Where to start this week
Check robots.txt for a blanket AI block and separate the two decisions. Take your three most commercially important pages and move the direct answer to the top. Then run the prompt audit once, so that next month’s version has something to compare against.
That is the whole of ChatGPT-specific work. Everything beyond it is ordinary search quality doing the heavy lifting, which is what we do.
