Share of Voice in AI Answers: The Formula and Its Limits
Sep 2, 2026 · 10 min read
How AI share of voice is calculated, what changes the percentage, and what an analyst needs to publish beside it.
Two dashboards can give the same brand very different share of voice scores without either calculation being wrong. Usually, the market has not moved. The denominator has.
AI share of voice is your brand’s share of all brand mentions across a fixed set of AI answers. Change the prompts, the counting rule, or the competitor list, and the percentage changes with them.
Similarweb’s Lululemon campaign proves the point. The same dataset produced 26.72% brand visibility and 11.87% brand mention share. One set of answers, two valid figures, and nearly fifteen percentage points between them.
A share of voice score is only as useful as the methodology published beside it. Here is how the calculation works, where the denominator changes it, and what the percentage cannot tell you.
Key Takeaways
- The denominator defines the metric. Most disagreements about share of voice come from what is included in that denominator.
- Two share-of-voice denominators are common: all brands named or a closed competitor list. Dividing by the answer count produces brand visibility instead, which is a related but different metric.
- In a closed-denominator calculation, the competitor set is an editorial judgment: adding two rivals can cut the figure without anything changing in the market.
- Contrarian: a share of voice figure with no published denominator cannot be interpreted properly. Ask for the methodology before reporting it.
- Share of voice counts names, not sources: it tells you how often engines say your name. It never tells you which page produced the mention.
What Share of Voice Means in an AI Answer
AI share of voice is your brand's slice of the brand names an engine produces when it answers your category's questions. Similarweb defines it as the percentage of all brand mentions across AI-generated responses that belong to your brand.
That definition is sound. It is also incomplete, because it names a numerator and leaves the denominator to the vendor.
Three quantities have to be fixed before the division happens.
- The prompt set: which questions were asked, in which market, on which engine, on which date.
- The counting rule: whether a brand named four times in one answer counts once or four times.
- The brand universe: which named entities are treated as brands, and which are ignored.
Change any one and the percentage changes. Nothing in the market has to move. A share of voice chart with a rising line and no stated method is a chart of your own tracking decisions. Our share of voice across six categories publishes those decisions beside the results.
The Denominator Statement
Clinical quality measurement solved this problem decades ago. The American Medical Association specifies a performance measure in three parts: a numerator statement, a denominator statement, and denominator exclusions. A fully specified measure carries all three, and the rate is never published alone.
Borrow the middle one. A Denominator Statement runs to one line: prompt set size, engine, country, date, counting rule, and competitor list. An analyst who reads it can reproduce the number. An analyst who cannot is taking the vendor's word.
How Share of Voice Is Calculated in Traditional Search
Search share of voice arrived from advertising, where the denominator was total category impressions or spend. Ahrefs states the general form as brand visibility value divided by total market visibility value. The visibility value itself depends on the channel you are measuring, in a guide published in August 2020 and last updated in October 2023.
The chosen channel determines what the calculation measures. In Ahrefs Rank Tracker, the visibility value is estimated clicks rather than appearances. A first-position result and a ninth-position result therefore carry different weights.
| Channel | Numerator | Denominator | What it weights by |
|---|---|---|---|
| Broadcast and print | Your impressions or spend | Total category impressions or spend | Money and reach |
| Paid search | Your ad impressions | Total eligible impressions for your targeting | Auction eligibility |
| Organic search in Ahrefs Rank Tracker | Your clicks across tracked keywords | All clicks to all results for those keywords | Position and search volume |
| Social listening | Your brand mentions | All brand mentions in the category | Raw mention count |
| AI answers | Your brand mentions | All brand mentions across tracked answers | Raw mention count |
How Share of Voice SEO Tools Set the Denominator
Ahrefs publishes its own arithmetic. In Rank Tracker, share of voice is your clicks against all clicks going to the SERPs you track. Its worked example uses a South African coffee SERP with 5,000 total clicks. The first result takes 1,500 of them and scores 30%.
Two things follow, and both carry into AI measurement. The number is volume-weighted, so one fat-head keyword can carry the whole figure. The number is bounded by the keyword list you chose. A share of voice SEO report describes your tracked set, not your market.
Ahrefs' help page adds a warning worth keeping: other platforms ship this same construction as a visibility score.
Why the AI Version Needs a Different Denominator
Clicks do not exist inside an AI answer. There is no ranked list, no click curve, and not always a link. So the click denominator that carries search share of voice has nothing to attach to. The metric falls back on counting names.
That leaves a methodological choice that many reports fail to state. AI share of voice commonly uses one of two denominators: every brand named or a closed competitor list. Dividing mentions by the number of answers produces brand visibility instead.
| Denominator | The question it answers | Where it flatters you | Where it punishes you |
|---|---|---|---|
| All brand mentions in the tracked answers | "How much of the category conversation is us?" | Concentrated categories with few named players | Fragmented categories where engines name eight brands |
| A closed list of named competitors | "How do we stand against the rivals we care about?" | Any category, because you chose the list | Nowhere, which is the problem |
| The number of answers tracked | "In what share of answers do we appear?" | This produces brand visibility, not share of voice | It does not reflect how many competing brands appear beside you |
Similarweb is unusually direct about the middle row. Some tools use a closed denominator that counts only the brands you added by hand. A challenger you were not watching then stays invisible, and your figure looks stronger than it is.
Similarweb's Lululemon campaign kept the denominator open. Its 11,295 mentions included Reddit and YouTube alongside Athleta, Nike, and Vuori.
That distinction matters. An open denominator counts what engines actually name, and some of it is not a competitor. A closed one counts a list you wrote.
Neither denominator is wrong. Publishing which one you used is not optional.
The Formula, Worked Through One Prompt Set
The share of voice formula is one line, and the terms have to be defined immediately around it.
Share of voice = your brand mentions ÷ total brand mentions in the same answer set
- Your brand mentions: answers in the set where your brand name appears, counted once per answer under the counting rule below.
- Total brand mentions: every brand mention in those same answers, under the same rule and the same brand universe.
- The answer set: one prompt set, one engine, one country, one date. Never a blend of runs.
Here is how to calculate share of voice end to end. The figures below are this article's own worked illustration, not a measurement of any real market. They exist so you can reproduce the arithmetic on the page.
The setup. A mid-market employment law firm runs 40 buyer prompts. One engine, one country, one day. The counting rule: one mention per brand per answer, no matter how often the name appears. The firm is named in 19 of the 40 answers. Those 40 answers carry 138 brand mentions in total.
| Denominator used | Arithmetic | Result |
|---|---|---|
| All brands named (open) | 19 ÷ 138 | 13.8% |
| Six-firm tracked list (closed) | 19 ÷ 90 | 21.1% |
| Brand visibility: answers containing the firm | 19 ÷ 40 | 47.5% |
The closed list holds six firms: the client, plus five named rivals accounting for 71 mentions. So 19 + 71 = 90. The other 48 mentions belong to national firms, legal directories, and marketplaces the list ignores.
One dataset and one day produce two defensible share-of-voice figures, 13.8% and 21.1%, plus a brand visibility figure of 47.5%. A slide that labels all three as share of voice is mixing different metrics. Any figure reported without its denominator is reporting a choice without explaining it.
Choosing the Competitor Set, and Why It Decides the Number
Professional services make the point sharply. The competitor set here is genuinely arguable. The firm competes with regional practices on some matters and national firms on others. Buyers ask engines about both.
Take the same 40 answers. Add two challengers to the tracked list, one with 16 mentions and one with 12. The denominator moves from 90 to 118.
The firm's share fell from 21.1% to 16.1%, and nothing in the market changed. Nobody published anything, nobody lost a deal, and the chart shows a five-point decline.
Three rules make the competitor set defensible.
- Fix the list before the first reading, not after it. A list edited once results arrive is a list edited to produce results.
- Publish it in full beside the number. Six names in a footnote cost nothing and settle most arguments before they start. Name what you left out as well, which is what the AMA's third part, denominator exclusions, exists to capture.
- Change it on a schedule, quarterly at most, and restate history when you do. Changing the denominator mid-quarter breaks the comparison between the two readings.
Similarweb's guidance points the same way. In fragmented categories even a market leader might hold 10% to 15% mention share. In concentrated ones the top brand can own 40% to 50%.
That spread is wider than almost any real quarter-on-quarter movement. Cross-category comparison is close to meaningless, and internal consistency is everything.
What a Share of Voice Percentage Cannot Tell You
Four limits belong beside every share of voice figure. None of them is a criticism of the metric. They define what the metric can and cannot support.
- It cannot name the source. Share of voice counts brand mentions across an answer set. It does not tell you which page, thread, or review produced any of them, and the mechanism behind the citations sits in a different report entirely.
- It cannot tell you what was said. A mention in a list of firms to avoid counts exactly the same as a recommendation. Sentiment is a separate reading and it is not optional.
- It cannot separate you from your denominator. A rising share can mean you gained mentions or that rivals lost them. Publish the raw counts alongside the percentage and the ambiguity disappears.
- It cannot survive a changed prompt set. Editing prompts between readings breaks the series. The old number and the new one are measuring different questions.
Similarweb draws the first limit itself. It separates share of voice from citation share, and notes that a brand can hold strong share of voice with weak citation share. Mentioned, never linked.
That gap is why a mention count cannot double as a work order. It tells you where you stand, and stays silent on what to do next.
Where the Number Moves and Where It Stalls
The size of one mathematical step is not the same as reliable resolution. In the worked set above, one additional answer naming the firm and no other brand moves the result from 13.8% to 14.4%, a change of roughly 0.6 points. But repeated runs may change several answers at once. Establish the reporting threshold from the variation between repeated baseline runs, not from the prompt count alone.
| What changed | Effect on the number | Read it as |
|---|---|---|
| You are named in more answers | Moves, slowly | A real gain, if the denominator held still |
| Engines name more brands per answer | Falls, sometimes sharply | A category effect, not your performance |
| You added rivals to the tracked list | Falls immediately | A reporting change, and label it as one |
| A rival exits the answers | Rises with no work done | Note it in the commentary before someone else does |
| You edited the prompt set | Any direction | A new series, not a continuation |
The number stalls in one common case: your mentions grow at roughly the rate the category's total mentions grow. Work is landing, the percentage sits still, and the program looks dead from outside. Keep the raw counts beside the percentage so the reason for any movement remains visible before taking the number to a board. Share of voice is one reading among several, and our guide to the metrics worth tracking explains what the others add.
How Klarivo Reports Share of Voice
Klarivo Monitor reports competitor share of voice across ChatGPT, Claude, Perplexity, Gemini, and Grok, tracks each engine independently, and updates on a schedule the client controls. It also reports query-level mention rates, sentiment, and a Top 10 Citation Sources list at domain level. The methodology matters as much as the feature list because a percentage cannot be interpreted without knowing how it was constructed.
That standard applies to every share of voice report, including ours. Publish the construction beside the number: the prompt set, the engine, the country, the date, the counting rule, and the competitor list. Keep the raw counts under the percentage. When the competitor list changes, restate the prior reading against the new denominator rather than re-basing quietly.
A number a client can argue with beats one they have to accept, and it is the only kind that survives a finance review. Ask us for that construction, and ask every other vendor on your shortlist for it too.
One limitation should remain explicit. Citation reporting is domain-level, so the surface names the domains engines drew on and never claims the individual page.
Changing what those domains say about you is separate work. That is where a mention count stops being a report and becomes a task list.
Want your share of voice reported with its denominator attached? Book a 15-minute Klarivo discovery call and choose a time that works for you. Tracking runs inside Klarivo Monitor, while our guide to how the program works explains what happens after the first reading.
Frequently Asked Questions About AI Share of Voice
How many prompts do we need before the number stops jumping around?
There is no universal prompt count that makes the number stable. Forty prompts may be enough to expose large gaps, but the count alone does not justify calling a one-point change meaningful. Run the frozen set at least twice, measure the spread between runs, and use that observed variation as the minimum threshold for reporting movement.
Should we report share of voice per engine or blended across all of them?
Per engine, always. Blend only for a headline you also break out. Each engine produces its own answer set, so a blended figure averages denominators that were never built to be comparable. Treat a wide gap between two engines as a question about those engines before you read it as a change in your standing.
Can we compare our share of voice against a number published by another category?
No. Category concentration makes direct comparisons unreliable. Similarweb puts fragmented-category leaders around 10% to 15% and concentrated-category leaders at 40% to 50%, so a 12% figure can be dominant or mediocre depending only on what you sell.
What should we ask a vendor that will not publish its denominator?
Four questions, in this order. Which prompts were run, and on what date. What counts as one mention. Which entities are in the denominator, and were they chosen by us or found by the tool. Are the raw counts exportable. A vendor answering all four is reporting. A vendor answering none is selling a chart.
Does share of voice still mean anything in a category with only two real competitors?
Less than you would like. With two players, share of voice is close to a mirror of your rival's number, and it moves in whole chunks because the denominator is small. Mention rate and sentiment carry more information in that shape of market. Report share of voice if leadership asks for it, and say plainly that a five-point swing may reflect a change in only one answer.
Start accelerating your Reddit presence
See how Klarivo can shift your visibility across AI, search and buyer communities.
Book a Demo
