AI brand mention tracking: what you can measure

AI brand mention tracking: four signals you can verify

An AI mention is not one event. It can be a citation, a named recommendation or a comparison, and each leaves a different trace. The console measures the traces that leave a record — and says which unit each number is counted in.

How to track brand mentions in AI search

1. Step 1

   Pick the questions, not the keywords

   Choose queries where an AI answer would plausibly name you: category terms, comparisons and alternatives. Those are the questions worth sampling; a brand keyword alone mostly confirms you already exist.

2. Step 2

   Sample the AI Overview layer

   Record presence and citation for those tracked queries, and keep the cited URLs. That baseline is what a later “are we being mentioned more?” question is measured against.

3. Step 3

   Record the referrals

   Install the probe or page beacon so AI referral visits land in the console with their landing page. A mention without a visit is still only half the answer.

4. Step 4

   Re-check after every deploy

   A robots rule or header change can silently block a crawler. Re-run the access checks after releases, or schedule them, so a lost mention has a timestamp next to a release.

Can you track brand mentions in ChatGPT?

Not as a per-brand mention log. ChatGPT does not expose one, and scraping conversations would not produce a representative sample. A vendor that sells a ChatGPT mention count is selling a simulation — or a guess — unless it states its prompt set and sample size.

What you can verify are the traces an assistant leaves: chatgpt.com referral visits with their landing pages, branded search demand in Search Console, and whether GPTBot can fetch the pages you want cited. The full answer, including the substitutes worth tracking, is in can you track ChatGPT brand mentions?.

What this page cannot measure

Every AI-visibility vendor draws this line somewhere. Here is where we draw it, so a report cannot quietly imply more than the data supports.

• Mentions inside ChatGPT, Claude or Perplexity conversations. None of them expose a per-brand mention log; any vendor count is a sample, a simulation or a guess.
• Sentiment of an AI answer. A model can describe you positively in one session and omit you in the next; a sentiment score would be a snapshot sold as a trend.
• Share of Model or share-of-voice percentages. Without a stable denominator the number moves when the sample changes, not when your visibility changes.
• A price per mention. An AI mention is not a billable unit; its value shows up later as referrals and branded demand.

A count implies a fixed universe of conversations. Assistants do not expose one, and a sample of prompts is a sample — useful as a trend, misleading as a total. If you are pricing a mention, the companion guide separates a citation from a recommendation and values the referral instead: what an AI brand mention is worth.

What to do instead

1. Confirm access: run the free AEO checker so GPTBot, ClaudeBot and PerplexityBot are not blocked before you measure anything.
2. Record referrals: install the probe or page beacon and read AI referral visits by landing page.
3. Sample the questions: add the comparison and category queries where an AI answer could name you, and keep the cited URL for every citation.