How do I check if ChatGPT knows about my website?

Ask it in a fresh conversation with memory and personalisation turned off, or logged out, so you are not testing what it knows because you told it. Ask what your product is and who it is for, then ask the category question without naming yourself, then describe the problem in a user's words.

What does it mean if ChatGPT has never heard of my product?

For a young domain it is the expected result and not alarming. It means you are not in the training data and have not been retrieved. Confirm the search crawler is not blocked and your pages are readable without script execution, then give retrieval something specific to find.

Why does ChatGPT describe my product incorrectly?

Because the material available about you is unclear, so the description assembled from it is unclear. This is the most fixable failure. Read your homepage as a stranger and count how many sentences state plainly what the thing does versus how many describe a vision, then rewrite the directory listings that were pasted everywhere.

I am described correctly by name but never appear in category answers. Why?

That points at third-party evidence rather than your own pages. Category answers draw heavily on comparison pages, directories, review sites and forum threads. If none of those mention you, you are not a candidate no matter how good your own site is.

Which prompt matters most?

The one describing the problem in a user's words, because it is what real users type. People with a problem describe the problem rather than naming a category. Every problem prompt where a competitor appears and you do not is a page you have not written.

Ask the assistants about yourself and read the answer carefully

This is the cheapest diagnostic available to you. It costs twenty minutes, needs no tools, and the shape of what comes back points directly at what to change.

Run the free check

Run four types of prompt against the assistants: your product by name, your category without naming yourself, the specific problem in a user's words, and a direct comparison. Each failure mode points to a different fix. Being unknown by name means a reachability or coverage problem, being absent from category answers means a third-party evidence problem, and being described wrongly means your own pages are unclear.

How to run the test properly

A few rules first, because a sloppy version of this produces conclusions that are worse than no conclusions.

Use a fresh conversation with no memory of you and no prior context, and if the assistant has personalisation or memory features, turn them off or use a logged-out session. Otherwise you are testing what it knows about you because you told it, which is not the question.

Run each prompt on at least two assistants, since they differ substantially. Record the answers verbatim in a document with the date, because the value of this test compounds when you repeat it in a month and can compare.

Do not argue with the answers or correct the assistant mid-conversation. You are taking a measurement, not having a debate.

The four prompts

Ask these in order, because each one tests a different thing.

By name: what is [your product name] and who is it for. This tests whether you are known at all and whether you are described correctly.
By category, without naming yourself: what tools are there for [what your product does]. This tests whether you are a candidate in your own category.
By problem, in a user's words: I am trying to [the situation your product solves], what should I use. This is the realistic prompt and the most important of the four.
By comparison: how does [your product] compare to [a known competitor]. This tests what third-party sources say about you.
Optionally, the trap prompt: ask something your product cannot do and see whether the assistant claims it can. A confident wrong claim about your product is worth knowing about.

Reading a blank

The assistant says it has no information about your product, or invents something unrelated.

For a young domain this is the expected result and it is not alarming on its own. It means you are not yet in the model's training and have not been retrieved.

The fix is in two parts. Confirm you are reachable, meaning the search crawler is not blocked and your pages are readable without script execution. Then give the retrieval something to find: pages that answer specific questions, and descriptions of you on sites you do not own.

One caution. If the assistant confidently invents a description of your product rather than saying it does not know, that is worth recording, because a plausible-sounding wrong description can circulate. The remedy is the same: give it accurate specific material to find instead.

Reading a wrong or vague description

The assistant knows you exist but describes you inaccurately, or in terms so general they could apply to anything.

This one is almost always your own fault and it is good news, because it is the most directly fixable failure. It means the material available about you is unclear, so the description assembled from it is unclear.

Go and read your own homepage as a stranger. Count the sentences that state plainly what the thing does and who it is for, against the sentences that describe a vision. If the second group is larger, that is your answer.

Then check your directory listings, which are often the source of a vague description because they were written quickly and pasted everywhere. Rewrite them, individually, specifically.

Reading absence from the category answer

You are described correctly when asked by name, but never appear when the category question is asked without naming you. This is the most common result for a competent new product.

It points at third-party evidence rather than at your own pages. When an assistant is asked which tools exist in a category, it draws heavily on aggregated sources: comparison pages, directories, review sites, forum threads. If none of those mention you, you are not a candidate, however good your own site is.

The fix is the listing work: get described accurately in places you do not own, with a different description each time, and participate genuinely in the communities where your category is discussed.

Reading the problem prompt

The third prompt is the one that matters most, because it is the one real users type. Somebody with a problem describes the problem. They do not know your category name and they certainly do not know yours.

If you appear here, you are in good shape, and this is the result worth protecting.

If you do not, look at what did appear and read those pages. Usually you will find they answer the specific question directly, near the top, in plain language, on a page devoted to that question. And usually you will find your own site covers the same ground somewhere in the middle of a page about something else, or not at all.

That gap is your content plan. Every problem prompt where you are absent and somebody else is present is a page you have not written yet.

Turning it into a habit

Run this monthly, on the same set of prompts, and keep the answers with dates. Twenty minutes a month.

The single measurement tells you where you stand. The series tells you whether the work is doing anything, which is the harder and more valuable question, and it will show movement long before the traffic numbers do.

Keep the prompt list fixed. The temptation is to improve the prompts each time, and improved prompts destroy the comparison. Add new ones by all means, and leave the originals unchanged.

Part of a larger guide

This page is one part of The AI channel. The other parts:

Questions people ask

How do I check if ChatGPT knows about my website?: Ask it in a fresh conversation with memory and personalisation turned off, or logged out, so you are not testing what it knows because you told it. Ask what your product is and who it is for, then ask the category question without naming yourself, then describe the problem in a user's words.
What does it mean if ChatGPT has never heard of my product?: For a young domain it is the expected result and not alarming. It means you are not in the training data and have not been retrieved. Confirm the search crawler is not blocked and your pages are readable without script execution, then give retrieval something specific to find.
Why does ChatGPT describe my product incorrectly?: Because the material available about you is unclear, so the description assembled from it is unclear. This is the most fixable failure. Read your homepage as a stranger and count how many sentences state plainly what the thing does versus how many describe a vision, then rewrite the directory listings that were pasted everywhere.
I am described correctly by name but never appear in category answers. Why?: That points at third-party evidence rather than your own pages. Category answers draw heavily on comparison pages, directories, review sites and forum threads. If none of those mention you, you are not a candidate no matter how good your own site is.
Which prompt matters most?: The one describing the problem in a user's words, because it is what real users type. People with a problem describe the problem rather than naming a category. Every problem prompt where a competitor appears and you do not is a page you have not written.
How often should I run this test?: Monthly, on a fixed set of prompts, recording the answers verbatim with dates. The single measurement tells you where you stand and the series tells you whether the work is doing anything, which will show movement well before traffic numbers do. Do not improve the prompts between runs or you lose the comparison.

Run the automated version

The free Get Found Check does the reachability and readability half of this for you, so the twenty minutes you spend goes on reading the answers rather than diagnosing the plumbing.

Run the free check Have Licheo do it for you