Skip to content

ResourcesAEO Handbook

AI share of voice: what it measures and how to calculate it

Share of voice asks how much of the conversation in AI answers is yours. The arithmetic is simple; the counting rules underneath decide whether the number means anything.

Updated Jack F, Co-founder

MeasurementAEOshare of voiceAI visibilitycompetitors

Part of the Answyn AEO Handbook.

Now readingWhat AI share of voice is

AI share of voice is your share of all brand appearances in a set of AI answers, counting each brand once per answer: your appearances divided by every brand's appearances in the same answers. Mention rate divides by answers instead, and citation share counts links. Before trusting the figure, check that the platform merges spellings of the same rival and leaves out names that are not rivals, then read it by stage and engine.

Share of voice is the measure people reach for when they want to know who is winning the conversation in AI answers. It is also easy to get wrong without noticing, because the formula is trivial and the decisions about what counts as a brand are not.

We build Answyn, which reports share of voice, so we have a view on how it should be counted. The figures below come from our own tracking of our own category, including the raw name lists where the problems show up.

What is AI share of voice?

AI share of voice is your share of all the brand appearances in a set of AI answers, with each brand counted once per answer. Visibility asks how often answers name you. Share of voice asks how much of the naming, across every brand the answers mention, is yours.

When a buyer asks an assistant which tools to consider, the answer names a handful of brands. Share of voice is your part of those lists, added up across many answers. A brand named in half the answers can still hold a modest share of voice if every answer names six others beside it.

It is one of the separate measures Answyn reports in AEO Hub, beside mention rate, prominence, sentiment and citation share. None of them is blended into a score, and each one opens back to the answers that produced it.

How do you calculate AI share of voice?

Take a fixed set of answers: the same questions, engines, market and window. For each answer, list the brands it names, each one once, however many times the answer repeats a name. Then divide:

AI share of voice = your appearances ÷ all brand appearances in the same answers

With each brand counted once per answer, your appearances are simply the answers that name you. Share of voice and mention rate therefore have the same numerator and differ only in what they divide by.

An arithmetic illustration, not measured data: if an answer names four brands, each holds a quarter of that answer. Across ten answers that name 40 brands between them, a brand named in five of the ten has a mention rate of 50% (5 of 10) and a share of voice of 12.5% (5 of 40).

Counting once per answer is a choice, and it is the one we make. A long answer that repeats one name six times has still put that brand on the list once. Counting every mention gives more weight to wordy answers and the brands they dwell on, and some tools also count citations as well as names. Each method can give a different figure on the same answers, so a share of voice can only be compared with one counted the same way.

MetricCountsDivided byThe question it answers
Mention rate (visibility)Answers that name youAll answers in the setHow often are we named?
Share of voiceAnswers that name youAll brand appearances in the same answersHow much of the naming is ours?
Citation shareCitations to your domainAll citations in the same answersHow often is the source ours?

Share of voice falls when answers name more brands, even if you are named just as often. That makes it good at showing a crowded field forming, and misleading if you read it without the mention rate beside it.

What skews AI share of voice?

The counting rules, mostly. Two problems show up in raw answers before any formula runs, and one decision is yours to make.

The first is spelling. The same product appears under several names. In Answyn's own tracking of its category (UK, 6 October to 4 November 2026), Consider-stage answers wrote Otterly five ways:

Answyn's own tracking, Consider-stage answers, UK, 6 October to 4 November 2026. One answer can use more than one spelling.
SpellingConsider answers using it
Otterly.AI262
OtterlyAI71
Otterly.ai39
Otterly AI28
Otterly20

Scrunch appeared as "Scrunch AI" in 100 answers and "Scrunch" in 41; AthenaHQ as "AthenaHQ" in 50, "Athena" in 7 and "Athena HQ" in 4. A tool that does not merge spellings splits one rival into several smaller ones, so Otterly reads as a brand named in 262 answers with four minor names further down the list.

You cannot fix that by adding the rows together, because one answer can use two spellings and would then be counted twice. The merged figure has to be recounted answer by answer, with every spelling treated as one brand. The lists above are our own raw rival lists, which keep the spellings apart, and that is why each figure here is per spelling.

The second is names that are not rivals. Answers name the assistants themselves as the place buyers search, and Learn-stage answers in our category named analytics tools such as GA4 and Microsoft Clarity. Count them as brands and they take share from real rivals, and from you. Answyn starts every project with the AI engines and the companies that make them excluded, and reviews each newly detected name by reading the sentences it appeared in, so tools, data sources and venues that are not real options for the buyer drop out. Any name can be restored, and marking one as not a competitor takes it out of share of voice, rankings and average position everywhere.

The third is a decision: whether a product and its maker are one rival. Our Consider answers named Ahrefs Brand Radar in 23.5% of answers and Ahrefs in 16.9%. Count them together if your buyers see one choice, or apart if they see two. Then keep the rule, or the trend line will move when nothing in the answers did.

Read it by stage and engine

One blended share of voice hides where each rival wins. In our category the leader changed with the buying stage:

Share of the answers at each stage that named the brand (mention rate). Answyn's own tracking of its category, UK, ChatGPT, Google AI Overviews and Claude, 6 October to 4 November 2026. Otterly.AI is one spelling only.
BrandLearnConsiderPurchase
Semrush13.7%48.8%31.0%
Profound5.8%52.1%20.2%
Peec AINot in the top three49.2%17.3%
Ahrefs12.1%16.9%9.7%
Otterly.AINot in the top three30.4%12.1%

Profound was named in 52.1% of Consider answers and 5.8% of Learn answers. Semrush led Learn at 13.7% and Purchase at 31.0%. Across the 25 questions, Semrush was named in 560 answers that did not name Answyn and Profound in 525, close to level, so a pooled figure would put them side by side and say nothing about where each one wins.

The table shows mention rates, the raw material of share of voice. Within a stage every brand is divided by the same total of appearances, so ranking the brands by share of voice gives the same order as ranking them by these rates.

Do the same by engine, because the engines answer from different material. In the same window all 380 youtube.com citations came from Google AI Overviews, ChatGPT's most-cited domains were vendors' own pages, and both answers that named Answyn were ChatGPT's. A share of voice averaged across engines blends answers built from different sources.

Tagging each question with a buying stage before you collect is what makes this possible, and customer journey analysis in AI search covers how. Answyn's AEO Hub groups the pre-purchase stages into Learn, Consider and Purchase, and most figures can be filtered to one engine.

What to ask a platform that reports it

Before you compare share of voice across tools, or trust one tool's trend, ask:

  • How are appearances counted: once per answer, or every mention? Is any weight given to position or citations? Ask for the formula in writing.
  • Are aliases merged for rivals as well as for your own brand? Ask to see the raw names behind one rival's figure.
  • Can you exclude names that are not rivals, such as the assistants themselves or analytics tools, and does the exclusion apply everywhere the metric appears?
  • Can you split the figure by engine and by buying stage, or only see a blend?
  • Can you open the answers behind the number, with each brand shown where the answer names it?

The last question decides the others. If you cannot open the answers, you cannot check how anything was counted.

Answyn's own answers: it counts each brand once per answer, and the aliases on your brand profile count as your brand. The AI engines and their makers start excluded, and any name can be excluded or restored. Most figures filter to one engine, and visibility is shown by Learn, Consider and Purchase. Every figure opens back to the answers, with each named brand shown in colour where the answer says it. For rival spellings, check the raw names in any tool, ours included: the variants above come from our own lists.

How the named platforms compare on engines, limits and access is in the 10 best AEO platforms for UK businesses in 2026. To have your category tracked this way in Answyn, enquire. Access is by invitation, and every plan is quoted for the business using it. See how Answyn pricing works.

A note on the data

All figures come from Answyn's own tracking of its category: the Answyn project (answyn.com), 25 questions asked without a brand name, UK market, 6 October to 4 November 2026. ChatGPT and Google AI Overviews were collected every day (750 answers each, 30 of the AI Overview checks showing no overview) and Claude 75 times, last on 30 October: 1,575 answers in all. Claude is not on Answyn's standard roster, and a plan adds it only by agreement; we collect it for our own project.

Learn covers 7 of the questions, Consider 14 and Purchase 4. Stage rates are the share of the answers at each stage that named the brand, counting only answers that came back, and name variants are counted per spelling. It is one category in one market over one month, so read the pattern rather than the levels.

Questions people ask

How do you calculate share of voice across AI answer engines?
List the brands each answer names, counting each brand once per answer, and divide your appearances by all brand appearances in the same answers. Keep the questions, engines, market and window fixed, and calculate it per engine and per buying stage as well as overall.
What is the difference between share of voice and mention rate?
Mention rate, or visibility, divides the answers that name you by all answers. Share of voice divides them by all brand appearances in those answers, so it falls when answers name more brands, even if you are named just as often. In a crowded category a brand can have a high mention rate and a modest share of voice.
Is AI share of voice the same as citation share?
No. Share of voice counts brand names in the answer text, while citation share counts the links an answer cites. In Answyn's own tracking answyn.com was cited 17 times in a window where Answyn was named in 2 answers, so the two can move apart.
Why does a competitor's share of voice look lower than it should?
Often because its name is split across spellings. In Answyn's Consider-stage answers Otterly appeared as Otterly.AI, OtterlyAI, Otterly.ai, Otterly AI and Otterly. A platform that does not merge them counts five small brands instead of one larger rival.
Should ChatGPT or Google count as brands in share of voice?
Usually not. Answers name the engines as the place buyers search, not as options to buy, so counting them takes share from real rivals. Answyn excludes the engines and their makers by default, and you can restore them if your product competes with the assistants.
What should we look for in a platform that measures AI share of voice?
How it counts appearances, whether it merges aliases for rivals as well as for you, whether you can exclude names that are not rivals, whether it splits by engine and stage, and whether you can open the answers behind each figure. Without the answers you cannot check any of the rest.
Does pinning a competitor change its share of voice in Answyn?
No. Pinning chooses which rivals the charts show; share of voice still counts every brand the answers name, once per answer. To take a wrongly detected name out of the numbers, mark it as not a competitor.

Run the numbers on your own brand.

Answyn puts the questions your buyers actually ask to every major engine, daily, and records who gets named and who gets cited.