Glossary · Beginner
What Is Mention Rate? Definition and Example
Mention rate is the percentage of AI answers, across a defined prompt set, that name your brand at all. It's the broadest GEO visibility metric: it tells you whether engines think of you when answering questions in your category, before asking whether they recommend or cite you.
How do you calculate it?
Mention rate = answers mentioning your brand ÷ total answers scored
Illustrative: 40 prompts × 2 engines × 2 runs = 160 answers. You're named in 48, so your mention rate is 30%.
What counts as a mention?
- Your brand or product name, including agreed variants and common misspellings.
- Not counted: generic category words that happen to match your name, unless clearly about you.
Write the matching rules down, especially if your name is a common word.
How should you segment it?
| Cut | Why |
|---|---|
| Engine | Visibility often differs a lot between engines |
| Intent | Learn, compare and buy prompts behave differently |
| Market/language | Answers vary by location |
| Branded vs unbranded | Branded prompts inflate the rate |
Example
Illustrative. A brand's overall mention rate is 30%, but 70% on learn prompts and 8% on buy prompts. Engines know it as an educational source rather than a vendor. The work shifts to comparison pages, reviews and clear product positioning.
Common pitfalls
- Comparing your rate with a tool's benchmark built on different prompts.
- Changing prompts between periods, which breaks the trend.
- Judging a change from a single run; answers vary.
Why it matters for GEO
Mention rate is the starting signal. Read it alongside recommendation rate and citation share to see what kind of visibility you have.
How often should you measure it?
Monthly is enough for most teams, with the same prompts, engines, settings and number of runs each time. Weekly checks tend to show noise more than change, because answers vary from run to run.
Related
Questions
Does a negative mention count?
For mention rate, yes; it's a count of being named. Track sentiment or accuracy separately.
How many prompts do I need?
Enough to be stable. Many teams start with 30-50 prompts and several runs each.
Should I include brand-name prompts?
Report them separately, as they almost always mention you.