AI Visibility Metrics That Actually Matter
The AI visibility metrics that matter most are mention rate (the share of AI answers that name your brand), average position in the answer, citation share (your share of the sources engines link to), share of voice against competitors, and sentiment. Each answers a different question, and each needs a clear formula and enough sampled answers before it means anything.
This page defines the five, shows the formula behind each one, and works through all of them on the same small set of answers so you can see how they relate. If you are new to the topic, start with the overview of what AI visibility is and how it is measured; this article goes one level deeper into the numbers.
Why AI visibility metrics need their own definitions
Search metrics were built for a ranked list of links. A generated answer is different: it can name five brands, cite eight sources, praise one product and warn about another, and it will rarely look the same twice. So a single “ranking” cannot describe it, and a metric borrowed from SEO without adjustment usually misleads.
There is no industry standard yet, and tools label similar ideas differently. Peec AI, for example, describes visibility, position, sentiment and sources as its core measures, while Ahrefs counts a mention once per response no matter how often the brand appears in it, and a citation once per response per page (Ahrefs Help, September 2026). Those counting rules change the totals, so check them before comparing numbers across tools or reports.
Consultant Aleyda Solis frames the same idea as presence KPIs in her AI search optimization checklist: prompt coverage, recommendation rate, linked citation rate, comparative win rate and representation accuracy. The names differ, but they map closely to the five metrics below.
The five core AI visibility metrics and their formulas
| Metric | Question it answers | Formula | Unit |
|---|---|---|---|
| Mention rate (visibility %) | How often do engines name us? | Answers mentioning the brand ÷ total answers | % |
| Average position | How high up are we when named? | Sum of the brand’s positions ÷ answers mentioning the brand | Rank (1 = first) |
| Citation share | How much of the evidence comes from us? | Citations to our domain ÷ all citations in the answers | % |
| Share of voice | How much of the conversation is ours? | Our mentions ÷ mentions of all tracked brands | % |
| Net sentiment | How are we described? | (Positive − negative mentions) ÷ all mentions | −100% to +100% |
Mention rate, also called visibility %
Mention rate is the base metric. Count every collected answer, count the ones that name your brand, divide. A brand counts once per answer even if the answer repeats its name. Merge spelling variants and product names into one brand before counting, or the rate will be understated.
It is the most reliable of the five because it does not depend on the order of a list, which changes constantly. In SparkToro’s study with Gumshoe, the same brand list came back in under 1% of repeated runs, yet the top headphone brands still appeared in 55% to 77% of 994 responses (Search Engine Journal, January 2026). Rates hold up across runs; single positions do not.
Average position in the answer
Position is the brand’s place among the brands named in an answer: first, second, third and so on. Average it only over answers where the brand appears, and always report it next to mention rate. A brand that shows up once at position 1 has a perfect average position and almost no visibility.
Two refinements make it more useful. The first-place rate (answers where you are named first ÷ all answers) shows how often you are the default recommendation. And position should be counted among all brands in the answer, not only the competitors you track, or an untracked market leader will quietly push everyone’s real position down.
Citation share and citation rate
Citations are the sources an engine links to. Citation share is your domain’s slice of all citations in the answers you collected. Citation rate is the share of answers that cite your domain at least once. Citation share tells you how much of the evidence base you own; citation rate tells you how often you are part of it at all.
Track citations separately from mentions. An engine can cite your guide while recommending a competitor, and it can recommend you based on a review site without linking to your pages. Understanding how AI search engines choose which sources to cite helps explain why the two often diverge.
Share of voice
Share of voice compares you with your competitors: your mentions divided by the mentions of every brand you track. Unlike mention rate, the shares of all tracked brands add up to 100%, so it moves when a competitor gains even if your own rate stays flat. Its value depends entirely on which competitors you include, so keep the set fixed between reporting periods.
Sentiment
Sentiment classifies each mention as positive, neutral or negative based on how the answer frames the brand. Net sentiment summarizes it as one number. Read the negative mentions themselves, too: they often repeat an outdated price, a fixed bug or a single old review, and the cited sources usually show where that framing came from.
A worked example of AI visibility metrics from ten answers
The log below is a hypothetical example: one prompt (“best accounting software for freelancers”) run ten times on one engine. Brand A is the brand being measured; B, C and D are tracked competitors. Ten runs are far too few for a real decision, but they keep the arithmetic easy to follow.
| Run | Brands named, in order | Brand A position | Citations in answer | Citations to Brand A | Brand A sentiment |
|---|---|---|---|---|---|
| 1 | A, B, C | 1 | 4 | 1 | Positive |
| 2 | B, A, D | 2 | 3 | 0 | Neutral |
| 3 | B, C, D | – | 4 | 0 | – |
| 4 | A, C | 1 | 2 | 1 | Positive |
| 5 | C, B, A, D | 3 | 5 | 1 | Neutral |
| 6 | B, D | – | 3 | 0 | – |
| 7 | A, B, C, D | 1 | 4 | 2 | Positive |
| 8 | B, C | – | 2 | 0 | – |
| 9 | B, A | 2 | 3 | 0 | Negative |
| 10 | D, B, C, A | 4 | 4 | 1 | Neutral |
Applying the formulas to this log gives the following results.
| Metric | Calculation | Brand A | Brand B (for comparison) |
|---|---|---|---|
| Mention rate | 7 answers ÷ 10 | 70% | 90% |
| Average position | (1+2+1+3+1+2+4) ÷ 7 | 2.0 | 1.4 |
| First-place rate | 3 ÷ 10 | 30% | 50% |
| Citation share | 6 ÷ 34 citations | 17.6% | not tracked |
| Citation rate | 5 answers ÷ 10 | 50% | not tracked |
| Share of voice | 7 ÷ 29 tracked-brand mentions | 24.1% | 31.0% |
| Net sentiment | (3 positive − 1 negative) ÷ 7 | +28.6% | not scored |
Brand B’s figures come from the same log: B appears in nine runs at positions 2, 1, 1, 2, 1, 2, 1, 1 and 2, giving an average of 13 ÷ 9 = 1.4 and five first places. C appears seven times and D six, so the 29 tracked mentions split 7, 9, 7 and 6.
The example shows why no single number is enough. Brand A’s 70% mention rate looks healthy on its own, but Brand B is named more often, sits higher when named and is the default first pick in half the answers. Share of voice and first-place rate expose that gap; mention rate alone does not. Brand A, on the other hand, is well cited: half of all answers link to its site, so the engine treats its pages as useful evidence even when it recommends B.
It also shows the sampling problem. A rough 95% margin of error for a 70% rate from 10 answers is about ±28 points. The same rate measured over 400 answers would carry about ±4.5 points. Use small logs to learn the method and large samples to make decisions.
AI visibility score and AI visibility index explained
Many tools combine the components into one AI visibility score or AI visibility index, usually scaled from 0 to 100. The common idea is to reward both frequency and prominence, so a brand named first in many answers scores higher than one named last in a few.
A simple, transparent version is a reciprocal-rank score: give each answer 1 ÷ position points (1 for first, 0.5 for second, 0.33 for third, 0 if absent), add them up and divide by the number of answers. In the log above, Brand A earns 1 + 0.5 + 1 + 0.33 + 1 + 0.5 + 0.25 = 4.58 points across 10 answers, a score of 45.8. Brand B earns 7.0 points, a score of 70.
Composite scores are convenient for trend lines and executive summaries, but they hide the cause of a change: a falling score could mean fewer mentions, lower positions or both. Vendors also weight the parts differently, so the same brand can get different scores in different tools. Report the composite with its components underneath.
Which AI visibility metrics to report and how often
| Metric | Stability across runs | Best use | Common mistake |
|---|---|---|---|
| Mention rate | High with enough samples | Headline KPI, trend over time | Judging it from a handful of manual checks |
| Average position | Low | Secondary signal next to mention rate | Reporting it alone, or counting only tracked brands |
| Citation share | Medium | Content and PR planning, per engine | Mixing engines that cite 1–2 sources with ones that cite dozens |
| Share of voice | Medium | Competitive benchmarking | Changing the competitor set between periods |
| Sentiment | Medium | Spotting wrong or outdated facts | Trusting automatic labels without reading examples |
Report each metric per engine before you combine them, because engines differ sharply in how many sources they cite and which ones. Collect weekly, report monthly, and note any change in prompts, engines or competitors as a break in the series. These metrics also sit naturally next to classic rank data: comparing them with keyword rankings tracked over the same period shows whether Google gains are carrying over into AI answers.
Collecting enough answers by hand is slow, which is why most teams automate it. Asheard, for instance, tracks mentions, citations, position and sentiment across ten AI engines with a location per prompt and reports share of voice against named competitors; the AI visibility tracking page shows how.
Frequently asked questions
What is a good mention rate in AI answers?
There is no universal benchmark, because mention rate depends on the prompts you track and how crowded your category is. Unbranded category prompts in a busy market might give a strong brand 30% to 50%, while branded prompts should come close to 100%. Compare yourself with named competitors on the same prompt set and focus on the trend rather than an absolute target.
What is the difference between mention rate and share of voice?
Mention rate measures your brand against the total number of answers, so every brand’s rate can rise at the same time. Share of voice measures your mentions against all tracked brands’ mentions, so the shares always add up to 100% and one brand’s gain is another’s loss. Use mention rate for your own progress and share of voice for competitive standing.
Is average position in AI answers a reliable metric?
Only as a secondary signal. The order of brands in AI answers changes on almost every run, so position from a few checks is mostly noise. Averaged over many answers, and always shown alongside mention rate, it can reveal whether you tend to be the first suggestion or an afterthought. Never report it on its own.
How is citation share different from mentions?
A mention is your brand name appearing in the answer text. A citation is one of your URLs listed as a source. Citation share measures how much of the evidence an engine used came from your domain. You can be cited without being recommended, or recommended without being cited, so track both and look at the gap between them.