Your Brand Is Missing from ChatGPT: the Likely Reasons

You typed the question a buyer would type. The best supplier of what you sell, for the kind of customer you serve. The model answered fluently, named a handful of companies, and none of them was you. One of them was a competitor you do not take seriously.

Before you buy a solution, get a diagnosis. There is a short list of reasons this happens, each one checkable, and they are not equally likely. Work down it in order. Most brands find their answer before they reach the bottom.

You are blocking the crawlers and nobody remembers doing it

Rule this out first: it is common, invisible from the outside, and completely disqualifying. When the scraping backlash arrived, a great many sites added blanket disallow rules for AI crawlers. Sometimes on legal advice, sometimes as a default toggle in a plugin or a CDN template nobody read. The decision was never revisited, and the people who now want AI visibility are not the people who made it.

How to check. Open yourdomain.com/robots.txt in a browser. Look for user agent blocks naming GPTBot, ClaudeBot, PerplexityBot, Google-Extended or CCBot. A Disallow: / beneath any of them means you have opted out of that system's ability to read you.

Then check the layer robots.txt cannot show you. Bot filtering at your CDN or firewall can turn these crawlers away before they ever see your rules, and it leaves no trace in the file. Ask whoever administers that layer what the bot policy is, and search your access logs for those user agent strings. If they never appear, something upstream is refusing them.

Fast to fix, with one caveat: allowing a crawler makes you readable from now on. It does not place you inside knowledge a model was already trained on.

You exist only on your own website

If the only place on the web that describes what you do is a domain you own, a model has no independent evidence about you. It has your marketing, which it has every reason to treat as marketing, and nothing else. Models assemble their picture of a category from what the entire web says about it, and a company living inside its own walled garden is close to invisible in that picture, however good the garden looks.

How to check. Search for your brand name while excluding your own domain from the results. Read what remains, discarding the directory listings that scraped your company registration, because those tell a model nothing beyond the fact that you legally exist. What you want is substance written by other people: a review with an actual opinion in it, a forum thread where someone explains why they chose you, a comparison article, documentation on a partner's site.

If there is nothing of that kind, you have found your reason. You have also found the slowest problem on this list, because you cannot write it yourself.

Your content is not really text

What renders beautifully in a browser is not necessarily what a crawler receives. Three patterns cause this, and all are quietly fatal. Your key information lives in images: pricing tiers, the specification table, the comparison chart, all rendered as graphics and therefore opaque to a system reading your source. Your documentation lives in downloadable PDFs, far less likely to be read and reused than the same content in HTML. Or your site renders entirely on the client, so the first response is a skeleton and the words only appear after JavaScript has run.

How to check. Fetch your own page the way a crawler would, with a command line tool rather than a browser, and read the raw response. Or turn JavaScript off in your browser and reload. Whatever text you can see is roughly what a machine gets. If your value proposition is not in there, for this purpose it does not exist.

You are described so generically that you attach to no question

Being readable is not the same as being memorable. Plenty of brands are perfectly crawlable and still never surface, because nothing in how they describe themselves connects them to a specific problem a person would actually ask about.

How to check. Take the sentence that introduces your company on your homepage. Remove your brand name from it. Now ask whether you could paste that sentence onto your closest competitor's site without anyone noticing. If you could, then as far as a model is concerned you are interchangeable, and interchangeable companies are the ones dropped when an answer has room for only a few names.

The remedy is uncomfortable rather than difficult. Say the specific thing. Name the use case you are genuinely good at. Say who you are not for. Precision is what makes you retrievable for a question. Vagueness is a decision to be relevant to none of them.

You are asking a question that cannot fail

Some of the brands convinced they are invisible have never actually tested. Asking a model what your company is will always produce an answer, because you handed it the name and it will find something to say. That measures nothing beyond its willingness to be polite.

How to check. Ask the questions people ask before they have heard of you. Who supplies this component in Europe. What is the best tool for this job at a company of this size. Which options are worth comparing for this problem. Those questions have no safety net, and your absence from them is your real position.

While you are there, ask each one several times, in fresh conversations. These models are probabilistic and the lists will differ. A single run is not a measurement. It is an anecdote, and strategies get built on anecdotes with alarming regularity.

You are visible in English and asking in another language

This one blindsides exporters. Visibility does not travel across languages, because a model's understanding of your category in German is assembled from German text, and in Spanish from Spanish text. A brand it recommends confidently in English can be entirely absent from the identical question asked in the language of a market where it earns a large share of its revenue, purely because that language's web has little to say about it.

How to check. Translate the same buyer question into every language you sell in, and run it in each. The competitors that come back will change, often to local companies you have never benchmarked against, because the model reflects what that language's web discusses rather than what your market map says.

You are simply too new

Worth naming, because it is the one cause that is nobody's fault. A model's trained knowledge is fixed at the point its corpus was assembled, and a company that launched afterwards is not in there and cannot be argued into it. How to check. Compare a platform that visibly cites live sources against one answering from memory. If you show up only where sources are listed, your website is not the problem. Time is, plus the second cause on this list.

What to do with the diagnosis

These causes are not equally expensive. Unblocking a crawler takes an afternoon. Getting your pricing out of an image takes a sprint. Building a reason for other people to write about you takes quarters, and no budget compresses it, because you are waiting on strangers to form an opinion.

What none of it does is guarantee the machine changes its mind. That is the sentence most of this industry declines to write. You can remove every obstacle here and still find a model favouring a larger, older, more discussed competitor, because that is what the text it read told it to do.

What you can have is an accurate picture instead of a guess. Checking by hand works once. Doing it across four platforms, dozens of buyer questions and every language you sell in, repeated often enough for the noise to average out, is not something a person keeps doing by hand. That is the job PSentry exists to do: it runs your prompts across ChatGPT, Claude, Gemini and Perplexity, per language and per market, and shows where you are named, which sources the models lean on, and who is recommended in your place. It runs that check twice each month, not in real time. It will not nudge any model's answer in your favor, and clearing every item here is no guarantee of being named: only a repeated look at whether you are.

Frequently Asked Questions

How long after a fix should I expect to appear?

There is no timetable to plan against. Systems that fetch live pages while answering can reflect a change quickly. Trained knowledge updates when the model does, on the vendor's schedule. Treat each fix as removing a reason to be ignored, not as an action with a delivery date.

I appear in ChatGPT but not in Gemini. What does that mean?

That the platforms disagree, which is the normal state of affairs. They were trained on different material, they retrieve differently, and some lean on live sources while others answer from what they absorbed. Uneven presence is the standard condition, not the symptom of a specific fault.

The model talks about my company but gets the facts wrong. Same problem?

A different one, with a different cause. Being absent means there is nothing to read about you. Being described incorrectly usually means what exists is outdated, contradictory across sources, or too vague to constrain a guess. The remedy is a current, factual page that states the correct thing plainly enough to be reused verbatim.

Could a competitor be doing something to keep me out?

There is no mechanism for that. Nobody can suppress a brand inside a generative answer, in the same way nobody can guarantee one. If a competitor is named and you are not, the likeliest explanation is that the web contains more, and clearer, text about them.

Does buying advertising on these platforms help?

Advertising and a model's own recommendations are separate systems. Paying for placement does not make a model choose your name when it answers from what it knows, and treating the two as the same thing is a reliable way to misread your results.