How to Measure AI Visibility: The 7 Metrics That Matter
AI visibility is measured with 7 metrics: share of voice, mention rate, citation share, answer position, prompt coverage, sentiment, and recommendation rate. Definition, formula, and a worked example for each β from one consistent scenario.
If AI assistants are the new front door to product discovery, "are we visible?" is no longer a yes-or-no question. ChatGPT, Claude, Gemini, Perplexity, and Copilot generate a different answer every time, mention different brands, and cite different sources. To manage that, you need numbers β not a gut feeling.
AI visibility breaks down into seven metrics. Each answers a distinct question, each has a formula, and each points to a different fix. Track one in isolation and you get a misleading picture; track all seven and you can see exactly where you are winning and where you are invisible.
The 7 metrics at a glance
| Metric | Question it answers | Formula (simplified) |
|---|---|---|
| Share of voice | How much of the category conversation is us vs competitors? | Your mentions Γ· all brand mentions Γ 100 |
| Mention rate | In what share of answers do we appear at all? | Answers naming you Γ· total answers Γ 100 |
| Citation share | When a source is linked, how often is it us? | Answers citing your domain Γ· answers citing any source Γ 100 |
| Answer position | When we appear, are we first or buried? | Average order of your mention within the answer |
| Prompt coverage | Across our topic set, where are we invisible? | Prompts where you appear Γ· tracked prompts Γ 100 |
| Sentiment | When AI talks about us, is it favorable? | (Positive β negative mentions) Γ· total mentions |
| Recommendation rate | When named, are we actually recommended? | Answers recommending you Γ· answers naming you Γ 100 |
One running example
Every number below comes from a single scenario, so you can trace how the metrics relate to each other:
You track 300 category prompts and run each one across 4 assistants β ChatGPT, Claude, Gemini, and Perplexity. That is 1,200 answers per measurement cycle. All examples reuse these same figures.
1. Share of Voice
Definition. Share of voice (SoV) is the percentage of all brand mentions in a category that belong to you. It is the single best measure of competitive dominance inside AI answers β your slice of the conversation.
Formula.
AI Share of Voice = (Your brand mentions / Total brand mentions in the category) Γ 100Example. Across the 1,200 answers, assistants name brands 2,400 times in total. Your brand accounts for 360 of those mentions. Your SoV is 15% β competitive, but roughly five out of six mentions in your category still go to someone else. Maya tracks this against named competitors in competitive benchmarking.
What good looks like. As a rough scale: 30%+ is dominant, 15β29% is strong, 5β14% is moderate, under 5% is low. Treat the bands as orientation, not law β and watch the trajectory more than the snapshot: a brand at 12% growing two points a month is in a better position than one at 20% and sliding.
2. Mention Rate
Definition. Mention rate is the share of AI answers that name your brand at all. Where SoV measures dominance relative to competitors, mention rate measures raw presence: do you show up, yes or no?
Formula.
Mention Rate = (Answers that name your brand / Total answers tested) Γ 100Example. Of the 1,200 answers, your brand appears in 360. Your mention rate is 30%. Note how this differs from SoV: you are present in three out of ten answers, and you hold 15% of all mentions β because when competitors appear, they are often named alongside more rivals. High mention rate with low SoV is the classic signature of a brand that is "always in the room but never the loudest." Presence across assistants is what AI visibility tracking monitors continuously.
What good looks like. Mention rate is category- and prompt-dependent. Track it per topic cluster; a rate that is high for your core product but near zero for adjacent use cases tells you exactly where to expand.
3. Citation Share
Definition. Citation share is how often your domain is the linked source when an AI answer cites its references, out of all answers that cite any source. Being named as a brand and being cited as a source are different things β citation is what earns referral traffic and signals authority to retrieval engines.
Formula.
Citation Share = (Answers citing your domain / Answers citing any source) Γ 100Example. Of the 1,200 answers, 720 include at least one source link. Your domain is one of the cited sources in 108 of them. Your citation share is 15%. If your mention rate is high but citation share is low, AI knows your brand but does not trust your website enough to link it β a content and authority problem, not an awareness problem. Source analysis shows which domains get cited in your place.
What good looks like. Citation share replaces click-through rate as a success metric. Inside an AI answer there is rarely a link to click, so being the cited source is how you capture the traffic that remains.
4. Answer Position
Definition. Answer position is where your brand sits within the answer β first named, third, or last. Assistants and users both weight the earlier items more heavily, so position is a quality dimension that mention rate alone misses.
Formula.
Answer Position = Average ordinal rank of your brand across all answers where it appears
(lower is better; 1 = named first)Example. Your brand appears in 360 answers. Averaged across them, you are the 4th brand named. An average position of 4.0 while three competitors consistently lead means you are visible but framed as an also-ran. Moving from position 4 to position 2 often lifts perceived credibility more than adding a few points of mention rate.
What good looks like. Aim for the top three. Position is heavily influenced by how directly and authoritatively your content answers the underlying question. (See "what not to measure" below for the trap of watching position in isolation.)
5. Prompt Coverage
Definition. Prompt coverage is the breadth of your topic map where you appear at least once. Where mention rate is measured across answers, coverage is measured across distinct prompts β it exposes whole subjects where you are completely absent.
Formula.
Prompt Coverage = (Distinct prompts where you appear at least once / Total tracked prompts) Γ 100Example. Of your 300 tracked prompts β features, comparisons, use cases, integrations, pricing questions β you appear in at least one answer for 120. Your prompt coverage is 40%. The 60% where you never appear is your invisible zone; it usually clusters around a use case or buyer question you simply have not published for. Prompt management is where you define and watch that topic set.
What good looks like. Coverage gaps are the most actionable metric on this list β each empty topic is a concrete brief for a page you have not written yet.
6. Sentiment
Definition. Sentiment measures the tone of your mentions: is AI describing you positively, neutrally, or negatively? A brand can be highly visible and still lose deals if the framing is lukewarm or critical.
Formula.
Net Sentiment = (Positive mentions β Negative mentions) / Total mentions
(ranges from β1 to +1)Example. Of your 360 appearances, 216 are positive, 108 neutral, and 36 negative. Net sentiment is (216 β 36) / 360 = +0.50. Healthy. If negatives spike β say, a recurring "limited integrations" line β that phrase is coming from somewhere in the AI's source data, and it is worth tracing and countering with sentiment analysis.
What good looks like. Around +0.5 is strong. Watch for specific recurring criticisms more than the aggregate score; a single fixable complaint repeated across answers is a direct product or content signal.
7. Recommendation Rate
Definition. Recommendation rate is how often, when your brand is named, the AI actively recommends it β as opposed to listing it as one option among many. It is the sharpest bottom-of-funnel metric: it separates "mentioned" from "endorsed."
Formula.
Recommendation Rate = (Answers that recommend your brand / Answers that name your brand) Γ 100Example. Your brand is named in 360 answers; in 90 of them the AI frames you as the recommended or best-fit choice. Your recommendation rate is 25%. Being named is table stakes; being recommended is what moves purchase intent. A high mention rate with a low recommendation rate means you are on the list but not the pick.
What good looks like. This metric responds most to third-party proof β reviews, comparisons, and authoritative endorsements β because assistants lean on external validation when they move from listing to recommending.
How many runs do you need?
Here is the part most "AI visibility" write-ups skip: AI answers are not deterministic. Ask ChatGPT the same question twice and you can get two different brand lists. Run a prompt once and you have not measured your visibility β you have taken a single sample from a distribution.
That has two practical consequences:
- One run is noise. A brand that appears in run A and vanishes in run B has not changed; you just saw two draws. Any metric built on a single pass is unreliable.
- You measure through repetition. Run each prompt several times per cycle and read the average and the trend, not the last result. A one-to-two point wobble in share of voice between runs is normal variance. A sustained multi-point move across many runs β that is signal worth acting on.
This is exactly why manual spot-checks ("I asked ChatGPT and we showed up") are misleading, and why measurement has to be systematic and repeated. It is also the single biggest reason to automate: getting a stable read means running hundreds of prompts, many times, across every assistant, on a schedule.
What not to measure
Just as important as the seven metrics is knowing which numbers will mislead you:
- A single-run snapshot. Covered above β one pass is a sample, not a measurement. Never make a decision on one run.
- Average position in isolation. Position only means something combined with mention rate. Being "position 1" in the two answers you appear in, while invisible in the other 1,198, is not a win. Weight position by how often you actually appear.
- A raw mention count with no denominator. "We were mentioned 500 times" is meaningless without the total. 500 out of 600 is dominance; 500 out of 50,000 is a rounding error. Always report share, not count.
- Vanity totals across unrelated prompts. Padding your prompt set with questions no buyer asks inflates coverage and mention rate while hiding weakness on the prompts that matter. Measure the prompts your customers actually use.
- Click-through rate. There is usually nothing to click. Chasing CTR inside AI answers optimizes for the wrong outcome; citation share and recommendation rate are the honest substitutes.
How the seven fit together
No single metric tells the whole story. They form a funnel:
- Prompt coverage and mention rate measure whether you show up.
- Share of voice and answer position measure how you stack up when you do.
- Citation share, sentiment, and recommendation rate measure whether that presence actually works in your favor.
A common pattern: strong mention rate, weak citation share and recommendation rate. You are known, but not trusted or chosen. The fix is not more awareness β it is citation-worthy content and third-party proof.
Read them together and each weakness points to a specific action:
- Low coverage β publish for the missing topics.
- Low share of voice or position β strengthen the pages that already surface.
- Low citation share β make your content the linkable source (structure, data, direct answers).
- Low sentiment β trace and counter the recurring criticism.
- Low recommendation rate β build reviews and honest comparisons.
The bottom line
"Are we visible in AI?" is really seven questions β measured through repetition, across every major assistant, on the prompts your buyers actually ask. Do that consistently and AI visibility stops being a guessing game and becomes a managed channel. The brands that treat these seven metrics as seriously as they once treated rankings and impressions will be the ones AI keeps recommending.
Measure all seven for your brand
See your share of voice, mention rate, citation share, answer position, prompt coverage, sentiment, and recommendation rate across ChatGPT, Claude, Gemini, and Perplexity β in one dashboard.