GUIDE · PART OF AI VISIBILITY TRACKING
AI mention tracking:
your brand, named in the answer.
A mention is the simplest thing you can measure in AI search, and everything else rests on it. Either the answer names you or it doesn't. This guide covers how to count mentions properly, why a single check isn't enough, and what a mention rate can and can't tell you.
What is AI mention tracking?
AI mention tracking is counting how often AI answer engines name a brand when they're asked questions relevant to it. Each tracked question is run on each engine several times, and every answer is checked for the brand's name. The result is a mention rate: the share of answers that include the brand. It is separate from citation tracking, which looks at the sources an answer drew on.
Key facts
- A mention is your brand named in the text of an answer. A citation is one of your pages used as a source
- Mention rate is the number of answers that name you, divided by the number of answers tracked
- Engines answer differently from one run to the next, so a rate needs repeated runs behind it
- Questions that don't name your brand show whether you make the shortlist. Questions that do show how you're described
- A mention rate describes the prompts you chose. Change the prompts and it changes
Last reviewed 15 October 2026 by the SEORCE team.
Why a single check tells you very little.
Ask an assistant about your category once and you'll get an answer. Ask again and you may get a different one, with different brands in it. That is the first of three things that make a simple count less simple.
The same question, a different reply
An engine writes each answer afresh. You might be named in seven runs out of ten, which a single check would report as either "yes" or "no".
Your brand, your product, a misspelling
Engines refer to companies in more than one way. If you only look for the exact brand name, you'll undercount.
Easy questions flatter you
Track only the questions you already win and the rate looks excellent. It tells you nothing about the ones you're losing.
What you're counting, and how to slice it.
Three decisions shape a mention rate. Make them once, write them down, and keep them the same from month to month.
Decide what a mention is.
It sounds too obvious to need deciding. Then you find an answer that names your product and not your company, and you have to choose.
- The brand in the textNamed in the answer itself, with or without a link.
- Products and variantsList the names you'd accept: the company, each product, common misspellings.
- A link alone isn't a mentionIf your page is used as a source and your name isn't in the answer, that's a citation.
For agencies billing abroad, Quillbook is a common pick. Ledgerly also invoices in several currencies and suits smaller teams.
Count unbranded and branded questions apart.
The two kinds of question measure different things, and adding them together hides both.
- Unbranded questions"Best invoicing software for agencies." Being named here means you made the shortlist.
- Branded questions"Is Ledgerly any good?" You'll nearly always be named. What matters is what gets said.
- By stageLearning, comparing, choosing. A brand can be strong at one and missing at another.
Repeat it, on every engine.
A rate built on one run per prompt is a coin toss dressed up as a statistic. The repetition is the whole point.
- Several runs for each promptSo the rate reflects the pattern.
- Every engine counted separatelyAn average across engines hides the one where you're missing.
- The same schedule each timeSo this month can be compared with last.
Counting mentions, start to finish.
You can do this in a spreadsheet for a handful of prompts. The method doesn't change when software takes over the repetition.
Choose the prompts
Mostly unbranded questions across the three stages, plus a few branded ones.
Prompt coverage →Run them, more than once
On each engine, on a schedule. Keep every answer.
Count
Answers that name you, divided by answers tracked. Do it for each engine and each stage.
Read the gaps
Find the questions and engines where you're missing. Those are where the work is.
AI visibility diagnosis →Check the names a machine can read.
Mention tracking depends on engines recognising your brand. The entity analyser shows whether Wikidata and Wikipedia have a record of it.
What a mention rate is telling you.
| What you see | What it suggests | Where to go next |
|---|---|---|
| High on branded questions, low on unbranded | Engines know who you are and don't put you on the shortlist. | Look at the sources they use for category questions. |
| High on one engine, low on the others | That engine reads a source the others don't. | Compare the citations engine by engine. |
| High at the learning stage, low at choosing | You explain the category well and aren't recommended. | Check which brands are named at the choosing stage, and why. |
| The rate swings from week to week | Too few prompts or too few runs. | Add runs before you read anything into the change. |
| Named often, described wrongly | The mention is there and the facts aren't. | Treat it as a brand monitoring problem. |
Every run counted, by prompt and by engine.
AI Beacon runs your prompts on 7 AI engines + Google AI Overview, AI Mode & Search, and records whether you were named each time. You see the result for each prompt and each engine, and a Visibility Score for the project built from those runs.
Frequently asked questions.
An AI mention is your brand named in the text of an answer from an AI engine. It doesn't need a link. If your page is used as a source and your name doesn't appear, that is a citation, which is counted separately.
Divide the number of answers that name your brand by the number of answers tracked. With 40 prompts on 5 engines, run 10 times each, you have 2,000 answers. If 820 name you, the mention rate is 41%.
There isn't a universal figure, because the rate depends on the prompts you chose. Two comparisons are meaningful: your own rate over time, and your rate against competitors on the same prompts.
Enough for the rate to stop jumping about. If the number changes a lot between two weeks with no change on your side, add runs or prompts before reading anything into it.
Only when the answer links to you and someone clicks. Plenty of mentions carry no link at all. Their effect shows up later, as direct visits and searches for your name.
Yes, for a few prompts. Ask each engine, note whether you're named, and repeat. It stops being practical once you want several runs of dozens of prompts on every engine.
Not automatically. You can ask each engine by hand. The free entity SEO analyser covers the groundwork: whether the public knowledge bases know who you are.
See where you're named, and where you aren't.
Add your site and a few buyer questions. Each one is run on every engine, and you'll see the answers that left you out.
Not ready for an account? Use the free browser extension or any of the free tools.