Get your business named in ChatGPT answers: the whole mechanism, and which parts are actually levers
There is no ranking API and no way to submit a site to ChatGPT. A prompt that scores every lever you have been sold as documented, inferred or undocumented.
- Works in
- ChatGPT, Claude, Gemini
- You need
- One sentence on what you sell and to whom · Three questions a buyer would ask an assistant · Anything you have been sold or told is a ranking lever
- Written for
- how to rank on chatgpt
Scored by our own engine
This page, run through the audit we sell. Measured 4 August 2026.
There is no ranking API, no submission form and no published ranking documentation for ChatGPT. That single fact rules out most of what is sold under the phrase “how to rank on ChatGPT”, and it leaves a smaller set of things that are genuinely yours to change. The prompt below scores those, marks each one as documented, inferred or undocumented, and names the things you have been told are levers that are not.
What is known, what is inferred, and what nobody has published
OpenAI, Anthropic and Perplexity all publish crawler documentation. None of them publishes anything about ranking. That asymmetry is the whole shape of this problem, and it is why the prompt insists on three evidence labels rather than one.
Documented is a short list. The crawler names, what each one feeds, and how to allow or block them in robots.txt are all published by their operators. So is Google’s rendering behaviour. That is close to the end of it.
Inferred is where the useful work sits. A retrieval system chunks a document, so a passage that depends on the paragraph above it is a worse candidate than one that does not. That follows from how these systems are built, and no operator has confirmed it. Acting on it is still rational, because the cost is an edit and the same edit helps a reader.
Undocumented is the honest label for most of the rest, including whether the assistant crawlers run JavaScript at all. Not confirmed, not denied, simply unwritten.
The three questions underneath the one question
Being named breaks into three separate jobs, and they fail independently.
Can a crawler fetch you. Binary, cheap to check, and the one place a single line in robots.txt undoes everything else. Start with the AI crawler checker.
Is there anything in the response to read. A client rendered site can be wide open and effectively empty, which is the site level and technical version of this question.
Would a passage of yours answer the question better than somebody else’s. Partly your page and partly the rest of the web, which is the half most retainers omit.
Two narrower pages sit under this one. ChatGPT’s search path retrieves live pages and is a different mechanism from what the model recalls from training, which is worth separating properly. And if you want the work in order rather than as a map, there is the procedure.
You are an AI visibility strategist. I want an honest map of how a business
gets named in a ChatGPT answer, and a score against it. Start from this fact
and do not contradict it anywhere in your output: OpenAI publishes no ranking
API, no submission form and no ranking documentation, so nothing in your
answer is a confirmed ranking factor.
PART ONE, THE SCORED RUBRIC. Produce a numbered list, not a table. One item
per lever. For each item give exactly four lines:
- LEVER: what it is, in one plain sentence.
- EVIDENCE: one of three words and nothing softer. DOCUMENTED means an
operator has published it. INFERRED means it follows from how retrieval
systems work but no operator has stated it. UNDOCUMENTED means nobody has
published anything either way. Never write DOCUMENTED unless you can name
the publisher and the document.
- SCORE: 0 to 3 for my business, with the specific reason. 0 means this is
working against me right now.
- CHANGE: the single next action, sized so somebody could start it this week.
Order the list by how much of it is inside my control, most controllable
first, rather than by how much difference you think it makes. You do not know
how much difference any of it makes.
PART TWO, THE THINGS THAT ARE NOT LEVERS. List anything I have been sold or
told about that does not work the way it was described, and say plainly what
it does instead. Include any of these that appear in what I paste: submitting
a site to ChatGPT, paying for inclusion in answers, a file that guarantees
citation, a schema type that guarantees citation, a keyword density target for
AI. If I have not mentioned something, do not invent a strawman for it.
PART THREE, WHAT IS NOT MINE. The parts of being named that do not live on my
site at all: what third party pages say about me, whether I appear in the
comparison articles and listicles an assistant is likely to retrieve, and
whether anybody has written about me recently. Be specific to my business.
Constraints you must obey:
- Do not state or imply that any action causes a ranking in ChatGPT. There is
no published ranking system to rank in.
- Do not invent statistics. No percentage of searches, no adoption figures, no
citation share, no traffic loss numbers. If I ask you for one later in the
conversation, refuse again.
- Do not claim any assistant reads any particular file unless you can name the
operator document that says so.
- If you are unsure whether something is documented, mark it UNDOCUMENTED. An
honest gap is more useful to me than a confident guess.
What I sell, and to whom: [ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]
Three questions a buyer would ask an assistant: [THREE QUESTIONS, ONE PER LINE]
What I have been sold or told is a ranking lever: [PASTE THE PITCH, THE ADVICE OR THE ARTICLE, OR WRITE "none"]
My site: [YOUR DOMAIN, AND THE THREE URLS THAT MATTER MOST]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]- Decides which levers matter. A local dental practice and a developer tools company are named in answers for almost opposite reasons, one through directories and reviews it does not control, the other through documentation it does. Without the buyer in the sentence the rubric comes back generic.
[THREE QUESTIONS, ONE PER LINE]- The questions a buyer would type into an assistant, in their words. "Which CRM works for a two person agency" is a question. "Best CRM" is a keyword, and it scores the wrong thing, because nobody asks an assistant a two word query and stops there.
[PASTE THE PITCH, THE ADVICE OR THE ARTICLE, OR WRITE "none"]- The proposal, the LinkedIn post or the agency deck that made you look this up. Part two works on it directly, and it is the fastest way to find out whether you were sold a mechanism or a vocabulary. Writing "none" skips that section cleanly.
[YOUR DOMAIN, AND THE THREE URLS THAT MATTER MOST]- Three URLs, not a sitemap. The rubric scores what those pages do, and three specific pages produce three specific actions. A whole domain produces advice you have already read.
How to run it
- 01Settle the fact before you run anything
Read OpenAI's own bots documentation first, because it is short and it is the thing most advice in this area quietly contradicts. It describes what the crawlers are for and how to block them. It does not describe how anything is ranked, because nothing about ranking is published.
- 02Write three questions a buyer would actually ask
Say them the way somebody types them into a chat window, in a full sentence, including the constraint that matters to them. The questions people ask an assistant are longer and blunter than keywords, and the length is the point: it is what makes a specific page a better answer than a generic one.
- 03Paste whatever made you look this up
The deck, the cold email, the blog post. Part two of the output separates the part that describes a real mechanism from the part that describes nothing. If nobody has sold you anything, write "none" and the section is skipped rather than padded.
- 04Read the EVIDENCE column before the SCORE column
A lever marked DOCUMENTED without a named publisher and document is a fabrication, and it is the failure this prompt is most likely to produce, because a clean confident taxonomy is what models have read. If you see one, treat the rest of the output with the same suspicion and rerun it.
- 05Fix the binary thing first
Crawler access is the only part of this that is on or off rather than a matter of degree. Check whether GPTBot, OAI-SearchBot and the rest are allowed in your robots.txt before you spend a week on content, because no amount of good writing survives a Disallow line.
- 06Take the top scoring change and stop there
The rubric will hand you eight actions and you will do none of them. Take the highest item that is inside your control, do it properly, and rerun the prompt in a month with the same three questions. A repeated measurement on a stable input is the only feedback loop available here.
Questions people ask
How do you rank on ChatGPT?
You cannot, in the sense the question implies. OpenAI publishes no ranking system, no ranking factors and no way to submit a site, so there is no ranked list to enter. What you can influence is narrower and real: whether the crawlers are allowed to fetch your pages, whether those pages contain readable text before JavaScript runs, whether a passage answers a question on its own, and what other sites say about you.
Can I submit my website to ChatGPT?
No. There is no submission form, no inclusion programme and no paid placement in ChatGPT answers. Anyone offering to submit your site or guarantee a citation is describing something that does not exist. What does exist is ordinary crawling, which you allow or block in robots.txt.
Does ChatGPT use Google rankings?
Not in any way OpenAI has documented. ChatGPT search has its own crawler, OAI-SearchBot, which exists to build the index behind links shown in ChatGPT. Whether and how that index relates to any other search engine is not published, so treating your Google position as a proxy for it is a guess.
How long does it take to be named in ChatGPT answers?
Nobody can tell you, and the honest reason is that there is no published index status to check. A crawler has to fetch the page, the page has to be selected for a given question, and neither event is visible to you. Treat any specific timeline you are quoted as invented.
Is being named in ChatGPT worth anything if nobody clicks?
That depends on what you sell, and it is a real question rather than a rhetorical one. For a business whose revenue depends on the visit itself, an answered question can be a lost transaction. For a business selling something the answer leads to, being the named recommendation is worth more than the click was.
Unsubscribe in one click. We never pass your address on.