How to get cited by ChatGPT

Someone asked ChatGPT who to read on your topic. A competitor appeared. You did not. The usual diagnosis is “we need GEO” or “we need an llms.txt.” Often the page is blocked from the search crawler, or the claim is not a sentence anyone can quote, or the team is counting clicks as citations.

There is no published ChatGPT ranking you can buy. How answer engines select sources is the selection playbook: crawl, understand, extract, trust. This page is the operator answer to the question that page does not own: how do I get cited?

ChatGPT is not one door

OpenAI documents three agents. They are not aliases.

OAI-SearchBot is for search. It is how a public page becomes eligible to appear in ChatGPT search answers. If you opt out, OpenAI says the site will not be shown in those answers. It can still appear as a navigational link. That is eligibility, not credit.

GPTBot is for possible training of OpenAI foundation models. Disallow it if that is your training policy. It is a separate switch. You can allow search and refuse training.

ChatGPT-User is a fetch a person triggered. OpenAI says it is not an automatic crawl, that robots.txt may not apply, and that it is not used to decide whether content appears in Search. Do not treat a user-triggered hit as proof you are in the search index.

The two robots.txt tags are independent. OpenAI says a search-side change can take about twenty-four hours. I re-read that page on 16 August 2026. If a vendor slide collapses the three names into one “ChatGPT bot,” throw the slide out.

Eligibility is not a citation

Allowing OAI-SearchBot answers one question: may this URL be a candidate in ChatGPT search? It does not answer will this URL be the source named in the answer?

Google is even plainer about its own surfaces. Its AI optimization guide, last updated 10 July 2026 when I fetched it, says generative features stay rooted in the core Search index and ranking systems. A page has to be indexed and eligible to appear with a snippet. From Google’s seat, AEO and GEO work is still SEO. Prioritize that over “AEO/GEO hacks.”

There is a second Google switch people miss. Search Console has a Search generative AI control. The default is include. If someone excluded the property, your links and content stay out of AI Overviews, AI Mode, and the generative features in Discover. That control is still rolling out to a subset of owners. Check it. Do not invent an opt-in ritual for a default that is already on.

None of that is a citation. A citation is a name.

The standard is already public

CITATION is the live capsule, not a product acronym I am expanding here.

A page worth quoting is discoverable, attributed, answer-shaped, evidenced, and actually used. Search eligibility is not enough. Answer engines take the first clean capsule they can verify.

That is the whole list I will stand on in public. The book that owns the longer operating system, The Findability Engine, is in production, English first. It is not for sale. Do not wait for the hardcover to write a sentence a machine can name.

GEO is the system around the page: retrieve, interpret, extract, verify. AEO is the craft of one URL. They fail together when the first screen is a brand manifesto and the claim lives in a newsletter no crawler can fetch.

What you can actually control

Discoverable. The right agent can fetch a stable 200 with the primary text in the HTML. No accidental noindex. No login wall in front of the sentence you want quoted. If ChatGPT search is the surface you care about, OAI-SearchBot has to be allowed on that URL. If Google’s generative features are the surface, the page has to be ordinary-Search eligible and not excluded in Search Console.

Attributed. One person name. One organization name. sameAs links you actually control. If the machine cannot tell who you are, it will credit someone it can.

Answer-shaped. One sentence that owns the claim, near the top, with the conditions attached. If you cannot point at the sentence, you do not have a page. You have a theme.

Evidenced. A primary source, a date, a limitation. “Schema helps AI” is not a claim. “Google states that structured data is not required for generative Search, and correct markup does not guarantee a rich result” is a claim a skeptic can check.

Used. Someone applied the thing. A citation, a referral, a decision. The last word in the capsule is used, not eligible.

What you cannot buy

There is no ChatGPT citation switch. No vendor has the ranking function. Anyone selling “we will get you cited in ChatGPT this quarter” is selling a hope with a invoice.

llms.txt is not a Google ranking file. Google Search does not use it. Keeping one for other tools neither helps nor harms your Google visibility. It does not replace crawlable HTML.

Structured data is not a citation spell. It can clarify an entity. Google says it is not required for generative Search and does not guarantee a rich result.

FAQPage is not a current rich-result tactic. Ordinary FAQ rich results stopped on 7 May 2026. Google removed the documentation on 15 June 2026. A visible question-and-answer block can still help a reader. Ship it for the reader, not for a carousel that is gone.

Do not manufacture mentions. Google’s own guide tells you to ignore inauthentic mention-chasing. A fake quote next to your name is not authority. It is a faster way to be wrong in someone else’s answer.

Measure three numbers, not one dashboard

A citation is not a visit. A visit is not a conversion.

Appearances. For ChatGPT search, you still mostly look with your eyes and a log: date, prompt, whether your name or URL was in the answer. For Google, Search Console launched Generative AI performance reports on 3 June 2026. They report impressions (links shown) in AI Overviews and AI Mode. The report is rolling out to a subset of properties, and it may stay empty if impressions are too few. That number is Google. It is not ChatGPT.

Visits. Some ChatGPT search links arrive with utm_source=chatgpt.com. Track the referrer. Zero clicks do not prove zero citations. People do not always click a source they already read.

Conversions. Keep the sale on its own line. A named source that never sends a qualified conversation is still a citation. It is not a pipeline.

How answer engines select sources already separates discovery, visibility, citation, and business impact. Do not collapse them here to make a prettier slide.

A fifteen-minute audit

Pick the page you most want ChatGPT to name.

1. Open robots.txt. Is OAI-SearchBot allowed on this URL if ChatGPT search is the goal? Is GPTBot set on purpose, not by accident?
2. Fetch the canonical URL. Status 200? Any noindex? Is the sentence you care about in the HTML without a click?
3. Write the claim in one sentence. Paste the live sentence that owns it. If you cannot paste it, you do not have a capsule.
4. Circle the person or organization named on that page. Follow one sameAs link. If it 404s, the machine will invent you.
5. Write the date you last checked the evidence. If there is no date, the claim is already stale.
6. Write three notes from the last thirty days: an appearance, a visit, a conversion. Leave a line blank if you did not see it. Blank is information.

If step 1 or 3 is empty, you do not have a citation problem. You have an eligibility problem or a page that cannot be quoted. Fix that before you buy a GEO retainer.

The stamp on the desk is unused. Permission is not a name. Write the sentence. Let the right crawler fetch it. Count the three numbers without lying to yourself about which one moved.