The definition
What is an AI visibility score, and what does it measure?
The short answer
An AI visibility score is a number that claims to say how present your brand is in AI answers. Two very different things are sold under that name. A signal score grades your own setup out of 100: schema, crawler access, press footprint, profiles. An answer measurement asks the engines real questions and counts how often you are named. A signal score says you are eligible. An answer measurement says whether you are named.
Published September 20, 2026 · updated September 22, 2026 · part of our public GEO method · written by Thibault Genon-Catalot, founder
Two families, one word
The same name for two different measurements.
Signal scores: is the door open?
A signal score reads your own properties and grades them. Instant Press's free AI Authority Score, for example, rates 13 signals out of 100: a Wikipedia article, a Google Knowledge Graph panel, schema markup, whether robots.txt lets the AI crawlers through, the press footprint, the links between your profiles (read on instantpress.co, September 8, 2026). Every one of those is worth having. Together they mean an engine that wants to quote you technically can. They are inputs, and they are all under your control, which is exactly why they are cheap to grade and free to give away.
Answer measurements: are you in the room?
An answer measurement ignores your setup and looks at the output. It takes the questions your buyers actually type, sends them to the engines several times each, and counts who gets named and who gets named first. The number it produces is a share of the answers, not a grade of your website. It costs real money to produce, because every question is a paid call to an engine, which is why almost nobody gives it away free.
Why the difference decides what you buy
A signal score tells you what to fix on your own site this week. An answer measurement tells you whether fixing it changed anything. Buy the first one for free from anybody, and never mistake it for the second. The clearest test of any vendor selling an "AI visibility score": ask whether the number comes from your website or from the engines' answers, and how many answers it is built on.
Side by side
What each number can and cannot say.
| Signal score | Answer measurement | |
|---|---|---|
| What it looks at | Your own properties: schema markup, a Wikipedia article, a Knowledge Graph panel, whether your robots.txt lets AI crawlers in, how many press placements carry your name, whether your profiles link back to each other | The answers themselves. Real questions sent to the real engines, several times, with a count of who gets named |
| What it produces | One number out of 100, usually with a colour and a list of missing signals | A share: how many of the answers about your market name you, against the brands that take your place |
| What it can tell you | Whether an engine that wants to quote you technically can: you are readable, identifiable and consistent across the web | Whether an engine actually does quote you today, on the questions your buyers type |
| What it cannot tell you | Whether a single engine has ever named you. The signals are inputs, not outcomes | Why you are absent. A measurement counts; it does not diagnose on its own |
| What it costs | Free, almost always. The score is the door to a sales call | Real engine calls cost money, so it is paid work. Ours is $1,500 for 300 answers |
| How fast it moves | Slowly. Add schema today, the number moves today, and nothing else changes | Constantly. Ask the same engine the same question three times and the list of brands changes |
Our own numbers
Why one number out of 100 cannot hold this.
In September 2026 we ran two full audits for B2B companies in two different markets, a software agency in France and a ticketing platform in Singapore. Each one is 20 or 21 buyer questions, sent to five engines, three times each: 615 answers in total, and 1,226 distinct brands named inside them. Here is what those answers do to the idea of a single stable score.
A score out of 100 hides every one of those facts. The useful unit is a share of answers, with the number of answers it rests on written next to it, and the date it was taken. That is what our method produces, and the reasoning behind each engine and each run is on that page.
Our own score
We have the signals. We are still not named.
Our own site has the whole signal checklist in place: organization and person schema, an open robots.txt for the AI crawlers, consistent profiles, a sitemap, 25 pages indexed. Any free signal score would grade us well. Until September 20, 2026 we asked the engines 27 questions a week, one engine per weekday; since September 21, 2026 we ask the 20 questions of our own audit. In the week of September 14, 2026 we were named in 0 of the 135 answers, which is exactly where we started on August 31, 2026. We publish that line every week, up or down, on /share-of-answer.
That gap is the entire point of the distinction. Signals make you eligible. Answers show whether you are named. A vendor who only ever shows you the first number is selling you the easy half.
Questions
Fair questions about scores. Straight answers.
Does a high AI visibility score guarantee that ChatGPT will name me?
No, and the companies selling free scores say so themselves. Instant Press, whose free AI Authority Score grades 13 signals out of 100, writes in its own FAQ that the score "makes you eligible to be cited; it does not guarantee the citation" (read on instantpress.co, September 8, 2026). Eligibility and citation are two different measurements. We run both on ourselves: our signals are in place and, in the week of September 14, 2026, we were named in 0 of the 135 answers we measured.
Is an AI visibility score the same thing as Domain Authority?
No. Domain Authority is a third-party guess at how well a page will rank in a list of links, built mostly on the link graph. An AI visibility score is about being named inside a written answer, which the engines assemble from sources that often do not rank at all: only 12% of the URLs cited by ChatGPT, Gemini and Copilot rank in Google's top 10 for the same prompt, against 76% for Google AI Overviews (Ahrefs, August 2025, 15,000 long-tail queries). A strong domain with no presence in those sources still gets left out.
Why does my score say 82 out of 100 while I am nowhere in the answers?
Because the score is grading your paperwork. Schema, a clean robots.txt and a Knowledge Graph panel make you readable and identifiable. They do not put your name in the passage the engine writes, which is assembled from ranked lists, review platforms, communities and media. 86% of the top 50 most-mentioned sources are not shared across ChatGPT, Perplexity and Google AI Overviews (Ahrefs, June 2025), so being absent from the handful of sources one engine reads is enough to be absent from its answers.
How many times should the engines be asked before a number means anything?
More than once, and on more than one engine. In the 615 answers we measured for two B2B companies in September 2026, asking the same engine the same question three times returned the identical list of brands 2 times out of 205. Across the five engines, 74% of the 1,226 brands named were named by a single engine, and 2 to 4% by all five. One screenshot of one answer is an anecdote.
What should I actually do with a free score?
Use it for what it is good at: a checklist of technical gaps you can close in an afternoon. Let the AI crawlers in, put your name and the same one-line description on every profile, mark up your company with schema, link the profiles back to each other. Then measure the outcome separately, by asking the engines your buyers' questions. The first part is free and worth doing. It is the second part that tells you whether it worked.
Free, 2 minutes
Before you buy any audit: run one free check.
5 buyer questions, 2 AI engines, and you see who gets named instead of you. The verdict shows on the page, and a copy lands in your inbox.
Where to go next
Take the free score. Then measure the answers.
Run any free signal score you like, close the gaps it finds, and keep the afternoon it costs you. Then find out whether it changed anything: the $1,500 audit sends 50 to 70 of your buyers' questions (10 per territory) to five engines three times each, 750 to 1,050 answers, delivered in 48 hours without a call. What the rest of the market charges for the same work is at /geo-price-comparison, the full price map is at /geo-pricing, and the reason a brand that ranks first on Google can still be invisible in ChatGPT is at /why-chatgpt-ignores-your-brand. The market numbers behind all of it, sourced and dated, are on /ai-search-statistics. Our own study of 615 AI answers, and what it says about measuring any of this once, is at /do-ai-answers-change-every-time. Three pages go further: does domain authority decide whether AI engines name you, does a high score guarantee citations, and how to improve your AI visibility.