How to measure brand mentions in AI answers

To measure brand mentions in AI answers, ask the questions your users ask, in the assistants they use, from the countries they live in, and record whether the first answer names your brand before you do. Repeat that every day and track the share over time. Everything else is detail around that one number.

The three metrics that matter

Metric How to calculate What it tells you
Named unprompted rate Chats whose first answer names the brand ÷ analyzed chats How often users hear about you without asking
Average position Mean rank of the brand among brands in the first answer Whether you are a top pick or an afterthought
Agreement rate Chats where the assistant supports the brand by the end ÷ analyzed chats How the assistant reacts when a user brings you up

Named unprompted rate

Only the first answer counts, and only if the brand appears before the user mentions it. If your opening question names the brand, the measurement is worthless. Write questions the way users ask them: "best accounting software for freelancers in Germany", not "is Ledgerly good?".

Average position

Assistants usually list three to seven options. Being first matters: many readers stop at the top recommendation. Record the position each time the brand is named, and average it over the period.

Agreement rate

When a user says "I'm leaning towards your brand", does the assistant confirm it fits, add caveats or steer towards a competitor? This shows how the model describes you once you are in the conversation.

How many chats you need

Answers vary between runs, so small samples mislead. As a rule of thumb:

  • 20 or more analyzed chats per assistant before comparing assistants.
  • A 7-day rolling rate for trends, so one odd day doesn't dominate.
  • The same questions over time, so changes reflect the assistant and not your sample.

Break it down

An overall rate hides the useful detail. Split the numbers:

  • By assistant: ChatGPT, Gemini, Perplexity and Copilot often disagree.
  • By country: local competitors and sources differ by market.
  • By question: you may lead "best X for agencies" and be absent from "cheapest X".

Keep the transcripts

Numbers say what happened; transcripts say why. Read the first answers where you are missing: which competitors were named, and how were they described? Those descriptions show the language and sources the assistant relies on.

GEO Boost stores every conversation in full and analyzes each one for these three metrics, per assistant, country and question. The pricing page shows what a daily campaign costs.

FAQ

Why measure only the first answer?

Because later answers react to what the user said. If the user names your brand, the assistant will talk about it. The first answer is the only one that shows whether the assistant brings you up on its own.

How often should I measure?

Daily, with results read as a 7-day rolling rate. AI answers change with the web and with model updates, so weekly snapshots miss most of the movement.

Do AI assistants give the same answer to everyone?

No. Answers depend on the wording, the language, the country of the connection and the moment. That is why a reliable measurement uses many chats from each target country.

Should I track competitors too?

Yes. Recording which brands the first answer names gives you the competitors the assistant actually recommends, which are not always the ones you compete with in search.

More articles · GEO Boost