Practical guide

How to track brand mentions in ChatGPT

A defensible tracker measures a declared panel of buyer questions, repeats the observations and keeps every mention traceable to the answer that produced it.

3 min read

Brand illustration for tracking ChatGPT mentions, with abstract interface shapes and no text.

Key takeaways

  • Build the panel around buyer problems, categories, comparisons and purchase decisions.
  • Record the interface, model when visible, market, language, account state, date and whether web search was active.
  • Repeat important questions so one unusual answer does not become the result.
  • Separate a neutral mention from a recommendation, warning, comparison, citation or navigational reference.
  • A mention rate needs a published denominator. It is not preference and it is not total market share.
Contents

Start with buyer questions

Build a representative panel around problems, categories, comparisons and purchase decisions, not prompts chosen because they already mention your brand.

  1. Record the ChatGPT interface, model when visible, market, language, account state, date and whether web search was active.
  2. Repeat important questions under comparable conditions so one unusual answer does not become the result.
  3. Separate a neutral mention from a qualified recommendation, warning, comparison, direct citation or navigational reference.
  4. Store the complete question, answer, visible sources and collection context behind every aggregate metric.
  5. Track rates over time and against a declared competitor set. Do not present the sample as total market share.
Five steps for tracking ChatGPT mentions: buyer questions, recorded context, repeats, mention types and stored answers.
Every aggregate number should trace back to the question, the answer and the collection context.

A mention is not one flag

Define the rule before collection. A common definition is an explicit brand or product-name appearance, recorded separately from recommendations, citations and navigational links. Recommendation wording matters more than a binary flag alone. A visible citation does not prove why the source was selected.

A tree separating a neutral mention, a recommendation and a citation.
Keep the surrounding wording. A mention can be neutral, a recommendation can be qualified, and a citation is a visible source.

How to state a mention rate

Divide answers containing a qualifying mention by all eligible collected answers in the declared sample. Publish the denominator and explain how errors, refusals or missing answers are handled. A change between two periods does not by itself prove causality.

Four cards on mention rate limits: the denominator, preference, market share and causality.
A mention rate describes the declared sample. It does not prove preference, total market share or causality.

Frequently asked questions

Can I search ChatGPT once to check my brand visibility?

One search can provide an example, but not a reliable baseline. Generated answers vary, so important questions need repeated observations under documented conditions.

What counts as a mention?

Define the rule before collection. A common definition is an explicit brand or product-name appearance, recorded separately from recommendations, citations and navigational links.

How is a mention rate calculated?

Divide answers containing a qualifying mention by all eligible collected answers in the declared sample. Publish the denominator and explain how errors, refusals or missing answers are handled.

Does a mention prove that ChatGPT prefers my brand?

No. The surrounding context may be neutral, negative or conditional. Recommendation context and evidence should be reviewed separately.

Sources

This guide is a ViziGEO working method for ChatGPT observations. It does not claim access to private model rankings. Related measurement rules are in the methodology.

  1. ViziGEO measurement methodology
  2. OpenAI: reproducible outputs with the seed parameter

Start a free AI visibility audit

Open ViziGEO and review a documented sample of questions, mentions, recommendations and citations. The audit records what was observed. It does not guarantee a mention, a citation or a recommendation.