How to Measure Your Brand’s Visibility in ChatGPT

To measure your brand's visibility in ChatGPT, run a fixed set of unbranded buyer questions and score every answer three ways: whether it names you, whether it recommends you, and whether it cites your site. Build the prompt set first and with care, because every number you report depends on which questions you ask. Keep the three scores separate, because they diverge. In a 400-prompt panel we track for a B2B software vendor, ChatGPT named the brand in 29% of answers and said something evaluative about it in 15%.

Why does the prompt set decide every metric?

Because the prompts are the denominator of every rate. Change the questions and the mention rate, the recommendation rate and the share of voice all change with them, for the same brand in the same week.

Our 400-prompt panel shows how far the number can move. ChatGPT evaluated the brand for 60 of the prompts and left it out of the other 340. A 20-prompt set drawn from the first group would report 100%, and one drawn from the second would report zero. The wording alone shifted the result.

The yes-or-no questions are the clearest case. ChatGPT named any of the ten tracked brands in only 1 of those 40 answers, so they measured almost nothing. Longer, more specific prompts also behaved differently: the brand was evaluated in 25.9% of prompts of 16 to 20 words and 11.0% of prompts of 11 to 15 words. These are patterns in one panel. They show that prompt design is a measurement decision.

How do you build a prompt set you can trust?

Write the set from how buyers talk, apply the same rules to every prompt, then freeze it.

Take the phrasing from sales calls, support tickets, reviews and forum threads. "Best project management software" is a marketer's prompt. "We're 12 people and our current tool is getting expensive, what else is there" is what a buyer types, and it returns a different set of answers. Spread the set across intent: category questions, situations with a persona and a trigger, criteria such as price or ease, and market qualifiers such as a city or a tax rule.

Map each prompt to a buying stage so you know what a win is worth.

Concentrate the scored set on the comparison stage. Keep brand-stage questions in a separate group: they show whether ChatGPT describes you accurately, and they should never count toward visibility.

Treat prompt volume as directional. Search volume is measured across an industry, and prompt volume is modelled. Tools such as Profound and Ahrefs estimate it, which helps you rank candidates, but no one can tell you exactly how often a prompt is asked. Check candidates against keyword and AI search volume where it exists, and keep a prompt without data only if a reviewer judges it a real buyer question.

Then freeze the set. Once you change the prompts you can no longer compare this month with last month. Put new questions into a second set with its own baseline.

What should you measure in ChatGPT?

Measure three things per answer and report them as separate rates.

Count each answer once per score, even if your name appears three times. Use completed answers as the denominator and log errors separately. A completed answer with no mention is a zero and stays in the count.

How do you run the test so it's repeatable?

Hold the conditions fixed and record them with every run.

ChatGPT can rewrite search queries using memory and can use approximate location to shape local results. See OpenAI's web search guide and Temporary Chat guidance.

Why run each prompt more than once?

Because the same prompt returns different sources on different runs. In AirOps research, only 2.3% of page and query combinations were cited in all three ChatGPT runs (AirOps, The Fan-Out Effect, April 2026, vendor research). Run each question three times and count how many runs include you. One run tells you what happened once.

What does a real result look like?

We track a panel of 400 unbranded buyer prompts for a B2B software vendor and nine of its competitors in the US. Each prompt runs once on six AI engines, which gives 2,400 answers. None of the prompts names a brand. The data comes from Ahrefs Brand Radar and its sentiment export, autumn 2026. The client and its category are withheld.

"Evaluating" means the answer made a statement about the brand that carries sentiment, such as a strength, a weakness or a fit. The mention count comes from the tool's latest snapshot and the evaluative data from an export about a week earlier.

The brand was named in about twice as many answers as it was evaluated in. A mention rate alone would have overstated how often ChatGPT gives buyers a reason to choose it.

Is ChatGPT visibility the same as visibility in other AI engines?

No. The same 400 prompts produced different results on every engine.

The engines also disagree about which questions the brand belongs in. At least one engine evaluated the brand for 146 of the 400 prompts. All six did so for only 2. For 86 prompts, another engine evaluated the brand while ChatGPT left it out. A ChatGPT score tells you about ChatGPT.

How do you read the results?

Read them by question and by topic. Headline rates hide where the problem is.

In the panel, 94% of the 554 evaluative statements about the brand were positive. The negatives were concentrated. All 17 statements about quality and reliability were negative, and 8 of 18 about price and value were. All 220 statements about features were positive. One engine, Copilot, produced 24 of the 34 negative statements.

That points to specific work: publish evidence on reliability and pricing, and check what Copilot's sources say. For the prompts where competitors appear and you don't, open the pages behind the strongest competing answers and compare them with yours.

Alongside the rates, report how many distinct prompts produced at least one recommendation. Repeated hits on one question can conceal weak coverage across the rest.

What should you do after measuring?

Turn the cited sources into a work list. A visibility score tells you where you stand, and the pages the AI engines cite tell you what to do next. In our process we export every cited URL from the panel, crawl each page, and record four things: what type of page it is, whether it mentions the brand, whether it links to the brand, and which competitors it links to. We add a domain authority measure so targets can be ranked by more than citation count.

The page type decides how you approach each target.

A mention without a link and a link from a page the engines already cite are the cheapest wins, so start there. For the full method, see how to mine LLM citation results and how to classify cited URLs.

How does this compare with Business Hub insights?

If you have access to Business Hub, compare its aggregate insights with your prompt panel. OpenAI counts an online-business mention there when ChatGPT references the business's website. The brand-name mentions in this guide use a different definition, so the two numbers will not match. See OpenAI's Business Hub insights guide.

Run the same panel again after each change and compare counts by question. Site visits and conversions measure what happened after a click and belong in a separate report.

Frequently asked questions

How many prompts do you need to measure ChatGPT visibility?

Ten well-chosen prompts run three times each is enough to start by hand. A tracked panel can be much larger. Ours uses 400 prompts, which is what made the per-topic and per-engine patterns visible. Quality matters more than count: 40 of our 400 were yes-or-no questions that measured almost nothing.

How often should you re-measure?

Re-run the same panel monthly and after any significant content change. Answers and cited sources shift between runs, so compare like with like and keep the prompt wording fixed.

Is a mention the same as a recommendation?

No. In our panel ChatGPT named the brand in 29% of answers and made an evaluative statement about it in 15%. Score the two separately.

For context on the paid and organic surfaces, see how to reach customers in ChatGPT. See what a full visibility report looks like or get an audit for your own buyer questions.