What Is an AI Visibility Score and How Does It Measure Brand Discoverability?
An AI visibility score is a single number that summarises how often, and how favourably, AI assistants surface a brand across a chosen set of prompts. Most scores start from how often the brand appears in answers. Many then blend in its position in lists, whether its pages are cited, the sentiment of mentions and its share against competitors. It measures discoverability in the sense that matters for buyers: the chance that someone asking a relevant question hears your name. It is not like Domain Authority. It rests on a sample of answers that vary from run to run, and it depends entirely on the prompts, engines, runs and weights behind it. A score is only as useful as its disclosed method.
What goes into an AI visibility score?
| Component | What it measures | Common pitfalls |
|---|---|---|
| Appearance or mention rate | Share of answers naming the brand | Inflated by branded prompts |
| Position | Where the brand appears in lists | Unstable between runs |
| Citation rate | Share of answers linking to the brand's pages | Often confused with mentions |
| Sentiment or role | Whether mentions are recommendations, neutral or warnings | Hard to label consistently |
| Share of voice | The brand's mentions against competitors' | Depends on which competitors are tracked |
| Accuracy | Whether facts stated are correct | Rarely included, though important |
Each tool picks its own mix and weights, and many do not publish them.
What does an AI visibility score measure, and is it like Domain Authority?
It measures how visible a brand is in a sample of AI answers for chosen topics. Domain Authority and similar scores summarise a site's link profile. They are calculated from a large, fairly stable dataset, change slowly and mean roughly the same thing across sites. AI visibility scores differ in four ways:
- They are sampled. A different set of prompts can give a very different score.
- They vary between runs. SparkToro found less than a 1 in 100 chance of the same brand list appearing twice for the same prompt, so scores based on few runs are noisy.
- They depend on the category. In an analysis by Gumshoe, leading headphone brands appeared in 90–100% of answers while leading brand design agencies appeared in 30–40%. So a score of 40 can be strong in one market and weak in another.
- They are not comparable across tools. Two tools can give the same brand very different scores because they measure different things.
What is a transparent AI visibility score made from?
A transparent score discloses enough for someone else to reproduce it:
- The prompt set: the prompts themselves, how they were chosen, and how many are branded and unbranded.
- Engines and settings: which assistants and models, whether search was on, and the location and language.
- Runs: how many times each prompt was run, and when.
- Components and weights: exactly how appearance, citation, position, sentiment and share of voice combine.
- Competitor set: which brands are counted for share of voice.
- Uncertainty: the number of answers behind the score and a range or confidence interval.
How to calculate the core component is covered in how to calculate AI visibility as an appearance rate across repeated runs.
Why do scores from different tools disagree?
- Different prompts, especially different shares of branded prompts, which inflate results, as covered in whether brand-name prompts inflate AI visibility scores.
- Different engines and models, including whether they use the consumer app or an API.
- Different run counts, from single checks to many runs per prompt.
- Different weights on position, citations and sentiment.
- Different competitor lists for share of voice.
- Different locations and languages.
When two tools disagree, compare their methods before trusting either number.
How should you read a score critically?
Ask these questions of any score:
- How many prompts, and how many were branded?
- How many runs per prompt, and over what dates?
- Which engines, and were they tested with search on?
- What are the components and weights?
- How much would the score have to change to be meaningful?
- Can I see the underlying answers?
If a vendor cannot answer, treat the score as a rough indicator only.
How can you build your own transparent score?
Keep it simple and publish the formula. An example you can adapt:
- Appearance rate across unbranded prompts, weighted 50%.
- Recommendation rate (named as the top choice), weighted 30%.
- Citation rate for your own pages, weighted 20%.
Calculate each per engine from repeated runs, combine them with the weights, and report the result alongside each component and the run count. Keep branded prompts and accuracy as separate measures, since they answer different questions. Recalculate monthly with the same prompts and settings.
How should leadership use a visibility score?
A single number is useful for trend reporting and for keeping attention on AI visibility. Present it with three things beside it: the component rates, the number of answers behind it and a comparison with the same competitors measured the same way. Tie it to decisions. A falling score should prompt a look at which topics and engines moved, and a rising one should be checked against control prompts before anyone claims credit. Pair it with business measures such as branded search, AI-referred leads and what new customers say about how they found you. A score that never connects to those measures becomes a vanity metric.
What should a score never be used for?
- Comparing brands measured by different tools. The numbers are not on the same scale.
- Judging a single month's work. Month-to-month changes are often within normal variation.
- Setting targets without a method. A target of "70" means nothing unless prompts, engines and weights are fixed.
- Replacing the answers themselves. Read a sample of real answers each month; the score cannot tell you what assistants actually say.
What might an illustrative case look like?
Consider an illustrative accounting software company whose two visibility tools report scores of 72 and 31 for the same month. The first tool's prompt set is 60% branded questions and uses one run per prompt. The second uses only category prompts with five runs each. The company builds its own score from 40 unbranded prompts, ten runs each, across three engines, weighting appearance, recommendation and citation as above. Its score is 34, with a range of about 29 to 39 once the variation between prompts is allowed for. Leadership now reports that number, with the components beside it, and uses the second tool's data because its method is closer.
What do people ask about AI visibility scores?
What is a good AI visibility score?
There is no universal benchmark. Compare against your own history and competitors measured the same way.
Can a score be gamed?
Yes, by choosing easy or branded prompts. That is why disclosed prompts matter.
Should scores include sentiment?
Include it as a separate measure or component, with clear labelling rules, since negative mentions are not visibility you want.
How often do scores change?
They can move from month to month because of sampling and engine changes. Look for trends across several months.
Can we compare our score with an industry average?
Only if the average was calculated with the same prompts, engines, runs and weights. Most published averages were not, so treat them as rough context.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources