All posts
Explainer

AI mention rate, explained

The base metric of AI visibility: what share of AI answers actually name your brand. How the number is worked out, how it maps onto SEO, whether yours is any good, and what to do when it’s low.

MT
The Mention Tracker team
Explainer · Aug 9, 2026 · 6 min read
AI Mention Analytics
51% ▲ 4 pts
ChatGPT Claude Gemini Perplexity Grok
Share of AI answers that mention your brand · Last 30 days
The AI Mention Analytics card on a live dashboard — here for the demo brand, Northwind Coffee. Every number quoted below is the one in this picture.

Search optimisation had one question at its centre: where do you rank? AI answers don’t have a ranking. They hand back a shortlist of two or three names, and you are either on it or you aren’t.

AI mention rate is the measure of that: the share of AI answers about your market that actually name your brand. It is the base metric of AI optimisation — placement, sentiment and share of voice all assume you were named in the first place.

How is AI mention rate calculated?

Every run puts your 15 tracked prompts to five models — ChatGPT, Claude, Gemini, Perplexity and Grok — and counts the answers that named you. No weighting, no proprietary score, no credit for prominence: a blunt number is one you can argue with.

You don’t write the prompts. You describe the business once — what you sell, where, who to — and the tracker drafts the fifteen questions your buyers ask, which you edit before anything runs.

The card above, worked through
15 prompts × 5 models = 75 answers in the Aug 3 run.
38 of them named Northwind Coffee.
38 ÷ 75 = 51%.
That 51% is all five models pooled, so none of the lines on the chart is the 51. We run 17 prompts in all — the two that name your brand inside the question stay out of the denominator, because they can’t test whether AI would name you unprompted.

Two rules decide what counts as a mention:

  • One answer, one mention. Named five times in a single answer still counts once — this measures presence, not volume.
  • Matching is fuzzy, but anchored. Name variants and your domain count. Where the brand name is also an ordinary word — Monday, Notion, Assembly — it only counts when the surrounding text ties it to you.

AI mention rate vs. SEO rank tracking: what maps across?

If you’ve run an SEO programme, most of the instincts carry over:

In SEO
In AI answers
Keywords
Prompts
Impressions
Mention rate
Position 1–10
Placement rank
Backlinks
Citations
One results page
Five models, five answers

The last row is where the analogy breaks, and it is why the card plots five lines: a search engine gave you one result page to work on, AI visibility gives you five that disagree — and predictably so. On Aug 3 above, Gemini named Northwind in 67% of its answers and Grok in 20%: same brand, same fifteen questions, same day. The models don’t read the same internet — some search the live web and cite as they answer, others reply from training data — so a lagging line is usually a sourcing problem with a name, not noise.

The quieter break is the row above — rankings reward your page, AI answers reward the reference. Models name brands they find corroborated in sources they already trust, very rarely your own homepage. Call it AIO, GEO or AEO; the job is the same: earn the citations, then measure whether the models started naming you.

Is my AI mention rate good?

There is no universal benchmark, and anyone quoting you one is guessing. The number depends almost entirely on how broad your prompt set is — fifteen narrow questions about your speciality score high, fifteen about the whole category score low. Same brand, same day, different number. So a mention rate is only meaningful against itself, and two comparisons carry real information:

  • You, last month. Same prompts, same models — so a move is a change in what AI says, not in what you asked.
  • Your competitors, on the same prompts. The only cross-brand comparison that controls for how hard the questions are.

And read the trend, not the pill. Across 75 answers one answer is worth 1.3 points, so the “▲ 4 pts” above is three answers changing their minds. These models aren’t deterministic — ask twice, get two slightly different shortlists. Three runs moving the same way is a signal; one run is weather.

How to read the chart

  • The y-axis fits your data. It isn’t pinned to 0–100 — the card above runs 5% to 85% — so read the labels before the slope.
  • A silent model gets a gap, not a zero. “You weren’t named” and “we never got to ask” are different facts, and the chart refuses to merge them.
  • 0% is a reading, not an error. It is the most useful state the card has — it names the exact prompts to go and win.

How do I get my brand mentioned in ChatGPT?

Mention rate follows citations, and citations are earned off your own site. That is the whole method:

  1. Work one losing prompt at a time. An average over fifteen questions isn’t something you can act on. A prompt you lose is. Take the worst one and treat it as the whole job.
  2. List what the winning answer cites instead of you. Not who it names — what it links: a subreddit thread, a roundup on a trade site, three review profiles, someone’s comparison post.
  3. Go and earn a place in those exact sources. Get listed where the directory is thin, reviewed where the reviews are, and answer the question publicly where the thread is. There is no shortcut that skips this.
  4. Make your own pages answerable meanwhile. Plain language, self-contained answers, specifics on the page rather than behind a form or a script. A page a model can quote is a page it can cite.
  5. Re-measure, and give it time. Sources get published, then crawled, then absorbed — weeks, not days. Which is why this is tracked rather than checked.

None of that needs our software. What the tracker adds is which prompt to start on, what the winners cited, and whether any of it worked — the part you can’t do by asking ChatGPT about yourself a few times. The longer version of this method, including the tactics that don’t work, is in how to show up when someone asks ChatGPT for a recommendation.

Common questions

What is AI mention rate?

The share of AI answers that name your brand. Mention Tracker puts 15 prompts to five models each run, counts the answers that named you, and divides: 38 of 75 is 51%.

What’s a good AI mention rate?

There is no universal benchmark, and anyone quoting you one is guessing — the number depends almost entirely on how broad your prompt set is. Only two comparisons mean anything: you last month, and your competitors on the same prompts.

Why do ChatGPT and Gemini give different answers about my brand?

Because they don’t read the same internet. Some models search the live web and cite as they answer; others reply from training data. In one run of the demo tracker Gemini named the brand in 67% of its answers and Grok in 20% — same brand, same fifteen questions, same day. That is why the chart plots one line per model rather than a single blended score: the blend is the headline, but the gaps between the lines are what you can act on.

Is mention rate the same as share of voice?

No. Mention rate asks whether you were named at all; share of voice compares how often against your competitors on the same prompts. You can hold a healthy mention rate and still be losing share of voice.

How often is it measured?

Weekly. The tracker re-runs your whole prompt set across every model, and the card gains one point per sweep — which is why movement reads run over run, not day to day.

Is AIO the same as GEO?

Two names for one job. AI optimisation (AIO), generative engine optimisation (GEO) and answer engine optimisation (AEO) all describe getting your brand named and cited inside AI answers, rather than ranked on a page of links.

Find out what the five models say about you.

Fifteen prompts, five models, tracked every week — with the losing prompts listed worst first, and the sources the winners cited.

Track my brand See the live demo Pricing →