How the AI visibility score is measured
The score is one number between 0 and 100 describing how often AI assistants name your business when asked questions you should be an answer to. This page says what it measures, how you can check it yourself, and where it is weak.
Last updated: October 12, 2026
1.What the score measures
We ask several public AI assistants a set of questions a customer might realistically ask, and we look for two things in each answer: whether your business is mentioned by name, and whether your site is cited as a source. The score combines the two across the assistants we ask, expressed out of 100.
That is the whole idea. It is a measurement of observed behaviour, not a model of it. It does not know why an assistant answered the way it did, and it does not predict what it will answer tomorrow.
What the score is not: a ranking, a guarantee, a prediction of revenue, or anything an AI vendor recognises. Nobody outside Tunnel uses this number. It is useful because it is measured the same way on every run, so its movement means something even where its absolute value does not.
2.Which assistants we ask
Five engines, all of them assistants a customer might actually use, and all of them ones we can genuinely reach:
| Engine | Citation list |
|---|---|
| OpenAI (ChatGPT) | No — mentions only |
| Anthropic (Claude) | No — mentions only |
| Google (Gemini) | No — mentions only |
| xAI (Grok) | No — mentions only |
| Perplexity | Yes — returns source URLs |
One thing about how we ask matters more than which models we ask. We send every question with web search switched on, because that is how people use these assistants: you ask, and the product looks things up before it answers. So what we measure is what an assistant can find and say about you today, not what a model happened to memorise. Until August 2026 we asked without search, and it made us wrong in one direction only: one business went from being named by none of three assistants to being named by two of them, and the size of the model made no difference, only the searching did. A scan that under-reports is worse than no scan, because it tells a business it is invisible to assistants that in fact name it first.
Which means a zero from us is still the pessimistic reading, though less so than before. A zero now says the assistants we asked did not name you when they looked — not merely that a model failed to recall you. What remains of the gap is the consumer product itself, which has personalisation and a ranking of its own that we cannot reproduce. The error runs in one direction, and it is the direction that makes us look wrong rather than you look bad. Searching costs more, which is why a repeat check is one you choose to run, for 2 tokens, rather than something that happens every day.
3.Which questions we ask
Five questions per run. If you have attached keywords to the site, they decide what the questions are about — your choice of battleground. Otherwise the questions are written from what we read on your own site, covering the kinds of question that make assistants name businesses.
The questions are shaped so that an answer is about you and your market, not about something that shares your name or a neighbouring country, and at least one of them asks the way a customer who has never heard of you would ask. That one is the hardest and the most honest.
Every run records the exact questions it used and where each one came from, so a score you disagree with can be traced to the questions that produced it. Both are shown on your own score in the dashboard, together with what each question returned: which assistants answered it, which of them named you, and whether the question already contained your name. That last one decides whether the row can be read at all — an assistant repeating a name we handed it looks identical to one that knew, so those rows are marked unreadable rather than counted as visibility. And a count is always out of the assistants that answered that question: one we could not reach is left out of the row, never counted as an assistant that did not know you.
Which brings us to the part that is easy to leave out. However a question arrives, the sentence itself is written by a model or by us, not by the person you are trying to reach. Attaching a keyword decides what a question is about; it does not put the question in a customer's words.
And none of them is evidence that anybody asked. A question can be plausible, or chosen by you, without a single real person ever having put it to an assistant. Making that different means buying search-demand data, and we have not: the licensing terms of the free sources do not permit it, and we will not build a paid feature on a number we cannot cite. We also cannot fill the gap from our own traffic, because we decided not to store the text of what agents ask us.
This is the single largest weakness in the score, and it is not a flaw in how the answers are counted. It is that the questions are a reasonable guess at demand rather than a measurement of it. Treat the score as "how visible am I on these questions", never as "how visible am I to the market".
4.What counts as a mention
A mention is your business named in the answer. Spelling, accents and the company form do not hide a match, so *"Cofetăria Bună"* is found as *"cofetaria buna"*.
A name that is also an ordinary word — "Tunnel", "Slack", "Delta" — appears in sentences all the time, so for these the same answer must also contain something that ties it to your business before the mention counts. Names specific enough to be yours count on their own.
This is deliberately conservative. A false positive is worse than a miss here: it tells you that you are visible when you are not, and you would stop working on the thing that needs work.
5.What counts as a citation
A citation is stronger than a mention: the assistant pointed at you as a source rather than merely naming you. If the assistant returned the sources it used, we look for your site among them; otherwise we look for your site written out in the answer.
Only Perplexity currently returns a source list, so for the other assistants a citation means your domain was written out in the answer.
6.How the number is made
Every assistant is asked every question. An answer that names you counts; an answer that names you and points at your site counts for more. The counts are combined across the five assistants and expressed out of 100. Serving your llms.txt from your own domain earns a small bonus — never a gate: not doing it costs nothing, and it cannot push a score above 100.
Alongside the score we record two rates that are easier to reason about than the composite: mention rate and citation rate, each the share of all answers that named or linked you.
The score is also shown as a letter grade from A to F. An A is genuinely hard — it means being named in almost every answer, by almost every assistant — but it is reachable.
7.How often it runs
Every new website and business profile gets its first check free, on every plan. After that, nothing runs on a schedule: every further check is one you ask for, on any plan, and costs 2 tokens. A negotiated Custom plan can still set a schedule of its own.
Assistant answers do not move fast enough for a frequent check to tell you anything, and every check asks with search on, which costs real money — so you decide when one is worth it. What has changed over time is the frequency, never the method: a cheaper check that under-reports is worse than a rarer honest one.
Every run stores its change against the previous run, so you can see movement rather than only a level. It withholds that change whenever the two readings were not taken the same way: a changed method, a changed set of questions, or a different AI model answering for an assistant. You are told which of those it was. A missing comparison is not a comparison of zero.
8.The accuracy audit
Separately from the score, every run compares what the assistants *said* with what your own site says, and flags what is wrong: a contact number that is not yours, a site that resembles yours but is not, a mention with no source attached.
You are notified when findings first appear, not on every run that re-confirms them.
9.Known limitations
Stated plainly, because you will find them anyway:
- The score covers assistants we can query. An assistant with no public way to ask it a question is not in the score at all. We would rather leave a gap than fill it with a number we did not measure — but it is a gap, and if your customers live on an assistant we cannot reach, the score is silent about them.
- We ask through each vendor's service, not the consumer product. Since August 2026 we send every question with web search switched on, which removes the larger half of this limitation. What remains is real: the model behind a consumer app is not always identical to the one we can call, and the consumer product has personalisation and a ranking of its own that we cannot reproduce. The error leans towards showing you as less visible than you are, but by much less than it did.
- We use each vendor's economical models. A larger model may answer differently — though when we tested it, searching decided whether a business was named and model size did not.
- Five questions is a small sample. Scores move for reasons that are noise, not signal. Watch the trend over a fortnight, not the change between two days.
- Mention matching is textual. An assistant that describes your business perfectly without naming it scores zero, and correctly so — but it is still a loss the number does not capture.
- Citations are mostly unavailable. Four of the five assistants return no source list, so their citations depend on the answer writing out your address.
- Answers are non-deterministic. The same question can come back differently twice.
10.Checking it yourself
You do not have to take the number on trust. Every run records the exact questions and, for each one, which assistants answered, which named you and which linked to you. Put the same questions to the same assistants and compare question by question, not one number against another. If what you see differs from what we recorded, the questions are the first place to look.
We would rather you checked. A score you cannot audit is a score you should not pay for.
11.Changes to this methodology
Changing how the score is measured changes every customer's history, so we treat this document as part of the product. When the method changes, this page changes with it, the date above moves, and scores from before and after the change are never subtracted from each other.
Questions about any of it: contact@tunnelpowered.com. Related reading: the agent and landline documentation.