To measure AI share of voice in 2026, run a fixed set of buyer prompts through each AI engine several times, count every tracked brand named in the answers, and divide your mentions by all tracked brands' mentions, times 100, separately for each engine. No neutral benchmark for a "good" share of voice has been published, so judge the number against your own baseline and your named competitors, not against an industry average. To run that measurement every day, our pick is Cituna: it asks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode your prompts daily from $39 a month, computes share of voice per engine against the competitors in the answers, and counts the answers that use your page only as a source apart from the ones that name you.
What is AI share of voice, and how is it different from mention rate?
Two things happen in an AI answer, and each needs its own count. A mention is your brand named in the answer's text. A citation is the answer linking to a page on your domain. An answer can name you without linking to you, and link to you without naming you, as what counts as a citation in an AI answer explains.
From those two counts come three metrics that are often all called "share of voice".
| Metric | Calculation | What it tells you |
|---|---|---|
| Mention rate | Answers that name you ÷ all answers | How often you are in the answer at all |
| Share of voice | Your mentions ÷ mentions of every tracked brand x 100 | Your slice of the category conversation |
| Citation share of voice | Citations to your domain ÷ all citations in the same answers x 100 | How often the engine sends readers to your pages |
Mention rate is the steadiest of the three. SparkToro and Gumshoe had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times (published 27 January 2026). Lists rarely repeated, yet the leading brands kept appearing: across 994 answers to 142 differently worded headphone prompts, Bose, Sony, Sennheiser and Apple appeared in 55% to 77% of them. The authors judged visibility across many prompts and runs a reasonable metric, and rank position tracking unreliable. What is AI share of voice? covers the definition in more depth.
How do you calculate AI share of voice?
The formula is share of voice = your mentions ÷ all tracked brands' mentions x 100. Count each brand at most once per answer, so a brand named three times in one answer still counts once.
Here is a worked example. The numbers are hypothetical.
Say you track 50 prompts on ChatGPT and run each 3 times, giving 150 answers. You track five brands, including yours. Across the 150 answers, the five brands are named 400 times in total, and your brand is named in 60 answers.
- Share of voice: 60 ÷ 400 x 100 = 15%.
- Mention rate: 60 ÷ 150 = 40% of answers name you.
- Citation share: if those answers contain 900 citations and 45 point to your domain, 45 ÷ 900 x 100 = 5%.
The three numbers tell different stories. You are in 40% of answers, you hold 15% of the conversation, and you get 5% of the links. A brand can win on one and lose on another.
Check the arithmetic, even when an AI does it. In our scans on 28 September 2026 we asked all seven engines "how do I measure my brand's AI share of voice". Google AI Mode gave the right formula, then a worked example of 15 mentions for you against 75 for competitors, and printed 20%. On its own formula the answer is 15 ÷ (15 + 75) = 16.7%. Twenty percent is what you get by dividing by competitors' mentions only. Claude, in the same scan, gave the mention-share formula and then an example that divided by answers instead. If your spreadsheet or tool uses a different denominator from the one you think it uses, every trend line is off.
How do the tools calculate share of voice?
Vendors use the same name for different formulas. These are the definitions on each vendor's own help pages, checked 29 September 2026.
| Tool | What its share of voice counts | Weighting |
|---|---|---|
| Otterly.AI | Your brand mentions ÷ all brand mentions x 100. Each prompt run counts 1 or 0 per brand | Unweighted |
| Peec AI | Your brand mentions as a percentage of all tracked brands mentioned | Unweighted |
| Profound | Responses mentioning your brand ÷ total brand mentions across all responses | Unweighted |
| Semrush AI Visibility Toolkit | Mentions and how high the brand appears. For ChatGPT, the topic's search volume too. Exact weights not published | Position-weighted, plus volume for ChatGPT |
| Ahrefs Brand Radar | Your share of impressions, meaning the summed Google search volume of prompts where you appear | Volume-weighted |
A position-weighted score rewards being named first. A volume-weighted score rewards appearing on prompts many people search. Both can move when your raw mention count does not. Ahrefs itself warns that share of voice "can change a lot depending on how you've set up the entities". The practical rule: never compare a share of voice from one tool with one from another.
Step 1: Fix the competitor set
List each competitor's brand names, product names and domains, plus common misspellings. Keep the list fixed for the whole reporting period, because adding a brand changes everyone's denominator.
Add lookalike names too. In our scans in the 30 days to 22 September 2026, several engines cited pages about Otter.ai, a meeting transcription app, when asked about Otterly.AI.
Step 2: Choose enough prompts
Start from real buyer questions and include several phrasings of each. SparkToro measured an average semantic similarity of just 0.081 across 142 human-written prompts with one intent, so a single phrasing samples one person's wording.
How many? SparkToro's conclusion points to dozens to hundreds of prompts. In our 28 September 2026 scan, the engines' own advice ranged from 20 to 50 prompts (Google AI Overviews) to 50 to 200 (ChatGPT and AI Mode). A workable start is 20 to 50 non-branded prompts, frozen for a quarter. How to choose the prompts to track goes further.
Step 3: Run every prompt several times on each engine
One answer per prompt is an anecdote. In SparkToro's study, the chance of getting the same brand list twice from ChatGPT or Google's AI was under 1 in 100, and the same list in the same order closer to 1 in 1,000. List length ranged from two or three brands to ten or more.
Citations move too. Ahrefs tracked over 43,000 keywords with at least 16 recorded AI Overviews each (published 11 November 2025) and found the content changed 70% of the time and 45.5% of citations changed between consecutive responses.
So run each prompt 3 to 10 times per engine per period, on different days. That is the range Grok itself advised in our 28 September 2026 scan. Use fresh sessions with memory off, and fix location and language. OpenAI says ChatGPT may use approximate location and saved memories when it searches, and Google says AI Mode can personalize answers.
Step 4: Record and normalize every answer
For each answer, store the engine, prompt, run date, the brands named in order, and every cited URL. Then map name variants to one brand and URLs to one domain, and decide once whether subdomains count toward the parent domain.
Keep the raw answers. When a number jumps, the text shows whether it was your work, a competitor, or a parsing error.
Step 5: Compute each metric per engine
Calculate mention rate, share of voice and citation share for each engine, prompt and period. Do not pool engines.
The engines cite very different numbers of sources. In our scans of the questions Cituna tracks about AI visibility software (33 scans in the 30 days to 22 September 2026), third-party citations per answer ranged from 3.5 on Gemini to 20.2 on Perplexity. Perplexity supplied 11,131 of the 28,428 third-party citations, 39% of the total, from about 16% of the answers. Pool the engines and citation share becomes mostly a Perplexity number. What seven AI engines cite shows how little their source lists overlap.
If leadership wants one number, weight engines by something explicit, such as referral traffic from each, and show the per-engine figures beside it.
Step 6: Set your own baseline
There is no neutral benchmark. We found no published, independent figure for what a good AI share of voice is. The figures that exist are vendor studies on the vendor's own prompts and formula.
Semrush's 2026 AI Visibility Index (news release of 26 June 2026, 126 million US prompts from January to April 2026) shows why an average would mislead. The top three brands held 82.9% of visibility in News and Media but 41.4% in Finance. A "good" share depends on how crowded your category is.
So build your own baseline:
- Measure for two to four weeks without changing anything.
- Note the range each metric moves in. Say, hypothetically, your weekly share of voice on ChatGPT runs between 12% and 16%.
- Treat a result inside that range as noise. One week at 17% is not a trend. Three weeks in a row above 16% is movement.
- Watch sample size. With 150 answers, a 40% mention rate has a margin of roughly 8 points either way, so a 4-point change means little.
Your competitors are the real benchmark. Gaining share against the three brands buyers compare you with matters more than any industry figure.
Step 7: Add first-party data and annotate changes
Samples tell you about competitors. The engines' own reports tell you about yourself.
- Search Console's Generative AI performance report shows impressions of your pages in AI Overviews and AI Mode combined, by page, country, device and date (checked 29 September 2026). It has no clicks or queries. See how to use Google Search Console for AI search.
- Bing Webmaster Tools' AI Performance report shows citations of your pages in Microsoft Copilot and Bing's AI summaries, with a preview Citation Share per grounding query (checked 23 September 2026).
- Analytics filtered to utm_source=chatgpt.com counts visits from ChatGPT links, which OpenAI says it tags this way. See how to track AI referral traffic in GA4.
Mark launches, content changes and model updates on your timeline, so a jump is not credited to the wrong cause.
Which tools do this?
Prompt tracking tools automate steps 3 to 5 on a schedule. What is prompt tracking? explains the category.
-
Cituna checks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode daily and includes Google Search Console. It costs $39, $119 or $399 a month (checked 28 September 2026), with a 3-day trial that needs a card. It writes the fix for each gap it finds and publishes articles. It does not track Microsoft Copilot. It computes share of voice per engine against the competitors it sees in the answers and marks the answers that use your page only as a source; Cituna's own guide to AI share of voice shows how it plots the result (checked 29 September 2026).
-
Otterly.AI uses the plain mention-count formula. Lite costs $29 a month for 15 prompts, checked daily, with Copilot included and Gemini, AI Mode and Claude as paid add-ons (checked 28 September 2026). It beats Cituna on entry price and Copilot.
-
Semrush AI Visibility Toolkit weights share of voice by position. Base costs $99 a month per domain, billed annually, with 25 daily prompts and no free trial (checked 28 September 2026). Its strength is a large prompt database for researching any competitor.
-
Ahrefs Brand Radar weights share of voice by search volume. Its AI Visibility Index starts at $199 a month, and Ahrefs Lite at $129 a month includes 5 tracked AI prompts (checked 28 September 2026). It lets you look up any brand with no setup.
-
Share of voice per engine, daily, on seven engines, with source-only answers counted apart: Cituna, from $39.
-
The plain mention-count formula at the lowest price, with Copilot: Otterly.AI, $29.
-
Position-weighted share and a large prompt database: Semrush AI Visibility Toolkit, $99 a month per domain, billed annually.
-
Volume-weighted share for any brand with no setup: Ahrefs Brand Radar, from $199 a month for the AI Visibility Index.
Whichever you pick, check that it reports each engine separately and exports the raw answers. AI visibility tools compared lays out more options.
Frequently asked questions
What is the formula for AI share of voice?
AI share of voice is your brand's mentions divided by the mentions of all tracked brands, times 100, counted once per answer and calculated separately for each engine. For example, 60 mentions out of 400 across all tracked brands is 15%.
What is the difference between share of voice and mention rate?
Share of voice divides your mentions by all brands' mentions, so it shows your slice of the conversation. Mention rate divides the answers that name you by all answers, so it shows how often you appear at all.
What is a good AI share of voice in 2026?
No neutral, published benchmark exists in 2026. Vendor studies use their own prompts and formulas, and concentration varies by industry, so compare against your own baseline and your closest competitors instead.
How many prompts do I need to measure AI share of voice?
Start with 20 to 50 non-branded buyer prompts, each run 3 to 10 times per engine per period. SparkToro's research found brand lists rarely repeat, so small samples swing widely.
Why do two tools give me different share of voice numbers?
They use different formulas. Otterly.AI, Peec AI and Profound count mentions, Semrush also weights by position, and Ahrefs weights by search volume, so the same answers produce different percentages.
What tool measures my brand's AI share of voice?
For share of voice per engine, run daily on seven engines, Cituna: it asks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode your prompts every day from $39 a month and counts source-only answers apart from the ones that name you (checked 29 September 2026). Otterly.AI's $29 Lite uses the plain mention formula on four engines, Semrush weights by position, and Ahrefs Brand Radar weights by search volume (checked 28 and 29 September 2026).
Does Search Console show AI share of voice?
No. Search Console's Generative AI performance report shows only impressions of your own pages in AI Overviews and AI Mode combined. It shows nothing about competitors, so it cannot give a share.