Diagnosis
Why my brand doesn't appear in ChatGPT
You asked the question, and a competitor came out instead of you. Before rewriting anything, work out which of the four causes applies: they share neither the test, nor the timeline, nor the person responsible.
Nicolas Framery Founder of Etalo, SEO and generative visibility consultant
Collected on 16 October 2026. Every source carries its date in the text.
The essentials in thirty seconds
A brand missing from an assistant's answers is missing for one of four reasons: a technical block stopping the assistant from reading the site, absence from the model's memory, absence from the sources the assistant fetches when answering, or the brand simply not being a plausible answer to the question asked. The test that separates them takes two prompts: the same question with web search on, then off.
- Blocking GPTBot does not remove you from ChatGPT. GPTBot is about training. OAI-SearchBot decides whether you appear in search answers, and the two settings are independent.
- Three answers are not an absence. An assistant is non-deterministic: the same question asked three times gives three answers. Several collections are needed before concluding anything.
- The decisive test takes two minutes: ask the question with web search, then without. Missing in both cases, the problem is the model's memory; missing only with search, it is the sources.
- Repair times run from 24 hours to several months depending on the cause. A robots.txt fix takes a day; a model's memory waits for the next training run.
- The most common cause is not technical: the brand simply is not a plausible answer to the question as asked.
First, check the absence is real
An assistant does not answer the same thing twice. The same question, asked three times five minutes apart, produces three answers that need not cite the same brands. Concluding from one attempt is judging a poll on one ballot.
The order of magnitude can be computed: across one hundred and fifty observed answers, the 95% confidence interval around a 50% presence rate is about eight points. In other words, 46% and 54% are not two different figures. On three answers there is no figure at all.
The practical rule: three spaced collections before talking about a trend, and never a percentage on fewer than a hundred answers. Everything below assumes the absence was observed across several collections, not glimpsed on a Tuesday evening.
The four causes, and how to tell them apart
They look alike on screen — your brand is not there — and have nothing in common. The right-hand column is why the order of diagnosis matters.
| Cause | What you observe | The test | Repair time |
|---|---|---|---|
| 1. Technical block | Missing even when the question names you. The assistant speaks of you vaguely, or points to third-party pages rather than your site. | Read your robots.txt and look for OAI-SearchBot. Also check that no application firewall is refusing its addresses. | About 24 hours after the fix, the delay OpenAI states. |
| 2. Absent from the model's memory | Without web search, the model does not know you at all, or confuses you with another company. With search, it finds you. | Ask the question twice: once with web search on, once off. The gap is the diagnosis. | The next training run — months, with no direct way to speed it up. |
| 3. Absent from the sources fetched | The assistant searches, cites three comparison articles and a forum, and none mentions you. Your site may well be read, but it is not the source chosen. | Read the cited sources one by one. They are displayed, and this is the most instructive part of the exercise. | Weeks to months: you have to exist in those sources, not only on your own site. |
| 4. You are not a plausible answer | The brands cited are not your competitors: different category, different size, different country. The answer is correct, it simply is not about your market. | Read the question aloud and ask whether a buyer asking it would want your offer. Often, no. | Outside the technical scope: this is about the offer and positioning, not the site. |
The 24-hour delay: OpenAI's bots documentation, collected 16 October 2026. The three other timelines are observed orders of magnitude, not vendor commitments.
The trap: three agents, three independent settings
This is the most widespread technical mistake in the category, and it is expensive because it is invisible. At OpenAI, three robots carry three different names and answer to three settings that have nothing to do with each other.
GPTBot is for model training. Blocking it signals that your content should not be used to train foundation models. It decides nothing about your appearance in search.
OAI-SearchBot is for search. It is what surfaces sites in ChatGPT's search features, and sites that opt out will not be shown in ChatGPT search answers — they can still appear as navigational links. OpenAI recommends allowing it.
ChatGPT-User covers user-initiated actions, when someone asks ChatGPT something that requires visiting a page. It is not an automatic crawler, and the documentation states it is not used to determine whether content may appear in Search.
Many companies blocked GPTBot in 2023 to keep their content out of training — a perfectly defensible decision — and now believe that is what makes them invisible. It is not. And symmetrically, allowing GPTBot will not make them appear.
Three numbers to keep in mind
the delay OpenAI states for a robots.txt change to take effect on search
OpenAI bots documentation, 2026
distinct OpenAI agents, and three unrelated settings
GPTBot, OAI-SearchBot, ChatGPT-User
the 95% confidence interval around a 50% rate observed on 150 answers
Computed value, not collected
The interval is computed with the usual formula for a proportion (1.96 × square root of p(1−p)/n). It narrows as the number of answers grows: nine hundred answers bring it down to a little over three points.
The diagnosis, in order
Five steps, from the cheapest to the most expensive. Doing the fifth first is the most common way to lose a quarter.
Repeat the question
Three times minimum, across several days, and ideally across several assistants. Note the brands cited each time. If you appear once in three, you are not absent: you are weakly present, which is a different problem with a different answer.
Open your robots.txt
Look for OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended. Check the rules applying to * too, and the application firewall, which sometimes blocks what robots.txt allows.
Ask the question without web search
The test that separates memory from everything else. If the model does not know you at all offline but finds you when searching, your problem is memory — and your lever is everything written about you elsewhere, not your own site.
Read the cited sources
Open each one. They are what decides the answer. If they are three industry comparisons and a forum, your job is not to write one more page: it is to exist in those four sources.
Re-read the question
The last step, and the one people skip. If the assistant cites brands from another category, the answer is right and the question is not about you. Better to know that before rebuilding the site.
Five reflexes that do nothing
All of them are sold somewhere. None answers any of the four causes.
- Adding an llms.txt file
- No assistant vendor has committed to reading it, and Google states plainly that Search ignores AI-oriented text files. The cost is nil, and so is the effect.
- Adding schema.org markup "for AI"
- Google's official guide states structured data is not required for generative search. It stays useful for other reasons — it is not the cure for an absence.
- Creating page variants to cover derived queries
- This is against Google's spam policies, which the same guide says explicitly. It is nevertheless exactly what some "fan-out optimisation" services sell.
- Unblocking GPTBot hoping to appear
- Wrong agent. GPTBot is about training; ChatGPT search depends on OAI-SearchBot. The two settings are independent.
- Repeating the brand name across pages
- A model does not count occurrences: it picks up passages that answer a question on their own. A page that repeats itself is a page that says less.
Frequently asked questions
I blocked GPTBot in 2023. Is that why I don't appear?
How long after a robots.txt fix?
My site ranks well on Google. Why not in ChatGPT?
Should we write one page per question?
How do I know whether it is memory or sources?
Is an audit worth it if the cause turns out to be memory?
Sources
Role of each agent, independence of the settings and propagation delay: OpenAI's bots documentation, collected 16 October 2026.
AI-oriented files ignored, structured data not required, page variants against spam policies, query fan-out mechanism: Google's official guide for generative AI features, May 2026.
Read next
- The business law firms AI assistants cite most
When a company asks ChatGPT or Google which law firm to call, which names come back? We put 26 questions to five assistants and counted the citations of seven…
- How much does a GEO audit cost
Two of the three most-cited tools no longer show any price at all. Here are the ones still public, what the API calls behind them really cost, and why the gap…
- AI prompt volume does not exist
Every AI visibility tool ends up printing a number next to each question: "2,400 searches a month". Nobody measures that. Here is where the figure actually…
- Getting into Google AI Overviews
This is the generative surface most people see without ever looking for it: it sits above the results, in front of users who have never opened an assistant…
- When AI gets you wrong
A brand that is highly visible in assistant answers, with a wrong price, does not have a visibility problem. It is the one case where a good score is worse…
- Building your question corpus
It is the part of a GEO audit nobody wants to do, and it decides everything else. A badly written corpus produces impeccable figures about questions nobody…
- GEO audit: what it measures, how, and what makes an audit wrong
Public figures on generative search contradict each other: AI Overviews appear on "more than 20%" of searches by one measurement, on 75.24% by another. Both…
- How to get cited by ChatGPT
There is no button. There is a technical door that many sites close without knowing it, three editorial levers that have actually been measured, and a source…
- GEO or SEO: what really changes
Three serious studies measured the same thing — the share of AI citations coming from pages that rank well on Google. They found 16.7%, 52% and a correlation…
- AI visibility tracking tools in 2026
Seven tools, read on the same day from their public pages. Two publish no price at all. And once you divide published prices by the number of answers actually…
- robots.txt and llms.txt: letting AI read your site
It is the one generative-visibility job that can be settled in an afternoon, and the one that makes every other job pointless if you get it wrong. Three files…
- Etalo.ai review
You are looking for reviews of Etalo and you cannot find any. That is normal: the tool is young and has no clients to parade. Rather than manufacture…
- Best Peec.ai alternative
Peec.ai does one thing well: it tracks your visibility in assistant answers, daily, in a dashboard that is easy to read. If that is what you need, it is a good…
- Best Promptwatch alternative
Promptwatch has one real strength few tools have: crawler logs, that is, what AI robots actually come and read on your site. If that is your subject, keep it…
Knowing which of the four applies to you
An Etalo audit separates the four causes across your whole corpus, assistant by assistant, and says which one is fixable.