AI share of voice is the share of AI answers in your market that name your brand, compared with your competitors. If you and four competitors are named 50 times in total across a set of prompts, and 10 of those mentions are yours, your AI share of voice is 20%. It is the clearest single measure of where you stand in ChatGPT, Gemini, Perplexity and Google's AI answers.
This guide sets out a repeatable method: how to choose prompts and competitors, how many times to run each prompt, how to score answers so a recommendation counts for more than a passing mention, and how to turn it into a share of voice audit you can repeat each quarter. For the wider set of measures it sits within, see our guide to AI visibility metrics.
What is AI share of voice?
AI share of voice is the proportion of brand mentions within AI-generated answers that belong to your brand, measured across a fixed set of prompts and a fixed set of competitors. It adapts the traditional SEO share of voice idea, which measures your share of search visibility for a keyword set, to answers that list or recommend businesses.
The difference from mention rate matters. Mention rate tells you how often you appear. Share of voice tells you how much of the conversation you hold. A mention rate can rise while your share falls, if competitors are rising faster.
How do you build the prompt set?
Group prompts by the intent behind them, because AI tools answer each kind differently and your share will vary between them.
| Intent | Example prompt | What it reveals |
|---|---|---|
| Recommendation | What are the best accounting firms for small businesses in Maryland? | Who the AI puts on the shortlist |
| Need-based | I run a restaurant and need help with payroll and taxes, who should I talk to? | Whether you are matched to specific needs |
| Problem | How do I fix a payroll tax mistake I made last quarter? | Whether your content is used as a source |
| Comparison | Is [Competitor] or [Your Firm] better for a small restaurant? | How you are positioned head to head |
| Category | What should I look for when hiring an accountant? | Whether you are named in general advice |
Twenty to forty prompts is a workable set for most businesses, spread across these intents. Leave pure brand prompts such as "is [your firm] any good" out of the share of voice calculation, since they name you by design. Track those separately for accuracy.
How do you choose the competitor set?
Pick four to eight competitors and keep them fixed for the year. Include the businesses you lose deals to, not just the ones you think of as rivals. Then run your prompts once before fixing the list and note who the AI tools actually name. These are often different: a directory, a national chain or a marketplace may appear more often than any local competitor, and it belongs in the set because buyers see it.
Record any other brand that appears as "other". If "other" grows over time, someone new is entering the answers and may belong in the set next year.
How many runs per prompt do you need?
At least three runs per prompt per platform, each in a fresh chat. AI answers vary between runs, so a single run can swing your share by several points by chance alone. Running each prompt three times across four platforms turns a 30-prompt set into 360 answers per round, which is enough to see a real change from one quarter to the next.
Keep the setup consistent between rounds: the same platforms, the same logged-in or logged-out state, the same location settings, and roughly the same time of month. Note whether each answer searched the web, since answers with live search behave differently from those drawn only on the model's training.
How should you score each answer?
Counting mentions alone treats "also worth considering" the same as "the best choice is". A simple weighted score fixes that. Score every brand in every answer using one of three levels:
| Level | Definition | Suggested weight |
|---|---|---|
| Mentioned | Named anywhere in the answer, including a list | 1 |
| Recommended | Named as a top choice or the answer to the question | 2 |
| Cited | Recommended or mentioned with a link to the brand's own website as a source | add 1 |
The weights are a convention, not a standard. What matters is that you use the same ones every round. Share of voice is then each brand's total score divided by the total score of all brands. Report the plain unweighted share alongside it, so anyone reading the audit can see both.
What does a worked example look like?
The following numbers are hypothetical, for illustration only. A regional accounting firm runs 30 prompts three times each on ChatGPT, Gemini, Perplexity and Google AI Mode, for 360 answers. Across those answers the scores add up as follows.
| Brand (example) | Mentions | Recommended | Cited | Weighted score | Weighted share |
|---|---|---|---|---|---|
| Your firm | 90 | 30 | 20 | 140 | 20% |
| Competitor A | 160 | 70 | 40 | 270 | 39% |
| Competitor B | 100 | 40 | 10 | 150 | 22% |
| Directory site | 80 | 10 | 30 | 120 | 17% |
| Other | 15 | 0 | 0 | 15 | 2% |
| Total | 445 | 150 | 100 | 695 | 100% |
In this example, the weighted score for each brand is its mentions plus one extra point per recommendation plus one per citation. The firm holds a 20% weighted share against Competitor A's 39%. Its unweighted share is 90 of 445 mentions, also about 20%. The useful finding is in the detail: if Competitor A wins most of the recommendation prompts while the firm appears mostly in category prompts, the gap is in how clearly the firm is positioned for specific needs, not in whether AI tools know it exists.
Note - Break share of voice down by platform and by intent before drawing conclusions. A brand can lead on Perplexity and barely register on Gemini, and the fix for each is different.
What goes into an SEO share of voice audit?
A full share of voice audit puts AI answers next to traditional search. For the same topics, record your share of organic visibility for a keyword set, using the visibility or share of voice report in your rank tracker, alongside your AI share of voice. Then compare the cited sources AI answers rely on with the pages that rank. Where they differ, the AI answers are drawing on third-party pages you do not yet appear on.
- AI share of voice, weighted and unweighted, by platform and intent
- Organic share of voice for the matching keyword set
- The ten most cited source URLs across all answers, and whether you appear on them
- Accuracy issues in how AI tools describe you
- Three to five actions, ranked by expected effect
How often should you measure it?
Quarterly is right for a full share of voice audit. Mention tracking can run monthly in between, using the routine in our guide to tracking ChatGPT brand mentions. Share of voice moves slowly, and measuring it weekly mostly measures noise.
The short version
Measure AI share of voice with a fixed set of prompts grouped by intent, a fixed set of competitors, at least three runs per prompt per platform, and a consistent score for mentioned, recommended and cited. Report it by platform and intent, quarterly. Share of voice shows where you stand; to connect it to revenue, see how to measure the ROI of AI SEO. If you want the audit run for you and the gaps closed, that is part of our AI visibility optimization service.
