AI Visibility Score: Two Different Numbers and How to Read Each

An AI visibility score is a 0 to 100 number computed by a third-party tool, either from the signals on your pages (a readiness score) or from how often ChatGPT, Google AI Overviews, Claude, Gemini and Perplexity name or cite your brand across a fixed set of prompts (a presence score). No AI engine publishes an official one, so every score is one vendor's formula.

That is why a score of 62 tells you almost nothing until you know which kind was measured, and why two tools can hand the same brand very different results. It is not a reason to ignore the metric. It is a reason to know which score you are holding, what went into it, and how much of next week's movement is noise. If you want the wider picture first (what AI visibility is and the four states a brand can be in), start with our plain-English guide to AI visibility. This piece is about the number itself.

What are the two kinds of AI visibility score?

The two kinds are readiness and presence: a readiness score inspects a page, a presence score inspects AI answers. Tools sell both as an "AI visibility score", and they answer different questions.

	Readiness score	Presence score
Question it answers	Is this page easy for an AI engine to understand and quote?	Do AI engines actually name or cite this brand?
Input	One URL: structured data, headings, entity clarity, crawlability, trust signals	Many answers: a prompt set run across one or more engines
Speed of change	Moves the moment you edit the page	Moves only when engines re-read the web and answer differently
Cost to run	Cheap: one fetch and a parse	Every prompt on every engine is a paid call
Blind spot	Cannot see whether anyone cites you	Cannot tell you why you are absent

Our free AI visibility check is a readiness score: it fetches one URL, scores signals such as entity clarity, structured data, content structure and crawlability, and never sends a prompt to ChatGPT or any other engine. The presence score inside the CrawlTide app is the other kind. Before you compare a score with anything, confirm both numbers are the same kind. A readiness 80 next to a presence 30 is not a contradiction; it is the most common pattern there is.

Keep readiness in proportion, too. Google's guidance on AI features and your website says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." A readiness score measures proxies (clear structure, a named entity, a crawlable page) that make a page easier to quote. It does not measure a ranking factor any engine has published.

How is an AI visibility score calculated?

Each vendor uses its own formula, and most do not publish the weights, so here are ours as a worked example. Formula as of October 2026 (CrawlTide presence score, version 2). The score is built from three components, calculated separately for each engine the app measures (ChatGPT, Google AI Overviews, Claude, Gemini and Perplexity):

• Mention rate: the share of answers that name your brand in the text.
• Citation rate: the share of answers that list your domain among their sources.
• Competitor pressure: the share of answers where none of your named competitors appear as a source. More competitor presence, fewer points.

With competitors configured, the weights are 41 for mentions, 35 for citations and 24 for competitor pressure. Without competitors, that third component is switched off rather than awarded or deducted, and the weights become 54 and 46. A few rules keep the number honest:

• A brand that appears in no answer at all scores 0, full stop.
• Prompts the engine never answered are dropped from the denominator, so an outage cannot lower your score.
• Google searches that showed no AI Overview are dropped too, because there was no answer to appear in.
• The overall score is the plain average of each engine's score, so an engine you sampled with more prompts does not outweigh the others.

A worked example with the arithmetic

Take a hypothetical engine with 20 measured answers. Your brand is named in 6, your domain is cited in 4, and a competitor's domain is cited in 10.

• Mention: 41 × 6/20 = 12.3, rounded to 12
• Citation: 35 × 4/20 = 7
• Competitor pressure: 24 × (1 - 10/20) = 12
• Score: 31

Now delete the competitor list and score the exact same 20 answers: 54 × 6/20 rounds to 16, 46 × 4/20 rounds to 9, total 25. Nothing in the AI answers changed, yet the AI visibility score moved by six points because the configuration did. That is the single most useful thing to understand about any vendor's number.

Why do AI visibility tools give the same brand different scores?

Because they differ in prompts, engines, definitions and arithmetic, and any one of those can move the result more than your own work did. The worked example already shows the weights. Here are the rest, roughly in order of how much damage they do to comparability:

• The prompt set. Twenty branded prompts ("is [brand] good for...") will flatter you. Twenty unbranded category prompts ("best tool for...") will not. The prompt set is the biggest lever on the result, and most tools choose it for you.
• The engine mix. A brand cited often by one engine and rarely by another gets a different average depending on which engines are included.
• What counts as presence. Some tools credit only a brand-name mention, some only a cited link, some both.
• How missing data is handled. If a tool scores an engine outage or a search with no AI Overview as "not visible", your score drops for reasons that have nothing to do with you.
• The weights. As above: identical answers, different arithmetic.

So compare a score only with itself: same tool, same prompt set, same engines, same competitor list. The moment any of those change, start a new baseline instead of reading the change as progress.

How much of a weekly change is just noise?

More than most dashboards admit. AI answers are not fixed: ask the same engine the same question twice and the sources can differ. A 2024 study, "Non-Determinism of 'Deterministic' LLM Settings" (Atil et al., revised April 2025), ran five LLMs configured to be deterministic over eight tasks, 10 runs each, and found accuracy variations of up to 15% across runs. Those were models set up to give identical output. Consumer chat assistants with search attached have even more moving parts.

The practical rule: a few points of movement between two runs is not evidence of anything. Run the same prompts several times before you change a page, note how far the baseline wanders on its own, and only treat a move larger than that spread as real. Our write-up on running an AI visibility test without fooling yourself covers repeat counts and control groups in detail.

What is a good AI visibility score?

There is no industry benchmark, and any tool that tells you "above X is good for everyone" is selling a rule of thumb. The CrawlTide app labels presence scores of 70 and above "Strong", 50 to 69 "Improving", 25 to 49 "Needs work" and under 25 "Low visibility". Those bands describe our formula only. A 45 in our app and a 45 elsewhere are not the same achievement.

Two comparisons are worth more than any band. The first is your own trend over the same prompt set. The second is how named competitors score on that identical set, in the same tool, in the same week. If a competitor sits at 60 on prompts where you sit at 20, that gap tells you something real; the absolute 20 does not.

How should you read readiness and presence together?

Side by side, because each combination points to a different next step:

• High readiness, low presence: your pages are easy to quote but engines are not choosing them. The usual causes are weak coverage of the questions people actually ask, or stronger sources on the same topic. Work on content and on being cited elsewhere, not on more markup.
• Low readiness, decent presence: engines know you despite your pages, often from your brand name or third-party coverage. Fixing structure and entity signals protects what you have.
• Both low: start with readiness on your five most important pages. It is cheaper, faster to change, and you can verify the fix the same day. Our guide to optimizing your website for AI search walks through the usual fixes.
• Both high: stop optimizing the score and watch the prompts you are still losing, one by one.

How do I check my AI search visibility for free?

Combine three free sources, each of which covers a different part of the picture:

• Readiness: run your key pages through the free AI visibility check. It is free and sends no prompts to any engine.
• Presence, by hand: write ten prompts your buyers would ask, run each in ChatGPT, Gemini, Perplexity and a Google search, and record whether your brand is named and whether your domain is cited. Repeat the same prompts a few days later before you read anything into the result.
• First-party citation data: Bing Webmaster Tools has an AI Performance report (public preview since February 2026) that shows how often your pages are cited as sources in Microsoft's AI answers. Google Search Console counts AI Overviews and AI Mode traffic inside the overall Performance report, and some properties also see a beta Generative AI report that shows impressions (not clicks) from those features.

The manual method stops scaling at a few dozen prompts across several engines, which is the point where a tool earns its cost. CrawlTide's AI visibility tracking runs your tracked prompts across all five engines on paid plans and scores it with the formula above.

Does a higher AI visibility score mean more traffic?

Not directly, and that is the trap in reporting the score on its own. Pew Research Center analyzed the March 2025 browsing of 900 U.S. adults and published in July 2025 that users clicked a traditional search result in 8% of visits when an AI summary appeared, against 15% when none did. Links inside the summary itself were clicked in 1% of visits.

Being named in an answer has value (brand recall, the follow-up search for your name), but it is a different value from a click. Report the AI visibility score next to Search Console clicks and branded search volume for the same period. If presence rises and branded searches follow, you have a story. If presence rises and nothing else moves, you have a number.

What an AI visibility score cannot tell you

• What happens outside the prompt set. A presence score covers only the prompts that were run. Real users ask millions of variations no tool samples.
• Why an engine chose a source. No engine explains its selection, so a score shows the outcome, not the cause.
• Personalized answers. Logged-in users, location and chat history can change what an engine says; a tool sees one neutral session.
• Revenue. A mention is not a visit and a visit is not a sale. Tie the score to clicks, branded search and leads before you call it a result.

Check the cheap score first

Run your most important page through the free AI visibility check to get its readiness score and a prioritized list of fixes. Fix what it flags, then measure presence on a fixed prompt set, several runs deep, before you call anything a win.