How Is an AI Visibility Score Calculated?

Last verified: August 14, 2026.

An AI visibility score is calculated by measuring how often your brand appears in a defined set of AI-generated responses, then optionally combining that presence with signals such as recommendation strength, citations, sentiment, share of voice, and cross-model consistency. The simplest reproducible formula is: responses mentioning your brand ÷ total eligible responses × 100. If your brand appears in 18 of 36 responses, its mention-based visibility is 50%. There is no universal industry formula, so before comparing scores, check exactly what the tool measures, which prompts and AI engines it covers, and how any composite score is weighted.

A practical decision map is:

If you need to know…	Measure…
How often your brand appears	Mention rate
How often AI actively suggests your brand	Recommendation rate or strength
Whether your site supports AI answers	Citation coverage
How AI describes your brand	Sentiment
How you compare with competitors	AI share of voice
Whether visibility is similar across engines	Cross-model consistency
One number for reporting	A transparent composite plus its component scores

The key principle is simple: the raw measurements should remain visible even when you calculate a headline score.

What does an AI visibility score actually measure?

"AI visibility score" is not a standardized metric.

Different platforms use the term differently. Peec defines its Visibility Score as the percentage of tracked AI responses in which a brand appears. Semrush describes its AI Visibility metric as a 0 to 100 competitive benchmark and keeps mentions, citations, topics, prompts, and audience estimates visible beside it. Ahrefs takes another approach in Brand Radar, reporting mentions, citations, impressions, and AI share of voice as distinct AI visibility metrics rather than collapsing all of them into the same number.

Keyword.com publishes a different formula: detection rate multiplied by a rank score, then scaled to 100. The public AI Visibility Index methodology is another composite example, combining mention frequency, position and prominence, citation quality, and query coverage with disclosed weights. These are methodology choices, not interchangeable industry standards.

That means a score of 60 from one methodology is not necessarily equivalent to a score of 60 from another.

Before trusting a score, check six things:

1. Which prompts were evaluated?
2. Which AI engines or answer surfaces were included?
3. What counts as a brand mention?
4. What is the denominator?
5. Which additional signals affect the score?
6. How are those signals weighted?

Without those answers, the score may still be useful as an internal trend, but it is difficult to compare externally.

How do you calculate the baseline mention rate?

The cleanest starting point is response-level brand presence.

Mention rate formula

Mention Rate = Responses mentioning your brand ÷ Total eligible responses × 100

Suppose a B2B company tracks:

• 12 buyer prompts
• 3 AI engines
• 1 response per prompt and engine

That produces:

12 × 3 = 36 eligible responses

If the company appears in 18:

18 ÷ 36 × 100 = 50%

Its mention rate is 50%.

How current published formulas differ

The most useful way to understand the category is to compare the formulas directly.

Published method	Core formula	What it rewards	Main comparison caveat
Peec Visibility Score	Mentioning responses ÷ total responses × 100	Consistent presence	Does not add position to the headline visibility percentage
Keyword.com Visibility Score	Detection Rate × Rank Score × 100	Presence plus earlier position	Rank is calculated only where the brand was detected
AI Visibility Index	40% mention frequency + 25% prominence + 20% citation quality + 15% query coverage	A broader composite	Components are normalized within the index methodology
Semrush AI Visibility	Normalized competitive benchmark from its prompt and topic dataset	Presence relative to competitors and topic demand	The public metric is not the same as a raw mention percentage

These formulas answer different questions. A brand can improve its raw mention rate without moving earlier in answers. It can improve position on the responses where it appears while remaining absent from most prompts. A normalized industry benchmark can also move when competitors change, even if the brand's own raw count stays similar.

Do not ask which formula is universally correct. Ask which decision the formula supports and whether the underlying observations are available for audit.

Count at the response level rather than counting every repetition of the brand name. Ahrefs, for example, counts a brand appearing multiple times inside one AI response as a single mention for that response.

The denominator also needs to stay stable. If your first report uses 36 responses and the next suddenly includes additional prompts, models, languages, or markets, the two percentages are no longer directly comparable without accounting for the methodology change.

A useful measurement record should therefore contain:

Field	Example
Exact prompt	"Best software for monitoring AI brand visibility?"
Intent	Commercial comparison
AI surface	ChatGPT Search
Market	United States
Date	2026-08-14
Brand mentioned	Yes
Recommended	Shortlisted
Citation present	Yes
Sentiment	Positive
Competitors mentioned	Brand B, Brand C

Prompt selection deserves particular attention. If you only measure branded questions, your visibility will probably look very different from a set dominated by category, comparison, problem, and buying-intent questions.

BrandJet's guide to building a prompt set for AI search monitoring covers how to structure that underlying sample.

Which signals should sit beside mention rate?

Mention rate answers whether you appear. It does not tell you how you appear.

For a useful visibility scorecard, separate at least five additional signals.

Recommendation strength

Compare these statements:

• "Acme is the strongest choice for this use case."
• "Options include Acme, Beta, and Gamma."
• "Acme offers feature X, while Beta focuses on feature Y."
• "Companies in this category include Acme."

Every answer mentions Acme, but they do not represent equal commercial visibility.

Start with a simple recommendation rate:

Recommendation Rate = Responses explicitly recommending your brand ÷ Total eligible responses × 100

If you want to score recommendation strength, define the rubric explicitly. For example, an internal framework might assign more value to being the first recommendation than to an incidental mention.

The exact weights are a methodology choice, not an industry standard.

Citation coverage

A brand mention and a website citation are also different events.

Use:

Citation Coverage = Responses citing your brand domain ÷ Total eligible responses × 100

If 9 of 36 responses cite your site:

9 ÷ 36 × 100 = 25%

You might also want to know whether mentions are supported by your own domain:

Mention-Supported Citation Rate = Brand-mentioned responses with a brand-domain citation ÷ Brand-mentioned responses × 100

If 12 of your 18 brand mentions include a citation to your site:

12 ÷ 18 × 100 = 66.7%

Ahrefs similarly distinguishes citations from brand mentions in its AI visibility methodology.

Keeping the two metrics separate reveals situations such as:

• your brand is recommended but another website is cited;
• your site is cited while a competitor is recommended;
• your brand is frequently mentioned but rarely supported by owned sources.

Sentiment

Sentiment asks what happens after your brand gets mentioned.

A response can describe the brand positively, neutrally, negatively, or with mixed language. Peec reports sentiment separately from visibility and describes it as an analysis of the tone surrounding brand mentions.

This distinction prevents a misleading result.

A brand appearing positively in 2 of 100 responses is not necessarily more visible than one appearing neutrally in 50.

Track presence first, then use sentiment to diagnose how that presence is framed.

AI share of voice

Visibility and share of voice use different denominators.

Visibility asks: In what percentage of eligible responses did we appear?

AI share of voice asks: Of the tracked brand mentions, what percentage belonged to us?

The basic formula is:

AI Share of Voice = Your brand mentions ÷ Total mentions among tracked brands × 100

Suppose your brand gets 4 mentions and competitors get 12:

4 ÷ 16 × 100 = 25% AI share of voice

If those 4 mentions occurred across 10 evaluated responses, your visibility could simultaneously be:

4 ÷ 10 × 100 = 40% visibility

Peec explicitly separates these two concepts.

BrandJet's focused guide on how to measure AI share of voice goes deeper into the competitive calculation.

Cross-model consistency

Do not calculate one average before checking each engine.

Imagine your mention rates are:

AI engine	Mention rate
ChatGPT	70%
Gemini	45%
Perplexity	25%
Simple average	46.7%

The headline average is 46.7%, but the difference between the highest and lowest results is:

70% – 25% = 45 percentage points

That 45-point spread is strategically important. It tells you the brand's visibility depends heavily on which engine the buyer uses.

Cross-model consistency is also different from run-to-run stability. The former compares engines. The latter asks whether repeating the same prompt on the same system produces stable observations.

Repeated-sampling research published in 2026 argues that generative-search visibility measurements should be treated as estimates drawn from a variable response distribution rather than fixed values, reinforcing the value of repeated observations for important measurements.

How can you turn these metrics into one composite score?

You can combine the components when stakeholders need one headline KPI, but the weights should always be disclosed.

For example, an illustrative internal formula could be:

AI Visibility Index =

• 40% Mention Rate
• 20% Recommendation Strength
• 15% Citation Support
• 10% Sentiment
• 15% Cross-Model Consistency

This is an example, not BrandJet's proprietary formula and not an industry standard.

Suppose the component scores are:

Component	Score	Weight	Contribution
Mention rate	50.0	40%	20.0
Recommendation strength	55.0	20%	11.0
Citation support	66.7	15%	10.0
Sentiment	72.0	10%	7.2
Cross-model consistency	66.6	15%	10.0
Composite		100%	58.2

The resulting illustrative score is 58.2 out of 100.

The number is useful only because you can see what produced it.

A transparent composite should pass this checklist:

• Every component has a clear definition.
• Every denominator is documented.
• Qualitative classifications use a disclosed rubric.
• Each weight is visible.
• Prompt and engine coverage stays consistent between reporting periods.
• The component scorecard remains available beside the composite.
• The methodology has a version and an effective date.
• Changes to prompts, engines, weights, or entity rules are recorded before the next comparison.

Keep share of voice separate unless you deliberately want competitive presence inside the composite. Because both mention rate and share of voice derive from brand presence, including both without considering overlap can unintentionally overweight the same underlying signal.

Version the methodology before reporting trends

A monthly score is only comparable with the previous month when the measurement contract is materially stable.

Record a methodology version such as AIV-1.2 and log changes to:

• prompt set and prompt categories;
• monitored engines and surfaces;
• countries, languages, personas, and search state;
• brand and competitor entity rules;
• repeated-run count and collection schedule;
• component definitions, normalization, and weights.

If one of those changes, calculate the old and new method in parallel for at least one reporting window when practical. Otherwise, label the change as a new baseline. A silent methodology change can look like growth or decline even when the underlying visibility did not move.

For a binary mention rate, report a confidence interval when the sample is small. The point estimate tells you what the sample observed. The interval shows how much uncertainty remains. This matters when a few responses can move the headline score by several percentage points.

Why do AI visibility tools produce different scores?

When two platforms disagree, first compare the measurement systems rather than asking which number is "correct."

Differences can come from:

• prompt selection;
• prompt volume;
• AI engines covered;
• search-enabled versus other answer modes;
• geography and language;
• collection date;
• run frequency;
• brand entity matching;
• competitor selection;
• citation definitions;
• recommendation classification;
• weighting and normalization.

Semrush, for example, describes its AI Visibility Score as a combination of topic coverage and mention consistency, while Peec's published Visibility Score is based directly on the percentage of responses containing the brand.

Those numbers answer related but different questions.

Use this decision rule:

Compare AI visibility scores directly only when the prompt universe, AI surfaces, market, time period, entity rules, denominator, and calculation are materially equivalent.

Otherwise, use each methodology primarily to measure change against its own historical baseline.

For competitor analysis, also inspect the raw competitive signals rather than relying only on the aggregate. BrandJet's guide to monitoring competitor mentions in AI search shows how to structure that comparison.

What should you report each month?

For a lean B2B team, a compact scorecard is usually more useful than one unexplained number.

Track:

1. Mention rate
2. Recommendation rate or strength
3. Citation coverage
4. Sentiment
5. AI share of voice
6. Per-engine mention rates
7. Cross-model spread
8. Prompt-level gains and losses

Attach the measurement context every time:

sample size + prompt set + AI surfaces + geography + date range

Then investigate the biggest movements.

If mention rate rises from 42% to 55%, ask which prompts created the increase. If share of voice falls while your mention rate stays constant, check whether competitors gained mentions. If the overall average is flat but one engine drops sharply, inspect that model separately.

That is the purpose of an AI visibility score: not to create a mysterious ranking, but to compress a repeatable set of observations into something you can monitor and diagnose.

For teams moving beyond one-time spreadsheet audits, AI search monitoring provides the broader measurement framework. You can also review BrandJet's AI Search Monitoring page when evaluating an ongoing monitoring approach.

FAQ

What is the basic AI visibility score formula?

The simplest formula is:

Responses mentioning your brand ÷ Total eligible AI responses × 100

If your brand appears in 30 of 50 evaluated responses, the mention-based visibility score is 60%.

Is AI visibility score the same as AI share of voice?

No. Visibility measures how frequently your brand appears in the evaluated response set. AI share of voice measures your portion of tracked brand mentions relative to competitors.

Do citations count toward AI visibility?

They can contribute to a composite score, but they should also be reported separately. A brand can be mentioned without its website being cited, and a website can influence an answer without its brand becoming the primary recommendation.

What is a good AI visibility score?

There is no universal good score because methodologies differ. A more useful benchmark is your brand versus relevant competitors under the same methodology and your own score over time with a stable prompt set, engine set, and calculation.