How to Track Brand Visibility in ChatGPT

Brand visibility tracking in ChatGPT means regularly recording the answers given to a curated set of customer questions and measuring how your brand appears in those answers. Searching for your own brand once does not tell you how visible you are across your category. That requires defining a prompt set, measurement conditions, and a method for counting the results together.

A good tracking setup separates three outcomes: whether the brand's name appears, whether it is recommended, and whether a source link points to your site. Thanks to this distinction, instead of seeing only a single score, you can understand which content or product information you need to work on.

1. Choose questions that represent the customer's decision

Do not build your initial list only from your brand name. "What is brand X?" measures what the AI knows about you. "Which solution suits a small team?" lets you observe your chance of being discovered within the category.

For an example B2B software brand, four question groups can be used:

Question group	Example	What you want to measure
Need	How can small teams organize customer requests?	Visibility in the problem space
Choice	Which support software suits a five-person team?	Making it onto the shortlist
Comparison	How does email-based support compare with live chat?	Presence in the decision criteria
Brand knowledge	What is included in product X's starter plan?	Accuracy of the product description

These questions are designed to demonstrate the method. Build your real list from sales conversations, support tickets, and customer research. Instead of twenty questions that closely resemble each other, a smaller set covering different decisions may be more readable.

2. Fix your measurement conditions

Asking the same question on a different date does not by itself produce comparable data. Language, target country, prior conversation context, and the product surface used can all affect the result. Start with a fresh conversation whenever possible and record your measurement conditions.

Each observation should include these fields: question ID, question text, language, target country, platform or model, date, answer text, source URLs, and result status. Do not label an answer obtained via the API as the same record type as an answer seen in the user interface.

If you use a tool, examine how it distinguishes newly collected results from results served out of the cache. An answer collected yesterday being shown again today is not a second independent observation. Likewise, failing to produce an answer because of a technical error should not be counted as "the brand did not appear."

3. Mark mentions, recommendations, and citations separately

Mention: The brand's name appears in the answer. Recommendation: The answer presents the brand as a suitable option for the relevant need. Citation: The answer contains a source link to a specific web page.

An AI answer may recommend your brand but cite your competitor's site. It may cite your site while not counting your brand among the product options. The work to be done in these two cases is not the same.

In a hypothetical monthly measurement, out of 200 fresh and valid answers, suppose the brand name appears in 50, the brand is recommended in 30, and a source link to the brand's domain is given in 20. The mention rate is 25%, the recommendation rate across all answers is 15%, and the own-site citation share is 10%. Do not count five links in the same answer as five separate brand appearances.

Add the definitions to your report. In particular, the "recommendation rate" may use all answers as the denominator in some reports and only the answers where the brand was mentioned in others. Comparing different calculations under the same name is misleading.

4. Turn the result into a content decision

Make the last column of your tracking table "next action":

Observation	Next check
Competitor recommended, brand not visible	Do you have a page that answers the question's need?
Brand present, product information wrong	Are the pricing, product, and help pages consistent with each other?
Third-party source gives outdated information about the brand	Is the source current; is there a clear verification page on your own site?
A blog is cited but no product interest forms	Does the article address the right reader; is the next step clear?

Read the source page. Do not open a new article based only on a keyword that appears in the URL. Adding an example, an explanation, or fresh data to an existing page that meets the same need may be enough.

5. Evaluate change with a fixed scope

After a content update, track the same prompt set and the same platform scope. If you added new questions, report the old set's result separately. This lets you separate the change in visibility from the change in scope.

Do not declare a single positive answer a success. Check whether the improvement spreads across different days and different relevant questions. Also keep the increase in citations separate from site visits, product evaluation, and sales outcomes; being linked is not a guarantee of a click or a sale.

For more detailed calculations, review the seven AI visibility metrics. When checking source access, also keep in mind that OpenAI defines its search and training bots separately; allowing a bot does not replace visibility measurement.

By reviewing Maya's approach to AI visibility tracking, you can evaluate how to connect your question, answer, and source data to your own content decisions.