The honest answer is that you do not know until you have asked properly. Plenty of owners open ChatGPT, type their own company name, get a friendly description back, and conclude they are visible. That test measures nothing.
Around 900 million people use ChatGPT every week, according to OpenAI (February 2026). A growing share of them ask the model for providers, suppliers and products instead of opening a classic search engine. Whether your company shows up in those answers helps decide who still finds you at all.
Does ChatGPT mention my company in the first place?
You find out by asking the question a customer would ask, not the one that already contains your name. Ask about your service in your region, not about yourself.
For example: instead of "What does Acme Ltd do?", ask "Which providers of [your service] are there in [your city]?". Only the second question tests whether the model brings you up on its own. The first only tests whether it can confirm a name you handed it.
Why you must keep your brand out of the prompt
The moment your company name sits in the prompt, the model mostly just echoes it back. You are then measuring self-confirmation, not visibility.
A language model picks up the name from your question and builds a plausible answer around it. That feels like confirmation, but it says nothing about whether the AI would suggest you to a stranger unprompted. And that stranger is the case that counts. Put your own name in the prompt and you are testing the model's politeness, not your reach.
Mention, citation and recommendation are three different things
Whether you appear is not a yes-or-no question. There are three stages, and confusing them leads you to read the result wrong.
- Mention: your name appears in the answer, with no source. The model saw you during training.
- Citation: the AI names or links a source, such as your website or a press article. This happens mostly when web search is active.
- Recommendation: the AI does not just name you, it puts you forward as a fitting option. This is the stage with the highest business value.
Merge these three into a single metric and you make bad decisions. A passing mention buried in a long list is a very different thing from a clear, sourced recommendation at the top.
The 15-minute self-test, step by step
Here is how to reach a first solid impression in a quarter of an hour. Write down every answer instead of just scrolling.
- Collect five to eight questions a customer would type. Use different phrasings for the same thing.
- Ask each question in a fresh, empty chat. Old chat history colours the answer.
- Repeat each question two or three times. You will see the answers drift.
- Turn web search on for one pass and off for another. The answers differ noticeably.
- Record whether you were mentioned, cited or recommended, and in which position.
After fifteen minutes you do not have a measurement, but you have a direction. That is enough to decide whether a thorough measurement is worth it.
Which AI systems you should test
ChatGPT is the start, not the finish. Your customers also ask other systems, and each pulls its answers from somewhat different sources.
Alongside ChatGPT, check at least Google's AI answers, plus Perplexity and one more assistant your audience uses. A provider who shows up strongly in ChatGPT can be missing from Perplexity, because there live web search, and therefore fresh, well-structured sources, weigh more heavily. Use the same questions everywhere, or the results will not be comparable.
Why a screenshot proves nothing
A single screenshot shows one of many possible answers, frozen in one moment. The same question can name a different company five minutes later.
Language models answer probabilistically. Two identical prompts produce different wording and sometimes different recommendations. That is why you measure visibility through frequencies across many runs, not through one image.
Then there is the technical side. According to Cloudflare (August 2025), over 2.5 million websites block AI training bots entirely. Blocking and discoverability are linked: shut every bot out wholesale and you can disappear from some answers without noticing.
One more number, to be read honestly. Gartner forecast in February 2024 that classic search volume would fall by 25 percent by 2026. By mid-2026 that decline has not materialised in that form. Classic search stays important, AI answers arrive as a second channel on top.
From self-test to systematic measurement
The self-test tells you whether a problem exists. It does not tell you how big the problem is.
A reliable measurement works with a battery of questions rather than a single prompt, with many repetitions, across several platforms and in your customers' languages. It keeps mention, citation and recommendation apart and reports frequencies instead of impressions. How such a measurement is built is described on our page on AI visibility.
There are no guaranteed placements, and anyone who promises them has not understood the mechanism. You can improve the conditions under which a model names you, and you can measure the progress. That much, but no more.
If you want to know how often the major AI systems already mention you, a pilot project is the fastest route to real numbers. Reach us through the contact page and we will look at your case.