Mention rate

Mention rate is the share of AI answers, across a fixed set of test questions, that correctly mention a target brand. It is the most basic metric for measuring generative engine optimization (GEO).

The formula

mention rate = answers that correctly mention the brand ÷ valid answers

Two words in that formula carry most of the weight.

“Correctly.” A mention only counts if it refers to your company. If the answer names a different company with the same or a similar name, it is not a mention of you, and it should be counted under name confusion rate instead. Skipping this check is the most common way mention rates get inflated.

“Valid.” Some answers should not be in the denominator at all: errors, refusals, empty replies, or answers to a question the platform clearly misread. Decide in advance what counts as a valid answer and write it down, so the denominator does not shift from one round to the next.

A worked example

Say you run 40 questions on one AI platform and collect 40 answers. Two are error messages, so 38 answers are valid. Your brand name appears in 12 of them, but in 3 the assistant is actually describing a different company with the same name.

• Correct mentions: 12 − 3 = 9
• Mention rate: 9 ÷ 38 ≈ 24%
• Name confusion: 3 of the 12 answers that used your name pointed to someone else

Reporting “12 out of 40” would have overstated the result and hidden the confusion problem.

What has to stay fixed

Mention rate is only meaningful when it is comparable over time. Before the first measurement, freeze:

• The question set: exact wording and numbering. Changing the questions changes the ruler.
• The platforms: record each AI assistant separately, such as ChatGPT, Perplexity, or the Chinese assistants Doubao and Tencent Yuanbao. Do not average them into one number; they read different sources.
• Sampling: how many times each question is asked, whether web search is on, and the date.
• Judgment rules: what counts as a correct mention and a valid answer.

Then measure once as a baseline and repeat under the same conditions (see baseline and retest).

How it relates to other metrics

Mention rate answers “does the brand appear at all?” It does not say where the brand appears or how it is described. That is why it is usually reported alongside:

• First-recommendation rate: how often the brand is the first option recommended. A brand can be mentioned in every answer and still never be first.
• Name confusion rate: how often the brand name points to the wrong company.
• Description accuracy: whether what the answer says about the brand is true (see AI hallucination).

Improvements to citation sources usually show up in mention rate first, because being retrieved is a precondition for being recommended.

Breaking it down

A single overall number hides useful detail. It is common to split mention rate by question type: category keywords, decision queries, long-tail questions and branded questions. A brand that is mentioned on long-tail questions but not on category questions has a different problem from one that is missing everywhere.

Limits

A question set is a constructed sample, not a record of real user behavior, so mention rate measures direction of change, not market share. AI answers also vary between runs, so small differences between two rounds may be noise. Treat results as “under these conditions, in this period, we observed this,” not as proof of cause.

To see your current mention rate on a starter question set, request a free GEO audit.