How to Measure AI Share of Voice: Methods, Tools and Benchmarks (2026)

Key takeaways

• AI share of voice is your brand's mentions (or citations) divided by every tracked brand's mentions in the same answers, multiplied by 100.
• Work it out separately for mentions and citations, and separately per engine and per prompt group, rather than blending everything into one number.
• A position-weighted variant counts a brand named first for more than a brand named fifth, using a simple weight such as 1 divided by position.
• No public industry benchmark exists for this number in 2026, so compare your own score against your top 3 named competitors and against last month.
• The manual version works with a spreadsheet; a monitoring tool earns its place once your prompt and engine list outgrows what you can run by hand.

If you would rather have the gaps behind a low share of voice found and the fix written for you, start a 3-day trial of Cituna at cituna.com; a card is required.

This guide sets out the formula for AI share of voice, works it through with clearly labelled hypothetical numbers, and shows a position-weighted variant, before covering how to set your own benchmark. For the wider idea of what visibility in AI answers means before you put a number on it, see what AI visibility is. If you have not yet run a manual check, how to track your brand's visibility in ChatGPT covers building the prompt list this formula runs on.

Share of voice comes in more than one flavour. Each answers a slightly different question, so pick the one that matches your goal before you start counting.

Mention share counts every time your brand is named. Citation share counts only the answers that link to, or credit as a source, a page you control. A position-weighted share counts a mention named first for more than one buried at the end of a long list.

Work out each of these separately per engine, since a score blended across ChatGPT, Perplexity and Gemini hides which one actually has a problem. Work them out separately per prompt group too, since a category question and a named comparison question rarely behave the same way, and blending them together hides which type of question you are actually losing.

How do you measure AI share of voice, step by step?

1. List the prompts and the brands you will track

Start with the buyer questions your prompt list already contains for manual tracking: category questions, shortlist questions, and named comparisons with a competitor. Alongside your own brand, list every competitor you want the score to account for, since share of voice only means something relative to a named set of rivals.

Keep this list fixed for a stretch of weeks so your trend line is comparing like with like. Add a new prompt or a new competitor as a deliberate change you can point to later, not a quiet edit.

2. Run each prompt across every engine you track

Run every prompt in a fresh conversation on each engine in your set, saving the full answer text rather than a yes or no. Do this on the same day, or as close to it as you can manage, so a single day's news or a model update does not skew one engine's answers relative to another's.

Repeat on a fixed schedule, weekly at a minimum, so you have more than one data point before you read anything into a change.

3. Log each answer as a mention, a citation or neither

For every answer, record whether your brand was named at all, and separately whether the answer linked to, or explicitly credited, a page you control as a source. Do the same for every competitor on your list, in the same answer.

An answer can mention your brand with no citation, cite your site without naming your brand outright, or do both. Log these as separate columns, because collapsing them into one score loses the distinction the next two steps depend on.

Note the position of your brand within the answer as well, whether it is named first, buried in a longer list, or mentioned only after a follow-up question. That column feeds the position-weighted variant in step 6, so it is easier to capture it now than to go back through saved answers later.

4. Calculate your mention share of voice

The formula is: your brand's mentions divided by all tracked brands' mentions in the same answers, multiplied by 100.

Say you ran 20 prompts across 5 engines in a week, for 100 total answers. Across those 100 answers, every brand on your list was mentioned 340 times in total, and your brand accounted for 68 of those mentions. Your mention share of voice is 68 divided by 340, times 100, which is 20 percent.

5. Calculate your citation share of voice

Use the same formula on the citation column instead of the mention column: your brand's citations divided by all tracked brands' citations in the same answers, multiplied by 100.

Using the same hypothetical week, say those 100 answers contained 54 citations in total across every tracked brand, and 9 of them pointed to your site. Your citation share of voice is 9 divided by 54, times 100, which is about 17 percent. Compare that against the 20 percent mention share from step 4 and you can see your brand gets named more often than it gets cited as a source, which points at a different fix than a low mention share would.

6. Weight each mention by its position in the answer

A brand named first in a list usually carries more weight than one named fifth, so a simple weight of 1 divided by the mention's position is worth adding once your unweighted scores are close. Add up each brand's weighted mentions across the same set of answers, then apply the same share formula to the weighted totals.

Say your brand was named first in 10 of those 100 answers (weight 1 each, for 10), second in 20 answers (weight 0.5 each, for 10), and third in 38 answers (weight roughly 0.33 each, for about 12.7), giving a weighted total of about 32.7. If every tracked brand's weighted total across the same 100 answers comes to 150, your position-weighted share of voice is 32.7 divided by 150, times 100, or about 22 percent, a touch higher than the 20 percent unweighted mention share because your mentions tend to land near the top of the answer.

7. Set your own benchmark instead of an industry average

No vendor or research body publishes a public industry average for AI share of voice in 2026, and a number picked up from a blog post or a sales page will not be comparable to your own prompts, engines and competitor set. Set your own benchmark instead.

Compare against your top 3 named competitors in the same answers, since that tells you who to study when they win a question you should be winning. Compare against your own score from last month or last quarter, since a rising or falling trend on your own prompts is more useful than a single snapshot. Keep the comparison per engine, since a 20 percent share on ChatGPT and a 5 percent share on Perplexity are two separate problems, not one blended figure.

Which tools measure AI share of voice for you?

Running this by hand works, but a handful of prompts across several engines quickly turns into more answers than one person can read and log every week. The tools below automate the tracking; facts and prices below were checked on 29 September 2026.

Cituna checks 7 AI engines every day: ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode. It does not track Microsoft Copilot. Its three plans cost $39 a month, $119 a month or $399 a month, and past the tracking it writes the fix for each gap it finds and publishes the articles, with Google Search Console and MCP included. If you want a tool that turns a low share of voice into a written fix rather than another report, see Cituna's plans and trial at cituna.com; the trial runs 3 days and a card is required.

Peec AI's Starter plan, from $95 a month (checked 29 September 2026), tracks 3 models chosen from a pool of 6 and reports a citation share figure alongside gap analysis of sources that name your competitors but not you, according to Peec AI's pricing page.

SE Visible's Basic plan, from $99 a month (checked 29 September 2026), tracks 5 engines daily, including ChatGPT and Google AI Overviews, and its API can pull a Share of Voice figure and average position for each tracked prompt, per SE Visible's pricing page.

Otterly.AI's Lite plan, $29 a month (checked 29 September 2026), tracks ChatGPT, Google AI Overviews, Perplexity and Microsoft Copilot daily and publishes a figure it calls a Brand Visibility Index alongside domain ranking, according to Otterly.AI's pricing page.

AthenaHQ's Starter plan, $295 a month (checked 29 September 2026), checks 11 AI models a day and pairs prompt and response analysis with competitor insight reporting, according to AthenaHQ's pricing page. Deciding whether any paid tool is worth it yet for your team is covered in whether to pay for an AI visibility tracking tool, and the criteria to weigh them against are set out in how to choose an AI visibility tool.

What should you do with your share of voice number?

Treat the first score as a baseline, not a verdict. Rerun the same prompts on the same schedule for a few weeks before you decide whether a change in your content moved the number, since a single week's answers can swing on wording and model updates alone.

Read the answers behind a low score before you act on it. A low mention share paired with a healthy citation share suggests your pages are trusted as sources but rarely named outright, which points at a different fix than the reverse pattern. If clicks matter to you as much as mentions, how a tool should connect to Google Search Console covers reading the two side by side.

Once you have a number you trust, in 2026 the fastest way to close a gap is still to find out exactly which prompts and engines are behind it and fix the page or the source that answer relies on. See what a Cituna trial finds for your brand at cituna.com; it checks the 7 engines above daily and writes the fix for each gap over a 3-day trial that needs a card.