AI Visibility Score Explained: Metrics That Actually Matter
Short answer: an AI visibility score is a summary of a few underlying measurements: how often you are mentioned, how often you are cited, how prominent you are when you appear, and how that compares with competitors. The summary is only as trustworthy as the prompt set and the formula behind it. Learn the components, and use the score as a headline rather than a decision tool.
I have looked at a lot of dashboards that show a single “visibility” figure, and the most common problem is not that the number is wrong. It is that nobody can say what it means. This post breaks the number into the pieces that matter and tells you which ones to act on.
- There is no standard formula. Compare a score only with itself, over time, in the same tool.
- Mention rate, citation rate, position and share of voice answer different questions.
- Prompt selection decides more than any formula.
- Report by engine, then combine. A blended number hides the engine where you are losing.
Why a single score is tempting and risky
Executives like one number. A score is easy to chart, easy to put in a slide and easy to set a target against. The risk is that it compresses several things into one, and the compression hides exactly what you would want to fix.
Imagine two brands, each scoring 40. Brand A is mentioned in 40% of answers but never linked. Brand B is mentioned in 25% and linked in 55%. Same score, very different problems, and different fixes. One needs better third-party coverage so it is trusted enough to be a source. The other needs to appear in more of the answers at all.
So before looking at any score, ask three questions:
- Which prompts is it computed from?
- Which engines, and are they weighted?
- What counts: a name in the text, a link, or both?
The components, defined
Here are the measures that go into most scores. I use the same definitions in our own reporting, and I recommend writing them down so every report means the same thing.
Mention rate
Definition: the percentage of answers, across your prompt set and runs, in which your brand name appears in the generated text.
Use it for: the broadest read of whether you are in the conversation at all.
Watch out for: a single mention in a long list counts the same as a top recommendation unless you also track position.
Citation rate
Definition: the percentage of answers that link to a page you own, or to a page that is about you, as a source.
Use it for: understanding whether the engine treats you as a source, and whether your pages or third-party pages carry your story.
Watch out for: engines differ in how many sources they show. An engine that lists 15 sources per answer will produce higher citation rates for everyone than one that lists three. Compare within an engine, not across.
Position or prominence
Definition: where you appear in the answer. First named, one of several in a list, or a passing mention at the end.
Use it for: separating a recommendation from a footnote. Position is often a better predictor of what a reader remembers than a raw mention.
Watch out for: it is subjective unless you define tiers in advance. Keep it to three levels: lead, listed, passing.
Sentiment or framing
Definition: whether the answer describes you favorably, neutrally or critically.
Use it for: catching the case where a high mention rate is hiding a problem, such as an answer that says your product is “limited” or “expensive.”
Watch out for: automated sentiment scoring is rough. Sample and read the actual text.
Share of voice
Definition: your mentions divided by all brand mentions in the same answers, as a percentage. It is relative by design.
Use it for: competitive reading. It tells you whether you are gaining or losing the category conversation, not only whether you appear. Our fuller treatment is in AI share of voice.
Watch out for: it depends on which competitors you include. Leave out the real rival and your share looks inflated.
Source mix
Definition: the domains and page types that engines cite for your prompts: review sites, Reddit threads, video, your own pages, competitor pages.
Use it for: deciding where to work. The source mix is the closest thing to a to-do list that AI visibility tracking produces.
Referral traffic
Definition: visits that arrive from AI engines, measured in analytics.
Use it for: connecting visibility to business outcomes. It is a floor, because many influenced buyers never click. See how to measure AI referral traffic in GA4.
What goes into a composite score
A tool’s visibility score might be built from any mix of the above. Typical shapes:
- Simple: mentions divided by total answers, as a percentage.
- Weighted: mentions weighted by position, so a first recommendation counts more than a passing mention.
- Multi-factor: mention, citation and sentiment combined with weights chosen by the vendor.
- Engine-weighted: a blend across engines, sometimes weighted by assumed usage.
None of these is wrong. They are different instruments. A simple score is easy to explain, and a weighted one is more informative but harder to audit. The practical rule: if a vendor cannot tell you how the number is calculated in two sentences, do not make decisions off it.
Prompt selection matters more than the formula
I would rather see a crude formula on a good prompt set than a sophisticated one on a bad set. Three common distortions:
Branded-heavy sets. If most of your prompts name your brand, mention rate will be near 100% and tell you nothing. The prompts that measure real discovery are the unbranded ones. See branded vs unbranded prompts.
Prompts you chose because you win them. It feels good and distorts the trend. Build the list from buyer language, not from your best results, as explained in how to choose prompts to track.
A changing prompt set. If you add or drop prompts between reports, you cannot compare periods. Freeze a core set and add a separate “experimental” set.
Noise: why a point change usually means nothing
Engines do not return identical answers. Run the same prompt ten times and you may see your brand in six, or in four. That is just variance. A few implications:
- A move of a few points between two reports is usually noise unless the prompt count and repeats are high.
- Use repeated runs and look at the average.
- Look for sustained direction across several periods, not single jumps.
- Large, sudden changes are worth investigating, because the engine itself may have changed how it retrieves sources. That happens: in ReddGrow’s own tracking of ChatGPT answers, Reddit’s share of citation rows fell from roughly 3% to under 1% across about a month in mid-2026, which moved every brand that depended on that source. See why ChatGPT cites Reddit.
By engine, then combined
Do not stop at a blended number. ChatGPT, Perplexity, Gemini, Claude and Google’s AI features all pull from different sources and show different numbers of citations. A brand can be strong in Perplexity and invisible in ChatGPT, and the blended score will average that into something uninformative.
My preferred report layout is a small table:
| Engine | Mention rate | Citation rate | Lead position | Share of voice |
|---|---|---|---|---|
| ChatGPT | ||||
| Perplexity | ||||
| Gemini | ||||
| Google AI |
Fill it with real numbers from your own tracking. The headline is a weighted average of those rows, with the weights stated.
Metrics that sound good and mislead
Impressions. Some tools estimate how many people saw an answer that mentioned you. These are modeled estimates, not counts, and they flatter dashboards. We argued this in AI search impressions are a vanity metric.
A score with no trend. A score of 62 is meaningless. A score that moved from 41 to 62 over four months on a fixed prompt set tells you something.
Cross-tool comparisons. Different tools, different scores. Do not compare yours with a number from a different vendor’s report.
A practical way to use a score
- Freeze the prompt set for at least a quarter.
- Write the formula down in a shared document. Include the engines, the weights and the number of repeats.
- Track mention rate, citation rate and share of voice separately, and let the score sit on top as a headline.
- Pair every movement with a cause. Did you publish something, win a thread, lose a source, or did the engine change?
- Act on the source mix, because that is where the levers are.
ReddGrow shows these components separately, including the Reddit threads behind your citations, in its AI brand visibility tracking. The broader tool landscape is covered in best LLM SEO rank trackers 2026.
The bottom line
Use the score as a smoke detector, not a diagnosis. It tells you something moved. The components tell you what, and the source mix tells you what to do about it. If you remember one thing, make it this: a number you cannot decompose is a number you cannot improve.
Frequently Asked Questions
- What is an AI visibility score?
- It is a single number, usually 0 to 100, that a tool computes from how often your brand is mentioned or cited across a set of prompts. There is no industry standard formula, so two tools can give the same brand very different scores. Always ask what goes into the number before comparing it.
- What is the difference between mention rate and citation rate?
- Mention rate is the share of answers where your brand name appears in the text. Citation rate is the share of answers that link to one of your pages, or to a page about you, as a source. They move independently, so a brand can have a high mention rate and a low citation rate, or the reverse.
- What is a good AI visibility score?
- There is no universal benchmark, because the score depends on your prompt set, your category and how the tool calculates it. The useful comparison is your own trend over time and your position against named competitors on the same prompts. A number in isolation says very little.
- How is share of voice different from visibility score?
- Share of voice is relative: your mentions as a percentage of all brand mentions in the same set of answers. A visibility score is usually absolute, such as the percentage of answers that mention you. You can have a flat score and a falling share of voice if competitors are growing faster.
- Can an AI visibility score be gamed?
- Yes, mostly through the prompt set. If you only track branded prompts, or prompts you already win, the score looks excellent and means little. A score is only as honest as the prompts behind it, so audit those first.