How to Check if ChatGPT Mentions Your Brand

On 2026-10-25 Leadee checked the 7 websites of Netbase JSC, the company behind Leadee, on three answer engines. The OpenAI part used gpt-5 with the web_search tool through OpenAI's official API and stopped after 25 runs, when the account ran out of credit. In those 25 runs a Netbase brand was named in 3 (12%, 95% Wilson interval 4.2% to 30.0%), and the same 3 cited the brand's own domain. All three were CMSmart (cmsmart.net, part of Netbase JSC) on US questions about WooCommerce product designer plugins. With 25 runs this is an indicator, not a measurement to plan around.

The bigger lesson came from the question type. Pooled over all three engines (Perplexity sonar, Google AI Mode and OpenAI), the brands were named in 62 of 84 US answers when the question contained the brand name (73.8%, interval 63.5-82.0%), and in 20 of 274 when it did not (7.3%, interval 4.8-11.0%). Asking "what is [your brand]?" mostly tells you the engine can read your name back. It does not tell you whether a buyer who has never heard of you will find you.

The second lesson was about counting. Leadee's raw extractor first counted 12 "AdminKit" mentions in that round. Most of them were adminkit.io, an unrelated Bootstrap dashboard template, not adminkit.net (part of Netbase JSC). One Google AI Mode answer even added "Not to be confused with AdminKit.io". Counted unambiguously, adminkit.net was named in 0 of 39 unbranded US answers. A naive name count would have reported a brand that was not there.

Write 5-10 questions a real buyer would type before they know you exist: "best [category] for [type of company]", "[market player] alternatives", "how do I [job your product does]". These unbranded questions are the ones that matter for new customers. Add two or three branded questions ("is [your brand] any good?", "[your brand] vs [market player]") as a separate group, and never mix the two groups in one number.

If you are not sure which questions buyers ask, our guide on finding the prompts your customers ask ChatGPT lists where to look, and the free ChatGPT Prompt Finder suggests phrasings for your keyword. Use the buyer's words, not your marketing words. Before the first round, run the free ChatGPT Visibility Checker so you know whether OpenAI's bots can read your site at all.

One answer is one sample. The same question can name different companies, in a different order, with different sources, a minute later. Ask each question at least three times, in a fresh chat each time, so earlier turns do not steer the answer. Note the date, the model or product you used, and the country you asked from.

If you test in the ChatGPT app, sign out or use a temporary chat where you can, so your own history and saved memory do not shape the answer (see the section on the app below). Copy each full answer into your notes before you move on — you will want to re-read it when you count.

For each answer, write down three things. Mention: is your brand named in the text? Citation: does one of the source links point to your own domain? Position: if the answer is a list, where are you in it (1 = first)? Also note which market players were named, so you can see who takes the place you want.

Keep mention and citation apart. ChatGPT can name you while linking a directory page about you, or link your blog post without naming you. What ChatGPT citations are explains the difference in detail, and the ChatGPT prompt tracking sheet gives a ready column layout for all of this.

Before you count, decide what counts. A mention is your full brand name or your domain in a form a reader could not confuse with something else. If another company, product or open-source project shares your name, check every hit by hand: does the description, the link or the context match your business?

This is where the AdminKit numbers above went wrong at first: 12 raw hits, almost all for a different product. If ChatGPT describes the other company under your name, that is a separate problem with its own fix, covered in how to fix wrong information about your company in ChatGPT. For counting, the rule is simple: when in doubt, it is not a mention.

Divide unambiguous mentions by answers, separately for unbranded and branded questions: for example 3 mentions in 25 answers is 12%. Then add a confidence interval, because small samples swing a lot. The Wilson interval is the usual method for this; for 3 out of 25 it runs from about 4% to 30%. In plain words: your true rate could be a little above zero or close to a third.

That width is the reason a single check proves little in either direction. If a later round's interval overlaps the first one, you cannot say anything changed. Leadee reports every rate this way; the method is on our methodology page.

Re-run the same questions every week or two, on the same day, with the same setup. Change the question list only on purpose and write down when you did. Over a month the runs add up, the interval narrows, and you can start to see whether a fix to your site or a new listing on a third-party page moved anything.

Doing this by hand takes roughly an hour a round for 10 questions asked three times each. When that stops being practical, automated tracking does the same through OpenAI's API on a schedule. The other ChatGPT guides and tools are on the ChatGPT SEO overview.

Leadee measures through OpenAI's official API only. OpenAI says its Chat Completions search model gives developers access to the fine-tuned models and tool used by Search in ChatGPT, so the API is the closest repeatable view of ChatGPT search. It is still not identical to what a person sees in the app.

The app can add things an API call does not have: the user's saved memory and chat history, personalisation settings, the model or mode selected, whether a given question triggers a web search at all, and the user's location. Any of these can change which companies get named. We have not measured how large that difference is, so we do not put a number on it. Use the API (or a tool built on it) for the trend, and treat what you see in your own signed-in app as one more sample, not the truth.

Have a second person re-count 10 answers from the notes without seeing your tally. If you disagree on more than one, your rule for "unambiguous" is not clear enough yet. Check that every counted citation actually points to your domain, not to a reseller, a directory profile or a namesake.

Then compare your unbranded rate with your branded rate. A brand that is named in most branded answers and almost none of the unbranded ones is in the same place most of our own brands were in October 2026: known to the engine, but not recommended to new buyers.