How to rank on ChatGPT
Most of the advice on this query is confident and wrong in the same two ways. Here is what the mechanism actually is, in the order that stops you spending a month on the wrong half.
There is no ranking in ChatGPT to hold. It does not order results, it writes an answer and sometimes names sources. What you can influence is whether you are retrievable and quotable, and the highest leverage fact is that ChatGPT search runs on the Bing index, not Google's.
The Bing problem, which almost nobody checks
ChatGPT search reaches the web through the Bing index. That single fact explains more first reports than everything else on this page combined, and it is almost never the first thing a team checks, because every tool they own points at Google.
The failure mode is specific. A company ranks respectably on Google for its money terms, has a healthy Search Console, and cannot understand why the most used assistant in the world never names it. The answer is frequently that Bing has indexed a fraction of the site, or ranks it several pages down, and nothing in the Google shaped toolchain would ever show that.
Two free checks, before any of the AI specific work. Search your own money terms on Bing and see where you actually are. Then verify the site in Bing Webmaster Tools, which costs nothing and is the only place the index coverage is visible. If you are thin in Bing, that is the project, and no amount of writing quotable passages will substitute for it.
Note the control in our crawler data: Bingbot is blocked on 1.6% of the sites we measured, so the index door is almost always open. The gap is coverage and ranking, not access.
Three OpenAI crawlers, and only one decides whether you can be cited
Blocking rates are from our own crawl of 1048 of the most visited websites in August 2026, published with the method and the data.
| Agent | What it does | Blocking it means | Blocked on |
|---|---|---|---|
| OAI-SearchBot | Retrieval for ChatGPT search. | You cannot be cited. This is the one that matters. | 10% |
| ChatGPT-User | Fetches a page because a user asked ChatGPT to open it. | You break a link somebody deliberately chose to follow. | 12.6% |
| GPTBot | Training. | You leave a dataset. It does not remove you from answers. | 21.4% |
The row that costs money is the third one. A site that blocked GPTBot for reasonable reasons about training, and left OAI-SearchBot allowed, has done exactly the right thing. A site that did the reverse, or blocked all three with one wildcard, has removed itself from citation while believing it made a decision about training. The AI crawler checker lists each one separately so the three cannot be confused.
What to actually do, in order
Each step invalidates the ones below it while it is unfixed, so the order is not a preference.
- 01
Check Bing before anything else
ChatGPT search runs on the Bing index. Search your own money terms on Bing, not Google, and see whether you are there at all. A site that ranks page one on Google and page four on Bing is not an AI visibility problem, it is a Bing coverage problem, and it is invisible to everybody working only in Search Console.
- 02
Verify in Bing Webmaster Tools
Free, takes minutes, and it is the only place you can see whether Bing has actually indexed the pages you care about. Most teams in this category have never opened it, which is why the Bing gap goes undiagnosed for months.
- 03
Allow the retrieval crawler specifically
OAI-SearchBot is the agent that decides whether you can be cited. It is blocked on 10% of the sites we crawled. GPTBot, blocked on 21.4%, is training and does not affect whether you appear in an answer. Blocking the first while allowing the second is the exact wrong way round, and it is common.
- 04
Make sure the page exists without JavaScript
Of the sites in our crawl that depend on client side rendering, 27.8% served almost no text to a fetch that ran no script. The assistant received an empty shell. This is invisible in a browser and total in its effect.
- 05
Write the answer where it can be lifted
One question per section, answered in the first sentence under the heading, with the figure next to the thing it describes. An answer spread over four paragraphs of build up does not survive the summarising step, even when the page was retrieved.
- 06
Get described on pages you do not own
Comparison articles, directories, review sites, press. Most of what an assistant says about a company comes from sources the company does not control, and this is the half that no tool and no on page work reaches.
Six things sold as fixes that are not
This query attracts more confident nonsense than any other in the category, so this section is longer than the one above it on purpose.
Paying to be listed in ChatGPT
There is no placement to buy, no submission form and no partner programme that inserts a brand into answers. Anyone selling one is selling something that does not exist. If a vendor cannot name the mechanism by which your brand enters an answer, there is no mechanism.
Getting into the training data
Training data is a snapshot taken before a cutoff and frozen into the weights. You cannot be added on request, and nothing you publish today changes what a model was trained on last year. Essentially all practical work in this field is about live retrieval instead, which is a different pipeline with a different crawler.
An llms.txt file as the fix
No operator of a major assistant has documented that its retrieval pipeline fetches your llms.txt while answering a question about you. It is cheap and defensible for other reasons, covered on our llms.txt page, and it is the last thing to do rather than the first. Publishing one while the retrieval crawler is blocked is decoration on a locked door.
Stuffing the page with the brand name
The research that coined generative engine optimization tested this directly across roughly 10,000 queries and found keyword stuffing had close to no effect on whether a generative engine used a source. Adding citations, statistics and direct quotations did.
A "ChatGPT schema"
There is no schema.org type that signals eligibility to an assistant, and Google states outright that no special structured data is needed for its own AI features. Valid schema that matches the visible page helps a machine understand what it read. No markup makes a model quote you.
One screenshot as proof of anything
Generative answers vary between runs. The same question can name you at ten in the morning and not at noon with nothing changed. A screenshot is one sample, and both the vendor showing you a good one and the competitor showing you a bad one are showing you a single roll of a die.
How to tell whether it worked
Not by asking once. Generative answers vary between runs, so the unit is a rate across a fixed set of questions, rerun on a cadence. Twenty buying questions asked monthly for six months is a finding. One question asked today is an anecdote with a timestamp.
Record four things each time: whether you were named in the answer, whether you were linked as a source, who was named instead, and which sources the answer drew on. The competitor list is the most actionable line, and the source list tells you whether your remaining problem is your own pages or the third party coverage you do not control.
And keep the deterministic half beside it. Bing coverage, crawler access and render output do not vary between two samples, so when a rate moves you can tell whether something real changed or you caught a different roll.
Prefer to work it by hand? The library has a prompt for auditing a site against ChatGPT retrieval, one for reconstructing what ChatGPT search actually favours, and one for getting named in answers. Related reading: AI search optimization surface by surface, LLM SEO and how AI visibility is measured.
Questions people ask about ranking on ChatGPT
How do I rank on ChatGPT?
There is no ranking to hold, which is the first thing worth correcting. ChatGPT does not order results; it writes an answer and sometimes names sources. What you can influence is whether you are retrievable and quotable: be present in the Bing index, allow OAI-SearchBot, serve content that exists without JavaScript, and answer narrow questions in passages that can be lifted whole.
Does ChatGPT use Google or Bing?
ChatGPT search runs on the Bing index rather than Google's. This is the single most useful fact on this page, because it means a site that performs respectably on Google and thinly on Bing can be strong in classic search and close to invisible in the most used assistant there is. Check your money terms on Bing and verify the site in Bing Webmaster Tools.
Can I pay to appear in ChatGPT answers?
No. There is no placement to buy, no submission process and no partner arrangement that inserts a brand into generated answers. What exists is eligibility, which is technical and checkable for free, and being the best available source for a question you never see, which is editorial and slow.
Which crawler do I need to allow for ChatGPT?
OAI-SearchBot is the retrieval crawler and the one that decides whether you can be cited. ChatGPT-User fetches a page because a user asked ChatGPT to open it. GPTBot is training and blocking it does not remove you from answers. In our crawl of 1048 sites, OAI-SearchBot was blocked on 10% and GPTBot on 21.4%, which means a lot of sites have blocked the harmless one and kept the important one open, or the reverse, without intending either.
How long does it take to appear in ChatGPT?
The mechanical fixes take effect as fast as the crawlers revisit, which is usually days. Being chosen as a source is slower and is not on a schedule, because it depends on your page being the best available answer to a sub question and on third party sources describing you consistently. Anybody quoting a timeline is guessing.
Why does ChatGPT name my competitor instead of me?
Read the sources it cited. If they are the competitor's own pages, they answered the narrow question better or you were not retrievable. If they are third party pages, comparison articles, directories, reviews, then this is a corroboration problem and no work on your own site fixes it. The distinction changes what you do next entirely, and the free checker shows you the source list.
Is ranking on ChatGPT different from ranking on Google?
It shares the foundation and diverges above it. Both need pages that can be crawled, indexed and understood. Then Google gives you a position in a list on its own index, and ChatGPT gives you a mention or nothing, drawn from Bing, with no position to hold and answers that vary between asks. Measurement differs accordingly: a rate across a prompt set rather than a rank.
Check the mechanical half in about a minute
The free audit runs 36 checks on your live pages, including whether each OpenAI agent is allowed and what a crawler without JavaScript receives. Free, no card, and the report is yours whether or not we ever speak.