How to Check if ChatGPT Knows About Your Website

Why check whether ChatGPT knows about your site at all?

Because some of your customers no longer reach the search results page. A person asks an AI, gets a finished answer with a couple of sources, and calls the number in it. If you are not there, you did not lose to a competitor in the rankings - you were simply not part of the conversation.

The check is free and takes about twenty minutes. It usually ends in one of three outcomes: the AI knows you and cites you, knows you from scraps of other people's sources, or does not know you at all.

What follows is how to check each case and what to do about it. I covered the general logic of AI search separately: what GEO is and why your site should show up in ChatGPT.

How do you check this yourself in twenty minutes?

Four checks, in increasing order of strictness. Run them in a new chat with no history, so the model does not lean on your earlier messages.

Ask about the company by name. "What do you know about the company [name] from [city]?" Look past how polite the answer is and check the facts: are the services, the city and the specialization named correctly. Invented details are a bad sign: it means there is no data and the model is filling the gap.

Ask about the service without naming the company. "Who does [your service] in [city]?" This is the main check: it shows whether you make it into the answer when a person is looking for the service rather than for you. The first check is about recognition, this one is about sales.

Repeat it in systems that search the web. The answers differ: some models answer from memory, others search the internet and add links to sources. Check several - ChatGPT, Gemini, Perplexity, and the AI answers inside a search engine.

Check where the facts come from. If there are source links, look at whether they lead to your site or to a directory, a listing, someone else's article. It often turns out that the AI knows about you from other people's mentions and does not read the site itself.

Write the results down. A month after the fixes, repeat the same set of questions - that is the only honest way to see the change.

Why do AI systems not see your site?

Three reasons, from the most common to the rarest.

The site serves an empty page. The most widespread one. Many modern sites are assembled by scripts right in the browser: a person sees the text, a robot gets an empty shell. It is easy to check - open the page, click "View page source" and look for your heading or a paragraph of text. If they are not in the source code, the page does not exist as far as a crawler is concerned.

The hosting blocks AI crawlers. Bots like GPTBot, ClaudeBot and PerplexityBot have to get a normal response from your server. On some hosting providers and cloud protection services they are closed off by default or served an error. Check the robots.txt file - make sure there is no ban on these bots - and ask your host whether it blocks them on its side. This is the most galling reason: the site is good, and for AI systems it does not exist.

There are no extractable facts. The site is reachable, but there is nothing specific to take from it: the services are described in general terms, there are no prices, no city, no answers to questions. The model has nothing to quote, so it will quote whoever does have it.

What do you do if they are not finding you?

In order, starting with what gives the most.

Make the pages readable without scripts. If the content is only visible after the code runs, you need prerendering or a static build. That is developer work, but without it the rest is pointless.

Unblock AI crawlers. Remove the bans from robots.txt if they are there, and write to your host if the block is on its side. Usually one request settles it.

Give direct answers. Headings phrased as questions, a short answer right under the heading, a frequently asked questions block. The model looks for a self-contained fragment it can drop into an answer - give it one.

Write the facts out in plain text. What you do, who for, in which city, what it costs, how to get in touch. Not as an image, not in a PDF, but as text on the page. What is not in the text does not exist for an AI.

Add markup and llms.txt. Schema.org for the organization and the services, plus a short machine-readable file about your business in the root of the site.

Show up where you get cited. Listings in directories and on maps, industry catalogues, mentions in local sources. The model often learns about a company from exactly there.

How do you tell it got better?

You have to check the same way you did at the start, otherwise there is nothing to compare against.

Repeat the same four questions a month to a month and a half after the fixes, in a new chat. Look at three things: whether you now appear in the answer to the question about the service without the company name, whether the facts about you got more accurate, and whether they link to your site directly rather than to someone else's pages.

Separately, look in your analytics at referrals from addresses like chatgpt.com and perplexity.ai - if they are there, you are already being cited with a link.

An honest expectation: this is a slow process. Models do not refresh their data instantly, and some answer from memory that was formed long before your fixes. The systems that search the internet at the moment of answering react faster.

Is it worth the effort right now?

Depends what you expect. If it is a stream of leads from ChatGPT - no, the share of such visits is still small, and promising you otherwise would be a lie.

But two arguments make the work worthwhile. First: almost everything AI search needs also improves ordinary SEO - content accessibility, speed, structure, direct answers. You are not paying for that separately, you are simply building the site properly. Second: the people who arrive from an AI answer arrive already holding an answer to their question and a recommendation - that is warm traffic, not a random click.

How do I do this?

I build sites so that the page is served as finished HTML: a crawler sees the text without running any scripts. Markup, llms.txt, a "question - answer" structure where it fits.

My own site and this blog are the testing ground: everything described here was checked here first. And I check by measuring rather than by assuming - the same as with speed, where guesses usually point at the wrong cause.

If you want to find out whether AI systems see you - write to me, send the link. I will run the checks and tell you exactly what is in the way: the technical side, missing facts on the pages, or a block at the hosting level.