How to Track AI Mentions of Your Brand (Without Guessing)
Your brand is being discussed in places you can’t see. Every day, buyers ask ChatGPT, Perplexity, Gemini, and Google’s AI Overviews for recommendations in your category — and those answers name specific companies. If you don’t know whether you’re one of them, you’re flying blind. Here’s how to actually track AI mentions of your brand.
Why traditional rank tracking doesn’t work here
Rank trackers answer one question: what position does my page hold for a keyword? AI answer engines don’t produce positions. They produce a paragraph that either mentions you or doesn’t. There’s no page two to climb out of — either the model names your brand as an option, or the buyer never learns you exist.
That changes the unit of measurement. Instead of “keyword → position,” you need prompt → mention. And because AI answers vary between engines, sessions, and phrasings, one check tells you nothing. You need repeated sampling over time.
The four things worth measuring
| Metric | What it tells you |
|---|---|
| Mention rate | Of the buyer prompts in your category, what share of answers name you at all? |
| Recommendation share | When competitors are also named, how much of the “airtime” is yours — first mention, or a footnote? |
| Citation source | Which pages the engine links or leans on when it describes you (often not your homepage). |
| Accuracy | Whether what the model says about you — services, location, pricing, credentials — is actually correct. |
That last one surprises people. It’s common to discover an engine confidently describing services you dropped two years ago, or listing an old address. Fixing factual drift is often the fastest visibility win available.
Step 1: build a real buyer prompt set
Write 20–40 prompts the way a customer would actually type them, not the way a marketer writes keywords. Mix four kinds:
- Category discovery — “best [category] in [city]”
- Problem-first — “who can help me with [the problem you solve]?”
- Comparison — “[competitor] vs alternatives”
- Branded — “is [your brand] any good?” and “what does [your brand] do?”
Branded prompts test accuracy. Unbranded prompts test whether you get discovered at all. Keep the set fixed — changing prompts every month destroys your ability to see trends.
Step 2: sample repeatedly, across engines
Run the same prompt set against each major engine on a consistent schedule — weekly is a good baseline. Ask each prompt in a fresh, logged-out session so personalization and chat memory don’t contaminate results. Then log, for every run: the engine, the prompt, whether you were named, which competitors were named, and any sources cited.
One run is anecdote. Twelve weeks of runs is a trend line you can act on — and it’s the only way to prove whether the AEO work you did in March moved anything by June.
Step 3: connect movement to causes
Tracking is only useful if it changes what you do next. When your mention rate rises or falls, look at what the engines are actually citing: review platforms, directories, industry roundups, your own explainer pages. Those citation sources are your leverage points. If a “best in city” roundup drives half your mentions, getting listed accurately in three more roundups is a higher-yield move than any redesign.
Pair your tracking with the structural fundamentals: clear entity information, answer-shaped content, and machine-readable markup. See what AEO is for the concept, how to do AEO for the playbook, schema markup for AI for the technical layer, and the AEO checklist to audit what’s missing.
Four mistakes that make tracking useless
- Checking once. AI answers fluctuate. A single miss isn’t a verdict, and a single hit isn’t a win.
- Testing while logged in. Your own chat history skews the model toward naming you.
- Only tracking branded prompts. Of course the engine knows you when you name yourself. Discovery happens on unbranded prompts.
- Ignoring competitors. Mention rate without competitive share hides whether you’re gaining or the whole category is.
A cadence that actually gets done
Weekly: run the prompt set, log mentions and competitors. Monthly: review the trend, fix any factual errors the engines repeat, and publish one page that answers a prompt you keep losing. Quarterly: revisit the prompt set for new buyer language and check which citation sources are gaining influence.
If you want to see where you stand before you build any of this, run a free AI Visibility Scan — it asks the major engines real buyer questions in your category and shows you who they recommend today, and where you land. From there, the AI Visibility Tracker keeps the same prompt set running every week so you can watch the line move instead of guessing.