“How do we rank in ChatGPT?” is a useful business question wrapped in search-engine language.
ChatGPT does not publish a stable results list, a position report, or a formula for brand recommendations. The outcome you can observe is simpler: when someone asks a relevant question, does ChatGPT mention your business accurately, cite one of your pages, or leave you out?
You can improve the conditions around that outcome. You cannot force it. The first step is understanding the three OpenAI user agents most relevant to organic search visibility, potential model training, and user-triggered visits.
TL;DR
- No conventional ranking: ChatGPT does not publish a stable results list or a formula for brand recommendations, so the measurable goal is accurate mentions and citations for relevant questions.
- Search eligibility: Allowing OAI-SearchBot in robots.txt and OpenAI’s published IP ranges makes pages eligible for ChatGPT search, but eligibility does not guarantee inclusion.
- Training is separate: Allowing or blocking GPTBot is a model-training decision, independent of whether pages can surface in ChatGPT search.
- What you can improve: Consistent business facts, pages that answer real buyer questions with evidence, and an accurate public record improve the conditions for visibility without forcing it.
- How to measure: A fixed panel of 30 to 50 buyer questions, run on a schedule and logged in detail, shows whether ChatGPT visibility is improving over time.
The three OpenAI user agents people mix up
OpenAI’s crawler documentation separates search, potential model training, and user-triggered visits. It also documents OAI-AdsBot, which checks landing pages submitted as ChatGPT ads. Because that bot is limited to submitted ad pages, it is outside this organic visibility process.
| User agent | What OpenAI says it does | What allowing it does not promise |
|---|---|---|
| OAI-SearchBot | Automatically crawls pages so sites can surface in ChatGPT search results | A mention, recommendation, citation, or a particular refresh time |
| GPTBot | Crawls content that may be used to train OpenAI’s generative models | That your content will be used, memorized, or turned into a brand recommendation |
| ChatGPT-User | May visit a page because a user asked ChatGPT or a Custom GPT to do so | Search eligibility; it is not an automatic web crawler |
The controls are independent. A publisher can allow OAI-SearchBot while disallowing GPTBot. OpenAI also notes that ChatGPT-User requests are initiated by a person, so robots.txt rules may not apply to them.
One caveat matters when you explain search access. OpenAI says a site that opts out of OAI-SearchBot will not be shown in ChatGPT search answers, but it may still appear as a navigational link. “Blocked crawler, invisible brand” is therefore too absolute.
What crawl access can and cannot do
Allowing OAI-SearchBot makes your pages eligible to surface in ChatGPT search. OpenAI recommends allowing the bot in robots.txt and allowing requests from its published IP ranges.
Eligibility is the right word. OAI-SearchBot is not proof that ChatGPT fetches your page in real time for every question. Allowing it does not establish how ChatGPT evaluates pages or chooses brands. OpenAI does not publish that formula.
The same restraint applies to GPTBot. Its job is potential training access. OpenAI says content it crawls may be used to make its generative models more useful and safe. The documentation does not say that allowing GPTBot causes a brand to become part of default model knowledge, and it does not promise future recommendations.
A practical process for ChatGPT visibility
1. Make a deliberate crawler decision
Check your robots.txt file and any CDN or firewall rules that can block automated traffic. If appearing in ChatGPT search matters to you, allow OAI-SearchBot and the published search-bot IP ranges.
Decide on GPTBot separately. That choice is about whether your content may be used for model training, not whether your pages are eligible for ChatGPT search.
Do not add ChatGPT-User to a checklist as though it were another automatic crawler. It represents certain visits initiated by users, and OpenAI says it is not used to determine Search inclusion.
2. State your business facts plainly
Use the same accurate description of the company, audience, services, location, and product names across the pages you control. A vague slogan can carry the brand voice, but it should not be the only explanation of what the business does.
Consistency reduces the chance that a visitor or a retrieval system has to reconcile conflicting facts. It still does not guarantee how ChatGPT will describe you.
3. Answer the questions buyers bring to the product
Write for the decision someone is making, not for the phrase “ChatGPT SEO.” Useful pages often cover comparisons, alternatives, pricing variables, use-case fit, limitations, and implementation details.
Put a direct answer near the question, then show the evidence behind it. The goal is not to create a sentence that sounds quotable. It is to give the reader enough context to decide whether the answer applies to them.
4. Build an accurate public record
Your own site is only one source people can consult. Keep legitimate directory profiles, review pages, interviews, and partner pages accurate. Correct errors where you have a real editorial path to do so.
I treat independent evidence as good reputation work. I do not call it a documented ChatGPT ranking factor because OpenAI has not published it as one. Do not manufacture forum mentions or reviews in an attempt to simulate corroboration.
5. Publish pages worth using as sources
A useful source makes a specific contribution: original data with a method, a clear comparison, a first-hand test, a documented process, or a precise answer backed by primary sources.
This is where most brand pages come up short. They say the company is experienced or innovative but give the reader nothing to verify. Add the example, calculation, screenshot, or limitation that makes the page useful even when no AI system cites it.
6. Keep dates and claims honest
Show an updated date only when someone has reviewed the content. Replace obsolete product names, screenshots, and instructions when they change. Do not change the year in a title and call that a refresh.
Current information is better for readers and safer if a tool uses the page as a source. I have not seen an OpenAI document promising that a visible update date improves ChatGPT search placement, so I would not present it as a ranking lever.
7. Measure with a fixed question set
I use 30 to 50 questions as a starting range when a team can review that many consistently. A smaller set of high-value buying questions is better than a larger panel nobody maintains. This is an operating sample for comparison over time, not a statistically representative benchmark. Run the same questions on a schedule and record the exact wording, date, model or surface, account state, answer, mentions, errors, and cited sources.
One answer can change with phrasing, context, or product updates. A repeated panel lets you see whether the pattern is improving. It also keeps you honest: crawler eligibility, brand mentions, accurate descriptions, and citations are different measurements.
Tactics I would skip
- Hidden prompt-injection text asks a system to recommend you instead of helping the visitor. It creates a trust problem and OpenAI does not document it as a visibility tactic.
- An llms.txt file can be a useful directory for tools that choose to read it, but OpenAI does not document it as a requirement or ranking control for ChatGPT search.
- A large batch of thin articles about your own company does not create independent evidence. Publish the smaller number of pages that answer real questions well.
- A single flattering answer is not a measurement. Save the prompt, context, date, and sources, then see whether the result repeats.
Does Google visibility carry over?
There is no official mapping from a Google rank to a ChatGPT mention. Good SEO still leaves you with useful assets: pages that can be crawled, clear site structure, accurate business information, and content built around customer questions.
Use that foundation. Then measure ChatGPT directly. Do not assume that position three in Google should produce a recommendation, or that a missing ChatGPT mention proves your SEO failed.
What to expect
OpenAI says robots.txt changes may take about 24 hours to reach its systems. That is a settings-update estimate, not a promise that revised content will appear in ChatGPT within a day or within a few weeks.
There is no reliable public timetable for a page to be mentioned or cited, and potential model-training effects are even less predictable. Make the site eligible, improve the information, and track a consistent panel. Report what changed without inventing a deadline.
If you want a baseline of what ChatGPT currently says about your business and which sources it uses, Book a free 30-minute call. Bring two or three buyer questions and we can inspect the answers together.
