Updated Jack F, Co-founder
Part of the Answyn AEO Handbook.
Now readingWhy ChatGPT leaves a company out
ChatGPT leaves a company out when it cannot reach the company's pages, when its own web searches never retrieve them, when it reads them and finds nothing that fits the buyer's criteria, when the pages it cites make the case for rivals, or when it believes something wrong. Check those five in order. In Answyn's own tracking, ChatGPT read an Answyn page in 49 runs and still left Answyn out of every one.
Why does ChatGPT leave a company out of its recommendations?
ChatGPT leaves a company out of its recommendations when something breaks between the company's website and the answer: the pages cannot be reached, its own searches never retrieve them, it reads them and finds nothing that fits the question, the sources it does read make the case for someone else, or it believes something wrong. Each of those has a check. Run them in that order, because each one assumes the one before it has passed: a page ChatGPT cannot reach cannot be retrieved, and a page it never retrieves cannot be read.
None of the five is a verdict on the product. In the runs we collect, ChatGPT answers a request for recommendations by searching the web and writing from the pages it retrieves: it searched in all 750 ChatGPT runs Answyn collected between 6 October and 4 November 2026.
We use our own brand as the worked example, because it shows up in most of these checks. Answyn sells AI visibility tracking, and in that window the engines we track named Answyn in 2 of 1,575 answers about its own category, both of them in ChatGPT.
Check 1: can ChatGPT reach your pages?
Start with whether OpenAI's crawlers are allowed to fetch your site, and whether they can actually get through. OpenAI documents four user agents, and they do different jobs (OpenAI, Overview of OpenAI crawlers, read 4 November 2026).
| User agent | What OpenAI says it does | What it means for being recommended |
|---|---|---|
| OAI-SearchBot | Surfaces websites in search results in ChatGPT's search features | The rule to check first. Sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links |
| GPTBot | Crawls content that may be used to train OpenAI's generative AI foundation models | Disallowing it tells OpenAI not to use your content in training. It is a separate setting from search |
| ChatGPT-User | Visits a page when someone asks ChatGPT or a custom GPT a question. It does not crawl automatically | OpenAI says robots.txt rules may not apply, because a person started the action, and that it is not used to decide what appears in search |
| OAI-AdsBot | Checks the landing pages of ads submitted to ChatGPT | Visits only pages submitted as ads |
OpenAI says each setting is independent: a site can allow OAI-SearchBot to appear in search results while disallowing GPTBot to keep its content out of training. A site that blocked every AI crawler in one go has also opted out of the search answers where recommendations are made.
A robots.txt file is only half of reach. A CDN or firewall can refuse a request that robots.txt allows, and OpenAI recommends allowing requests from its published IP ranges as well as allowing OAI-SearchBot in robots.txt. After a robots.txt change, OpenAI says it can take about 24 hours for its search systems to adjust.
How to check: read your robots.txt for OAI-SearchBot and any wildcard rule that catches it, then request a page as that agent and see what comes back. Our free AI crawler access check does both for each named AI agent, OAI-SearchBot, GPTBot and ChatGPT-User included. Server logs, or LLM Analytics, show which bots actually request which pages.
Answyn passes this one: ChatGPT retrieved Answyn pages in at least 49 runs in the window.
Check 2: do ChatGPT's own searches find you?
In the runs we collect, ChatGPT rewrites the buyer's question into searches of its own, a process called query fan-out, and the pages those searches return are the pages it reads. It usually ran two or three searches per answer in the runs we read, and the wording kept moving: over 30 runs, a single question produced between 26 and 95 different search wordings, depending on the question.
The searches we read looked for facts. One that ChatGPT ran on 4 November 2026, for a question about which AEO platform a small brand team should choose, read "AEO GEO platform comparison 2026 small marketing teams pricing Otterly Peec AI Profound Scrunch AthenaHQ". It asked for pricing, and it added the names of five rivals that the question never mentioned. It did not add Answyn.
So the check is whether you have pages that answer what those searches look for. A page that never states your price or how to get access, who the product suits, what it covers and where it works gives a search for those facts little to match. For the question this page answers, ChatGPT retrieved no Answyn page in any of its 30 runs. We had no page on it until this one.
How to check: list the questions your buyers ask, write down the facts each one needs, and find the page on your site that states each fact in plain text. Where no page does, the searches have nothing of yours to return.
Check 3: does ChatGPT read your pages and still leave you out?
Being read and being named are separate results, and this is the check where Answyn fails hardest. Between 6 October and 4 November 2026, ChatGPT retrieved an Answyn page in 49 runs across 5 questions and left Answyn out of every one of those answers. In the whole window, it named Answyn in 2 answers.
The pages it read were not obscure. For the question about which AEO platform a small brand team should choose, it read our list of the best AEO platforms for UK businesses in 17 of 30 runs. For a question about where in the buying journey an assistant stops mentioning a brand, it read our handbook page on the customer journey in AI search in 18 of 30.
We cannot see inside the model, so what follows is observation, not a proven cause. Two patterns fit.
- The answer was built on criteria the page did not answer. For the small-team question, the answers ranked tools on entry price, prompt limits and the engines covered at entry level. Our list gave the published prices of rivals, checked on 22 August 2026, and "Enquire" for Answyn, because Answyn publishes no price. Read that way, our own page supplied the facts the answer ranked rivals on, and not the same facts for us.
- The page was a list where the brand is one of many. A roundup of ten platforms gives an answer ten names, and being one of them on our own list did not make Answyn the one the answer picked.
How to check: you need to know which pages the engine retrieved, as well as which it cited. When a page you own is retrieved and you are not named, read the answer for the criteria it ranked on, then check whether the page states your position on each one in a passage that makes sense on its own.
Retrieval is the hard part to see from outside. For tracked ChatGPT questions, Answyn records the searches ChatGPT ran and the pages it retrieved as well as the pages it cited, and the Answyn MCP reads those records on plans that include it. That is where the 49 runs above come from.
Check 4: who does ChatGPT cite instead?
The pages an answer cites when it leaves you out show where the answer was assembled, and so where the fix lives. In Answyn's category, ChatGPT's most-cited sources were of three kinds: vendors' own product, pricing and feature pages; roundups of the best tools; and official documentation from Google, OpenAI and Microsoft. For research-style questions it also cited studies with numbers.
Roundups need the closest look. One of them, technologyadvice.com's "Best AI search monitoring tools for 2026", was cited in 99 ChatGPT runs that left Answyn out, across 15 of our 25 questions. If the answer to your buyers' question is assembled from third-party lists you are not on, no page on your own site will fix it. The fix is on the list: ask to be added, with facts the editor can check. A list cited in an answer that leaves you out is not proof the list leaves you out too, so open it before you write to anyone.
How to check: sort the pages cited for each question you lose into your own, competitors' own, roundups and reviews, documentation, and forums or video. Competitors' own pages are something your pages can compete with. Roundups need outreach first. AEO Hub keeps the citations behind every collected answer, so the sort can run on all of them, not only the one you happened to read.
Check 5: what does ChatGPT believe about you?
The last check covers the answers that do mention you, including those where a buyer asks about you by name: is what ChatGPT says true? Answyn has a recent example in its Brand Facts. In 6 answers between 10 October and 2 November 2026, ChatGPT told the reader that Answyn's plans come with a trial, an offer Answyn no longer makes. There is no self-serve sign-up or free trial: access to Answyn is by invitation, starting with a 30-minute call.
Fix the source of truth first. Find where the old fact still lives, on your own pages and on the third-party pages that describe you, and correct it there. Then recheck the same question over several weeks, because a single clean answer proves little. We set out the method claim by claim in how to check what AI assistants say about you.
In Answyn, Brand Facts lists the claims the engines made about you, and you mark each one correct or incorrect, with a correction where it is wrong. Factcheck rechecks one question over time on ChatGPT or Google AI Overviews and saves each answer with its sources. Neither changes what ChatGPT says. They tell you what to correct, and whether the claim has stopped.
What should you fix first?
Once you know which checks failed, fix them in this order, skipping any that passed. It puts the cheapest fixes first and new writing last.
- Access. If OAI-SearchBot is blocked, or a firewall refuses it, fix that before anything else. It is also the cheapest fix.
- Wrong facts. Correct them at the source, because they reach every answer that already names you.
- Pages that are read but not credited. Update the page ChatGPT already retrieves so it states your position on each criterion the answers rank on, in a passage that stands on its own. An update usually costs less than a new page.
- Third-party lists. Where roundups carry the answer, ask to be added before writing anything of your own.
- New pages. Where no page of yours is retrieved at all, write the page that answers what the searches look for.
Some of it may not be a page at all. In our own case, the small-team answers rank on entry price, and Answyn has no entry price to rank: it publishes no list prices and quotes every plan for the business using it. That is a business decision. A page can state the access route and how pricing works plainly, but it cannot win a shortlist ranked on price. Turning gaps into a content plan starts by setting aside gaps like this one.
Before you pay for anything new, check the advice against the evidence. We grade the common claims, llms.txt and schema among them, in how brands earn citations in AI answers. If you want an agency to make the fixes, our ranking of UK AEO agencies compares six on published criteria. It ranks Tilio first, and Tilio and Answyn are the same company.
Answyn keeps a daily record of the answers behind checks 2 to 5 for the questions you track. Access is by invitation, and every plan is quoted for the business using it. Enquire if you want it run on your own category, or read how Answyn pricing works.
A note on the data
The worked example is Answyn's own tracking of its own category: 25 tracked questions in 8 topics, asked in the UK market between 6 October and 4 November 2026, giving 1,575 answers. That is 750 from ChatGPT, 750 from Google AI Overviews (30 of which showed no overview) and 75 from Claude, last collected on 30 October. Answyn was named in 2 of the 1,575, both by ChatGPT, for a question about the top answer engine optimisation software for UK marketing teams, on 12 October and 4 November. Of 11,097 citations in the window, 17 went to answyn.com, all in ChatGPT.
For the question this page answers, ChatGPT's answers leaned on OpenAI's own help centre, cited in 28 runs that left Answyn out, and on OpenAI's crawler documentation. Google's AI Overviews, asked the same question, leaned on LinkedIn posts, cited in 25 runs. The two engines built their answers from different sources, which is a reason to check each one separately. For Google's engine, see how to check whether AI Overviews use your website.
The example in check 5 comes from Brand Facts: the claims ChatGPT made in answers that discussed Answyn in the same window, which can include tracked questions that name Answyn, beyond the 25 category questions above.
It is one brand in one category and one market, so treat the pattern as the useful part rather than the levels. We sell AI visibility tracking, and we have used our own weak numbers because they are the ones we can show in full. OpenAI's crawler documentation was read on 4 November 2026, and it changes, so check the current version before you edit robots.txt.
Questions people ask
- Does blocking GPTBot stop ChatGPT recommending my company?
- Not in search answers, according to OpenAI's documentation. GPTBot collects content that may be used to train OpenAI's models, OAI-SearchBot is the crawler for ChatGPT's search features, and OpenAI says the two settings are independent. Sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links, so that is the rule to check first.
- How long after I change robots.txt will ChatGPT see my pages?
- OpenAI says it can take about 24 hours from a robots.txt update for its search systems to adjust. Being allowed in is not the same as being retrieved, though. That depends on whether ChatGPT's own searches find your pages, so judge the change over weeks of answers rather than the next morning.
- Why does ChatGPT recommend our competitors and not us?
- Usually because the pages it retrieves make the case for them. In Answyn's category, the sources ChatGPT cited most were vendors' own product and pricing pages, roundups and official documentation, and its answers for small teams ranked tools on entry price and prompt limits. If your pages do not state those facts, or the roundups it reads do not list you, the answer has nothing to recommend you with.
- ChatGPT reads our website. Why doesn't it mention us?
- Because being retrieved is not the same as being named. A mention needs your name in the answer's own words, and the answer names the brands whose pages make the case on the criteria it is ranking by. Check 3 above shows what that looked like in Answyn's own tracking, where a list of ten platforms was read often and Answyn's own row gave no price to rank.
- Will llms.txt or schema markup get ChatGPT to recommend us?
- Neither is one of the five checks here, and we would not start with either. We grade the published evidence on both in how brands earn citations in AI answers, alongside the other common AEO claims.
- How do I fix something wrong that ChatGPT says about my company?
- Correct it where it lives: on your own pages first, then on the third-party pages that describe you. Then ask the same question repeatedly over several weeks, because one clean answer proves little. A tool that records the claims can tell you when the wrong one stops appearing, but it cannot change what ChatGPT says.