AI Share of Voice: What It Measures and Misses

Last updated October 15, 2026 · ~6 min read

The short answer: AI share of voice tells you how often assistants name you for the prompts someone chose to track. It doesn’t tell you whether those are your buyers’ questions, whether you were recommended or just mentioned, or whether the answer got your product right. Use it as a trend on a fixed set of real buyer questions, split into mentions, citations and recommendations, and read the answers behind it.

Hasmord competitor research for a demo product: buyer searches where a competitor’s answer is shown, the evidence it uses, and the gap to close. — Competitor research in Hasmord: where a competitor is the answer, and the evidence it uses (demo workspace).

How AI share of voice is calculated

Tools differ, but most follow the same pattern: pick a set of prompts, ask one or more assistants, and count how often each brand appears.

Answer share: the answers that mention you, divided by all answers in the set.
Mention share: your mentions, divided by the mentions of every brand tracked, so the shares add up to 100% across you and your competitors.
Weighted versions count a first-place or highlighted mention more than a passing one.

Two reports can both say “share of voice” and mean different things. Before you compare numbers, check which formula and which prompt set each one uses.

Where one percentage misleads

Someone chose the denominator. As Dan Taylor argued in Search Engine Land, buyers can ask an assistant almost anything, so a tool samples a limited set of prompts and turns the results into one percentage. Change the prompts and the share changes.
A mention is not a recommendation. In Lily Ray’s analysis of 100 B2B “best software” queries, reported by Search Engine Land, Google’s AI Overviews cited a brand’s own listicle and still recommended someone else in 69% of those cases. A count of appearances can’t see that.
The same prompt gets different answers. Assistants vary from run to run and between their apps and APIs, so a single snapshot is one reading, not a position. See why the same prompt gets different picks.
A high share can hide a wrong answer. Being named with the wrong size, price or limit counts the same as being named correctly.

What to track instead (or alongside)

Measure	What it tells you	What it misses
Mention share	How often assistants name you for your tracked questions	Whether the mention was positive, a recommendation, or correct
Citation share	How often your own pages are used as a source	Answers that describe you from a retailer or reviewer page instead
Recommendation share	How often you are the product the answer suggests for the buyer’s conditions	Why you lost when you didn’t win
Answer accuracy	Whether the facts about your product were right	Nothing on its own about volume; pair it with the shares above

How to build a share-of-voice report you can act on

Start from real buyer questions. Use the questions your sales team and support hear, with their conditions attached, not generic category prompts. See from sales-call questions to better AI answers.
Fix the set and version it. Keep the same questions, assistants and products from one report to the next. When you add questions, start a new series instead of mixing them in.
Ask on a schedule. Several rounds smooth out run-to-run variation. Record the date, assistant and full answer each time.
Split mentions, citations and recommendations. Report them separately; a rise in one can hide a fall in another.
Read the answers you lose. For each question where a competitor is recommended instead, note the fact or source that decided it. That is the to-do list.
Report coverage and limits. Say how many questions and assistants the number covers, and that it is a sample, not the whole market.

More on the method in AEO tracking and AI visibility tools: what they measure.

Where Hasmord fits

Hasmord asks AI assistants your buyers’ own questions in rounds through the Anthropic, Google, OpenAI and Perplexity APIs, and records whether each answer picks your product and what it weighed. Duplicate wordings of a question are merged, so one question isn’t counted twice. Citations shows whether assistants used your pages as a source, and Competitor Research shows what competitors answer well that you don’t. The work that follows is fixing the facts behind the questions you lose, with every change approved by your team. API answers can differ from what people see in an assistant’s app.

Find your product opportunities Start free Talk to us

Frequently asked questions

What is AI share of voice?

It is the share of AI answers, across a set of prompts, that mention your brand compared with competitors. Tools calculate it differently: some divide the answers that mention you by all answers, others divide your mentions by the mentions of every tracked brand.

Is AI share of voice a reliable metric?

It is reliable as a trend on a fixed, documented set of questions and assistants. It is unreliable as a market-wide number, because someone chose the prompts, answers vary from run to run, and a mention is not the same as a recommendation.

How is AI share of voice different from citations?

Share of voice counts mentions of your brand in the answer text. Citations count the pages an assistant used as sources. You can be mentioned without your page being cited, and cited without being recommended.

How do I improve my AI share of voice?

Find the buyer questions where a competitor is recommended instead of you, read what decided each answer, and fix the missing or unsupported facts on your pages. Then ask the same questions again. Chasing the percentage itself usually leads to content that doesn’t answer anyone.

AI share of voice: what it measures, what it misses, and what to track instead