AI visibility score: what it measures, and how it is calculated

An AI visibility score is a single index, usually 0 to 100, that summarizes how often and how prominently your brand appears in AI-generated answers across a set of tracked prompts. It blends mention frequency, citation links, competitor share and sometimes sentiment into one number. Every vendor calculates it differently, so the score is useful as a trend line for your own brand and close to meaningless as a cross-tool comparison.

That caveat is the whole reason this post exists. Teams see a score of 34 in one dashboard and 71 in another and assume one of them is broken. Neither is. They asked different questions of different models and weighted the answers differently. Here is what actually goes into the number, and what to do about it.

What goes into an AI visibility score

Strip away the branding and almost every vendor is measuring the same four or five things. The differences are in the weighting, not the inputs.

How is an AI visibility score calculated?

The mechanics are simpler than the dashboards suggest. A tracker holds a fixed list of prompts, say 50 buyer questions. Once a day or once a week it sends every prompt to each engine it covers, captures the full text of each answer, and parses it for brand names and links. Each answer becomes a row: mentioned yes or no, linked yes or no, position in the response, competitors named alongside you.

The score is then an average across that grid, weighted by the vendor formula and normalized to 0 to 100. If you are mentioned in 20 of 50 answers on one engine and 10 of 50 on another, your raw mention rate is 30%. What the dashboard shows you might be 42, because citations were weighted double and one engine counted for more than the other. That gap between the raw rate and the published index is exactly why the numbers do not travel between tools.

Two design choices distort the result more than anything else. The first is the prompt set. A list stuffed with prompts containing your own brand name will produce a flattering score that predicts nothing, because nobody discovering you for the first time types your name. The second is engine mix. Tracking ChatGPT alone is cheap and common, and it will read very differently from a blended score across ChatGPT, Claude, Gemini and Perplexity, where only about 11% of cited domains overlap between engines.

What is a good AI visibility score?

There is no benchmark worth quoting, and any vendor that gives you one is selling something. The score is a function of the questions you picked. Ten broad category questions in a competitive market might leave a strong brand at 25. Ten narrow questions about a niche workflow might put the same brand at 80.

Two comparisons are genuinely useful. Compare yourself against the named competitors on your own prompt set, since that is measured under identical conditions. And compare this month against last month, with the prompt list unchanged. Change the prompts and you have reset the baseline, so freeze the list before you start treating the trend as real.

Why the score moves when you did nothing

AI answers are non-deterministic. Ask the same model the same question twice and you can get two different brand lists, particularly in categories where a dozen products are roughly equivalent. On top of that, model updates land without warning, vendors adjust their weighting, and retrieval layers refresh their index on their own schedule.

The practical rule: one reading is noise, four weekly readings are a signal. If your score drops six points on a Tuesday, do nothing. If it drifts down across a month while a competitor climbs, go read the pages the engines are citing instead of yours. That is where the answer is.

What the score cannot tell you

This is the limitation that catches teams out. A tracker reports that you are absent from an answer. It does not tell you why, and it cannot change it. The cause is almost always one of two things, and both live on your page rather than in the dashboard.

The first is ranking. Roughly 92% of pages cited in AI answers already rank in the top 10 for the query, because engines build their shortlist from pages the search layer already trusts. If you sit on page three, no amount of prompt monitoring will help; the fix is ordinary on page SEO.

The second is extractability. Plenty of pages rank well and still never get quoted, because the answer is buried four paragraphs down, the heading does not match the question, or the comparison is written as prose instead of a table. Models lean on corroboration too, so it pays to watch where your brand gets mentioned across the wider web, not only inside AI answers. The mechanics of that selection are worth reading in full in the breakdown of how answer engines choose which sources to cite.

How to check your AI visibility score for free

You can reproduce the core measurement in an afternoon, and it is worth doing before you spend $99 a month on a dashboard.

1. Write ten buying questions. The questions someone asks before they know you exist. No brand names, including your own.
2. Ask each one in four engines. ChatGPT, Claude, Gemini and Perplexity, logged out or in a temporary chat so memory does not feed your brand back to you.
3. Log three columns. Mentioned, linked, and which competitor won. Forty rows total.
4. Divide. Mentions over 40 is your mention rate. Links over 40 is your citation rate. That is the same measurement the paid tools normalize into an index.
5. Repeat monthly. Same prompts, same wording. The trend is the only part that matters.

If you would rather compare the paid options first, the AI visibility tracker comparison lays out what Profound, Otterly, Ahrefs Brand Radar and Semrush each track, with their public prices as of July 2026.

How to actually move the number

Once you know which questions you lose, the work is on the page. Rank for the query first, then make the answer easy to lift: the direct response in the first 40 to 60 words under a heading that repeats the question, a real table where you are comparing things, schema that names you as the source, and an honest last-updated date. Freshness matters more here than most people expect, with 76.4% of the most-cited ChatGPT pages updated within the previous 30 days.

That is the loop worth running. Measure by hand or with a tracker, find the questions where a competitor is named instead of you, then rewrite those specific pages. The answer engine optimization playbook covers the format layer surface by surface, and Seomake's free audit inside its ai seo software will read a live URL the way an extractor does and write the answer block, headings and schema it is missing.