What Is an AI Visibility Score? (How It's Measured)

An AI visibility score is a number that summarizes whether ChatGPT, Claude, Gemini and Perplexity recommend your business when someone asks a relevant question — and, just as importantly, how they respond when they don’t. It plays a similar role to a rank-tracking number in traditional SEO, but it measures a different thing: not where a page sits in a list, but whether an AI system chose to say your name out loud. Here’s what actually goes into one, honestly, including where the model breaks down.

Why a plain yes/no isn’t enough

The simplest way to measure this would be binary: did the engine mention your business, yes or no. That’s a reasonable starting point, but it collapses two very different outcomes into one bucket. A business that gets recommended first, by name, with a positive reason attached is in a completely different position than one that only comes up when a customer explicitly compares it to a competitor by name — and both of those are different again from a business the model has never heard of. A score built only on “mentioned or not” can't tell those three situations apart, even though they call for different action.

The three states behind an honest score

A more useful model separates visibility into three states, evaluated per engine:

• Found. The engine names your business unprompted, when someone asks an open recommendation question — “what’s the best plumber in Denver?” with no company named. This is the strongest signal: the model volunteered you.
• Knows you. The engine doesn’t volunteer you in a recommendation answer, but recognizes and discusses you correctly when asked a direct comparison question — “how does [your business] compare to alternatives?” The model has information about you; it just isn’t leading with it.
• Not found. Neither question surfaces you. The engine doesn’t recognize the business at all in this context.

Recommendation outranks comparison: if you’re named in the open recommendation answer, you’re found regardless of what the comparison answer says. The practical value of this model is that “knows you” and “not found” call for different fixes — one is a visibility problem, the other is closer to a ranking or positioning problem once you’re already on the model’s radar.

Turning three states into one number

Each engine gets a score based on its state: being found scores highest, with a small reduction the further down the answer you’re named (first named scores the most, easing down slightly per rank); knows you scores meaningfully lower, since recognition without recommendation is real but weak; not found scores zero. Those per-engine scores are then averaged across all four engines into one 0–100 visibility score.

The averaging matters more than it sounds. A score built from a single prompt, checked once, is noisy — the same business can plausibly score very differently on back-to-back runs if only one prompt and one moment are sampled. Scoring across both a recommendation and a comparison question, per engine, and averaging the results is what keeps a single unlucky (or lucky) answer from swinging the headline number.

Why the score moves even when you haven’t changed anything

AI answers are generated, not stored. A generated answer can shift because a competitor picked up a new review, a directory listing changed, the underlying search index refreshed, or the model itself was updated — none of which are things you did. That’s uncomfortable if you’re expecting a score to behave like a fixed grade, but it’s also the reason tracking on a schedule matters more than a one-time check: a single score is a snapshot, a trend over several weeks is a signal.

It’s also why locking the business inputs you’re scored against — name, website, category — for a period after you save a change is worth doing. Otherwise it becomes hard to tell whether a score moved because of the market or because you edited an input mid-week, which makes the trend line meaningless.

What an AI visibility score can’t tell you

Worth being honest about the limits. A score is built from the specific prompts it’s run against — it only reflects the questions actually asked, not every question a real customer might phrase differently. It also can’t promise a fixed benchmark the way an SEO tool might claim a “domain authority of 40 means X”: the category is new enough that there’s no industry-wide baseline yet. The most reliable use of the number is relative — is it moving in the right direction for your own business, on the actual questions your customers ask, not a universal ranking against every other company scored the same way.

How to read your own score

• Look at the per-prompt, per-engine breakdown before the headline number — that's where the actionable detail lives.
• Check which state you're in on each engine: not found calls for different work than knows you.
• Track the trend over several weeks rather than reacting to a single run.
• See who's named instead of you on the prompts where you're not found — that's often the fastest read on what's missing.

How Surfaced measures it

Surfaced runs this exact model every week across ChatGPT, Claude, Gemini and Perplexity, using the real questions your customers ask, with web search on. For each engine you can see whether you’re found, known, or not found, your rank when found, the sentiment, and who was named instead — rolled up into one score you can track over time rather than guess at from a single check.

Starter ($39/mo) covers one brand and up to 5 prompts. Agency ($149/mo) covers unlimited clients and prompts with white-label reporting. There’s also a free check with no signup if you want a first read today.

Related reading

For the mechanics behind why engines pick who to name, see how AI chatbots decide which businesses to recommend. For the checklist to evaluate any tool that produces a score like this, see best AI visibility tools. Or go straight to the tracker itself.

Frequently asked questions