How to Get Cited by ChatGPT, and What Nobody Can Promise You

A customer mentions, almost in passing, that they found their shortlist by asking ChatGPT. And you realise there is now a layer between you and your buyers that you cannot see into. With Google you could at least look up where you sat. Here there is no position, no report, and nothing to appeal to. So you search for how to fix it, and you find a video about a leak, a study of 1.4 million prompts, an audit of two million sessions and an advert for a tracking tool, all confidently describing a mechanism none of them has been shown. That is a genuinely difficult position to make a decision from. The good news is that a real, ordered list of things you control does exist, and it is shorter than the noise suggests. Start there, and treat everything sold beyond it with the suspicion it has earned.

Make yourself eligible first, because nothing else counts until you are

Every other step on this page is wasted if an assistant cannot read your site, so this is the one to do this week.

OpenAI's publisher documentation is unusually plain about the entry condition: any public website can appear in ChatGPT search, and for your content to appear in summaries and snippets, you must not be blocking `OAI-SearchBot`.

That is the whole gate. Not a subscription, not a file, not a schema type. A public page that a specific crawler is allowed to fetch.

Check three things and you have covered it. Your robots file does not disallow `OAI-SearchBot`. Your firewall, security plugin or CDN is not quietly blocking it either, which is where most accidental blocks actually live rather than in robots.txt, and is worth checking as part of ordinary search engine optimization. And the pages you care about return normally and are indexed, which you can confirm with URL Inspection in Search Console.

Do the same for the other assistants your buyers use. OpenAI publishes its full list of agents, and Google's guidance is that its AI features need no additional technical requirements beyond ordinary Search eligibility, which means Googlebot access is the entry condition there.

Allow the right crawler and stop doing the wrong one's job

This is the single correction most likely to change your outcome, and it catches experienced people.

The advice circulating almost everywhere is to allow `GPTBot`. `GPTBot` concerns content being used for future model improvement. It is not the agent that makes your pages available to ChatGPT when it searches the web on someone's behalf. That is `OAI-SearchBot`.

The practical consequence is worse than doing nothing. A business allows GPTBot, ticks the item off, tells itself it is now visible to ChatGPT, and stops looking. Months later it is still absent and nobody revisits the assumption, because the job was marked done.

The same confusion runs one platform over. `Google-Extended` is frequently described as the control for AI Overviews. It is not. Google states that Googlebot governs crawling for Search including its AI features, so blocking Google-Extended does not remove you from AI Overviews, and blocking Googlebot removes you from ordinary search results at the same time. Google has written about these publisher controls directly.

Two crawler names, two decisions, and getting them backwards is the difference between being readable and believing you are.

Give the assistant something worth lifting

Once you are readable, the question becomes whether there is anything on your page an answer would want to quote. This is where most sites lose, and it is a writing problem rather than a technical one.

The most useful evidence here is peer-reviewed rather than promotional. Aggarwal and colleagues published a controlled study at KDD 2024 that tested nine text-editing strategies against a generative engine and measured how visible each made a source. Adding citations, direct quotations and statistics improved the visibility measure in that setting. The preprint is public if you want the method.

Take that as a strong hint and not as a law, because the study measured a controlled setting rather than a small business competing on the live web, and it did not establish that any of those edits causes a citation.

What it points at, though, matches what practitioners keep reporting independently: a direct claim, followed by the reasoning that supports it, in short paragraphs. Answer the question in the first sentence under the heading, then justify it. A page that buries its answer three paragraphs down is hard to lift a sentence out of, and a page that hedges everything gives an assistant nothing quotable at all.

The thing you have that nobody else does is your own figures. What a job actually costs, how long it took, what happened when the weather turned, what you charge and why. Original, checkable specifics are both the hardest thing for a summary to replace and the easiest thing for it to quote.

Get your business facts to agree with each other

Assistants assembling an answer about a local business are pulling from more than your website, and disagreement between sources is a quiet reason to leave you out of one.

Your website, your Google Business Profile, and every directory that has ever listed you should carry the same business name, the same contact route, the same hours and the same service area. Details drift when a business moves, changes hours or rebrands, and old versions survive in places nobody thinks to check.

None of this is exotic and all of it is checkable in an afternoon. It is also the part with the clearest payoff for a local business, because a recommendation engine handling "who should I call near me" has very little else to work with. That is local SEO and Google Business Profile management doing a job they were always doing, for a new audience.

Google's guidance on helpful content is worth reading alongside this, because it names the opposite behaviour explicitly: content produced mainly to attract search visits, and summaries that add nothing, are listed as warning signs.

Find out whether you are cited, without buying a tool to tell you

You can measure this yourself, starting today, and the result will be better than a single score from a dashboard.

Two supporting sources are worth wiring up beside it. Check your analytics for referrals arriving from assistants, which will be small numbers and are worth watching for direction. And Microsoft has begun publishing an AI Performance report in Bing Webmaster Tools, which is currently the strongest first-party citation measurement any operator publishes. Search Console does not produce a site-wide ChatGPT citations report, so nobody is holding one back from you.

One caution about reading your own results. A run that returns nothing tells you about those prompts, on that day, in that region. It is not proof of invisibility, and treating it as proof will send you rebuilding things that were fine.

Know what is folklore before you pay for it

A market has grown up around this question faster than the evidence has, and some of what it sells is confidently made up.

There is no confirmed AI schema type that produces citations. An `llms.txt` file is not a Google visibility requirement, and Google's own guidance on optimising for AI features says to ignore unnecessary AI text files as a Search tactic. "Writing in chunks" is a plausible-sounding habit with no established causal evidence behind it. A tool score is a sample of changing output, not a measurement of your standing.

None of that means the tools are worthless or the people selling them are dishonest. It means the claims run ahead of what anyone has demonstrated, and you should buy accordingly. We keep a separate assessment of the tools people use to track this if you want the landscape before you spend. If your pages are not yet crawlable, your crawler access is wrong, or your business facts disagree with each other, a tracking subscription will simply document your absence more precisely.

Fix the free things first. Buy measurement afterwards, if you still want it.

Frequently Asked Questions About Getting Cited by ChatGPT

Can anyone guarantee my business will be cited by ChatGPT? No, and OpenAI says so itself. Its ChatGPT Search documentation states there is no way to guarantee top placement. Anybody promising a citation is promising something the operator has publicly declined to promise.

Is allowing GPTBot enough? No, and this is the most common mistake. GPTBot relates to future model improvement. `OAI-SearchBot` is the agent that lets your content appear in ChatGPT's summaries and snippets. Allowing only GPTBot leaves you where you started while feeling finished.

Do I need an llms.txt file or special AI schema? No. Google's guidance says to ignore unnecessary AI text files as a Search tactic, and no operator documents a schema type that produces citations. Ordinary structured data that matches your visible content is still worth having on its own merits.

How do I check whether ChatGPT mentions my business? Run twenty real buyer questions signed out in a clean browser, and record the answer, the linked sources, the date, the prompt and your region. Repeat monthly. There is no report to request, from any operator, that does this for you.

Why does my competitor get mentioned and I do not? Often it is eligibility rather than merit: they are crawlable and you are partly blocked, or their business facts agree across sources and yours disagree. Sometimes it is that they have published something specific and quotable and you have published a description. Occasionally there is no visible reason, which is the honest answer nobody selling this will give you.

Does being cited actually send traffic? Not much of it. Being named in an answer is closer to a credibility signal than a traffic source, so it is worth having and it is not a substitute for the pages that take a booking. Treat it as one more place your name has to be right, for the reasons we set out in what zero-click search means for a small business.

How long does any of this take to show up? Nobody publishes a timescale, because nobody publishes the mechanism. What you can do is date your changes, keep running the monthly log, and judge by whether the pattern moves across several months rather than by one striking result.

Final Thoughts

The part you control is real, it is short, and almost nobody puts it in order. Let the right crawler in, make sure the pages are genuinely reachable, publish answers that lead with the answer and carry your own figures, get your business facts agreeing with each other, and keep a monthly log so you can tell whether any of it moved. That list is the work, and none of it requires a subscription.

The part you do not control is also real, and it is better to hear it plainly than to find it after spending. No operator publishes a citation formula. OpenAI, having documented how to become eligible, says in the same breath that there is no way to guarantee top placement. That sentence is worth more to you than any playbook built on a leak, because it tells you exactly how far confidence can honestly go.

If you would rather have someone go through the crawler access, the business facts and the pages themselves and tell you which part is actually holding you back, that is ordinary work for us at Web Leveling. We build fast custom sites and run the generative engine optimization and search work behind them, owned outright by the business that paid for it. Tell us what is bothering you and we will come back within one business day with a straight read, including when the answer is that you are already doing the parts that matter.

Terms