AI visibility, measured
How often does AI name your brand?
An AI visibility tool asks the assistants the questions your buyers ask, records who gets named, and turns that into a number you can track. This one runs 16–18 questions across several models, repeats the commercial ones 3 times, and shows you the raw answers — not just a score.
What an AI visibility tool actually does
It answers one question: when a buyer asks an AI assistant about your category, does your company come up? The mechanics are the same everywhere:
- Build the questions. Not your brand name — the questions a buyer asks before they know you exist: “best X in London”, “who does Y for mid-size firms”.
- Ask the models. Through official APIs, in two modes: from model memory, and with web search enabled. The two give different answers, and both matter.
- Read the answers. Was the brand mentioned? In what role — recommended, listed, or mentioned in passing? With what sentiment? Who was named instead?
- Repeat. Model answers are not deterministic. A single run tells you almost nothing; the spread across runs is the finding.
- Score and track. One number to watch over time, with the interval that says how much of it is noise.
What makes this one different
It shows the working
Every number on the report is traceable to the answers underneath it. The formula is published on the methodology page, and it is the same formula the code runs — if the page and the calculation ever disagree, that is a bug we fix.
It covers models others skip
Alisa AI and GigaChat are not in any Western tracker, and for anyone selling into Russian-speaking markets that is the majority of the audience. More on that gap.
It refuses to invent numbers
Share of voice is not shown when there are fewer than three non-branded questions to count it on — a figure calculated on two answers would flatter you, and a flattering wrong number is worse than a missing one.
It costs nothing to try
No sign-up, no card, no sales call. You get a link to the report and you keep it. If you want the numbers walked through, that is a separate conversation you can decline.
What you get back
- Visibility index, 0–100 — five parts with published weights, not a black box.
- Mention rate with a confidence interval — Wilson, 95%. The interval is the honest part: it tells you when a change is real and when it is noise.
- The gap table — the exact questions where an AI named a competitor and not you, with the competitor named.
- Cited sources — which sites the models link to when answering in your field. The rows where competitors appear and you do not are the shortest route to being cited yourself.
- Every raw answer — full text, model, mode, sentiment, confidence.
Honest limits
Language-model answers are probabilistic and may be inaccurate, incomplete, outdated or invented; we do not verify them. This report is not a ranking, an assessment of quality, a recommendation or advertising. Consumer apps can answer differently from the APIs — their output depends on the account, the chat history, the region and the version. And a score measured today is a measurement, not a forecast.
Check your brand now
Free, no sign-up, with the formula and the intervals in the open.
Run a free checkCommon questions
- How long does a check take?
- Minutes, not days — you get a link to the report and it fills in as the models answer.
- Which models are asked?
- ChatGPT, DeepSeek, Alisa AI and GigaChat, in memory mode and with web search where the model supports it. The mode is printed next to every answer.
- Do I need to name competitors?
- No, but it helps: naming them adds two comparison questions and makes share of voice meaningful.
- Can I compare my score with another company's?
- Carefully. The share of commercial queries that receive an AI answer differs by category, so the safer comparison is your own brand over time, and against the competitors you asked us to track.