Why ChatGPT does not recommend you, even when you rank first

Companies that arrive with “why doesn’t ChatGPT recommend us” often have surprisingly good SEO. They are on page one, the site is technically sound, the content exists. And the model still names three competitors and not them.

That is neither chance nor bad luck. What AEO is and how a model picks its sources is covered separately. This article is something else: a list of specific faults, ordered by how often we see them.

1. You blocked the AI crawlers without knowing it

The most common cause of all. GPTBot, ClaudeBot, PerplexityBot, Google-Extended and others have their own names and can be blocked individually in robots.txt. Some content management systems and security plugins block them by default, and administrators occasionally add them out of caution.

The result is simple: the model cannot reach your site, so it does not mention you.

How to check: open yourdomain.com/robots.txt and look for those names. Time to fix: minutes. This is the only item on the list you can resolve this afternoon.

2. Your content exists only in JavaScript

Google renders the page and waits for it to fill in. Most AI crawlers do not. They take the HTML as the server sent it and read what is in there. When text, prices or specifications are added by a script in the browser, the model sees an empty shell.

Links that are not <a href> but onclick fare the same way. The model will not follow them, so it never reaches your subpages.

How to check: turn off JavaScript in your browser and load the page. What you cannot see, the model cannot see either.

3. You are a website, not an entity

A model does not work with documents but with things: companies, people, products. To recommend you, it has to be confident about who you are, what you do, and whether you are the same company written about elsewhere.

For that it needs anchors: Organization structured data, the same name and address everywhere, links to verifiable profiles, named people. Without them you are a handful of disconnected pages, and the answer goes to somebody who is legible instead.

4. Your site talks about itself instead of answering questions

“We are a dynamic company with years of experience” answers no question anyone actually asks. A model assembles its answer from sentences that can be lifted and that claim something.

This can be lifted: “A custom company website costs anywhere from hundreds to tens of thousands of euros, depending on scope and integrations.” Specific, verifiable, answers the question. You need many sentences like that.

5. Nobody writes about you anywhere else

This is the uncomfortable part and it cannot be solved with code. A model does not draw only on your website. It pulls from reviews, directories, professional discussions, articles and profiles. When there is no mention of you anywhere else, it has exactly one source: you. A company claiming about itself that it is the best.

A competitor with eighty reviews and three mentions in articles has five independent confirmations. The model picks them, and it is right to.

6. Your answer is in a PDF or an image

A price list as a picture, specifications as a scan, a case study as a downloadable PDF. That works for a human; for a model it mostly does not exist. Whatever you want cited has to be text in HTML.

7. No date, no author, no source

Models favour content where they can tell when it was written and who stands behind it. An article without a date is suspicious when the question is “what are the current prices”, so it goes unused. A named author with a profile makes the same text considerably more citable.

8. You are measuring it with the wrong ruler

The last reason is not a fault in the site but in expectations. When a model uses your content, the visitor often does not click. They got their answer and left satisfied. In Analytics that looks like a decline.

This is why AI visibility is not measured in traffic but in how often and in what context models mention you on questions from your field. While you watch clicks only, it will feel like nothing is happening even if you appear in answers every day.

How to check it in twenty minutes

1. Write five questions a customer would ask about your field. Not your company name, but “who builds custom e-shops in Slovakia”.
2. Ask them in ChatGPT, Perplexity and Gemini. Note who gets named.
3. Open your robots.txt and check the AI crawlers.
4. Turn off JavaScript and look at your own homepage.
5. Search your company name and count how many results are neither your site nor your social profiles.

After twenty minutes you know whether you have a technical problem, a content problem, or the problem of nobody writing about you. Those are three different repairs, and the order is exactly that: technical, content, mentions.

What can be promised and what cannot

The first two items you fix in a day. Items 3, 4, 6 and 7 are weeks of work and they are in your hands. Item 5 takes months and does not depend on you alone.

Nobody can guarantee that ChatGPT will cite you. Models change, sources get reassessed, and a warranty on somebody else’s algorithm is a promise that cannot be kept. What can be done is removing the reasons it passes you over today.

The full process, including content and measurement, is described under our AI visibility service.