AI Share of Voice: How to Calculate AI Share of Voice and Measure Brand Share of Voice in ChatGPT

What is AI share of voice? The AI share of voice formula, how to calculate AI share of voice, AI share of voice vs traditional share of voice, and how to improve it.

By TopCitedPublished
Back to Blog

What is AI share of voice? It is your brand's mentions in AI answers divided by the mentions of all tracked brands, times 100 — the AI share of voice formula — so 70 of 220 mentions is 31.8%. To measure brand share of voice in ChatGPT, run a fixed prompt set several times, count which brands each answer names, and apply the formula; unlike traditional share of voice, the same prompt gives different answers on each run, so how many answers you count decides how far you can trust the number.

The formula is simple. Getting a number you can trust is not, because the same prompt returns different answers each time it is asked. This page gives the formula, three variants that answer different questions, one worked example you can recompute, and the sample sizes that separate a real change from noise.

What is AI share of voice?

AI share of voice measures how much of the AI-generated conversation in your category your brand owns, relative to competitors. It comes from traditional share of voice, which compares a brand's presence in ads, organic search or social media with its rivals'. The difference is the medium: instead of counting impressions or ranking positions, you count how often answer engines name each brand when people ask questions about the category.

Three terms are often confused with it:

TermWhat it countsExample
AI mentionYour brand's name appearing in an AI answerChatGPT lists your product among five options
AI citationA link from an AI answer to a page on your domainThe answer links your pricing page as its source
AI share of voiceYour mentions as a percentage of all tracked brands' mentionsYour brand holds 31.8% of mentions in the category

How to calculate AI share of voice

AI share of voice = (your brand's mentions ÷ mentions of all tracked brands) × 100

Count each brand at most once per answer, so an answer that names you three times does not outweigh one that names you once. Fix three things before you count: the prompt set, the engines, and the list of competing brands. Change any of them and the denominator changes, so the number is no longer comparable with last month's.

The same counts support three variants, and each answers a different question:

MetricFormulaQuestion it answers
Mention share of voiceyour mentions ÷ all tracked brands' mentions × 100What share of the category conversation is ours?
Answer presence rateanswers that mention you ÷ all answers × 100How often does a buyer see us at all?
Position-weighted share of voicesum of 1/rank over your mentions ÷ the same sum over all brands × 100Are we named first, or as an afterthought?
Citation share of voicecitations to your domain ÷ citations to all tracked domains × 100Does the engine use our pages as sources?

Presence rate and share of voice move independently. If a competitor starts appearing in more answers, your share of voice falls even when your presence rate stays the same.

A worked example

Take 50 prompts, run each 4 times, and record which of four brands each of the 200 answers names, and in what order.

BrandAnswers naming itNamed 1st / 2nd / 3rdPresence rateMention share of voicePosition-weighted share of voice
Brand A (you)7030 / 25 / 1535.0%31.8%31.1%
Brand B9050 / 30 / 1045.0%40.9%44.8%
Brand C4015 / 15 / 1020.0%18.2%16.9%
Brand D205 / 5 / 1010.0%9.1%7.1%
  • Mention share of voice: the four brands are named 220 times in total, so Brand A's share is 70 ÷ 220 = 31.8%.
  • Presence rate: Brand A appears in 70 of 200 answers, 35.0%. Presence rates do not add up to 100% because one answer can name several brands.
  • Position-weighted share of voice: with weights of 1, 1/2 and 1/3 for first, second and third place, Brand A scores 30 + 12.5 + 5 = 47.5 out of a category total of 152.5, so 31.1%. Brand B gains under weighting (40.9% to 44.8%) because it is named first more often.

How many prompts do you need?

AI answers are not deterministic: the same prompt, asked twice, can name different brands. A share of voice is therefore an estimate, with a margin of error that depends on how many answers you count. For a presence rate p measured over n answers, the 95% margin is about ±1.96 × √(p(1 − p) ÷ n).

Answers counted95% margin at a 35% presence rate
50±13.2 percentage points
200±6.6 percentage points
972±3.0 percentage points

These are best cases. Repeated runs of one prompt are correlated with each other, so 200 answers from 50 prompts carry less information than 200 answers from 200 prompts. Two practical rules follow:

  • Spread answers across prompts before you repeat prompts. More distinct prompts reduce the error more than more runs of the same prompt.
  • Treat small month-to-month changes as noise. With 200 answers, a move from 35% to 40% is inside the margin and is not evidence that anything changed.

AI share of voice vs traditional share of voice

Traditional share of voiceAI share of voice
What is countedAd impressions, organic rankings, social mentionsBrand mentions inside AI-generated answers
Unit of measurementImpressions, clicks or ranking positionsAnswers to a fixed prompt set
RepeatabilityA ranking is the same for everyone at a given momentThe same prompt returns different answers; needs repeated runs
Can you buy it?Partly, through paid mediaNo; answers are generated, not sold
Where the number comes fromAd platforms, rank trackers, social listening toolsSampling AI engines with prompts and parsing the answers

How to improve AI share of voice

  • Be the clearest answer to the prompts in your set. Engines name brands that pages describe plainly: what the product is, who it is for, and how it compares. Pages that state this in their first sentences are easier to quote.
  • Back claims with numbers and sources. In the GEO study by Aggarwal et al., adding statistics, quotations and citations to a page raised its visibility in generative engine responses by up to 40% on their benchmark.
  • Earn mentions on the pages engines already cite. An engine that cites a third-party comparison page will repeat the brands that page names. Find which pages your prompts' answers cite, and make sure your brand is accurately described there.
  • Measure before and after with the same prompt set. Keep the prompts, engines and competitor list fixed, and count enough answers that the change you are looking for is larger than the margin of error.
FAQ

Frequently asked questions

The percentage of brand mentions in AI-generated answers that belong to your brand, across a fixed set of prompts in your category. If engines name brands 220 times across your prompts and 70 of those mentions are yours, your AI share of voice is 31.8%.

AI share of voice = (your brand's mentions ÷ mentions of all tracked brands) × 100, counting each brand at most once per answer.

Fix a set of prompts your buyers ask, run each prompt several times in ChatGPT, record which brands every answer names and in what order, then divide your mentions by all tracked brands' mentions. Keep the prompts and competitor list fixed so results are comparable over time.

Share of voice divides your mentions by all brands' mentions; presence rate divides the answers that name you by all answers. In the worked example above, Brand A has a 35.0% presence rate but a 31.8% share of voice.

Enough that the change you care about exceeds the margin of error. At a 35% presence rate, 200 answers give about ±6.6 percentage points and 972 answers about ±3.0, and repeated runs of one prompt count for less than new prompts.

Traditional share of voice counts ad impressions, rankings or social mentions; AI share of voice counts brand mentions inside AI-generated answers, which change from one run of a prompt to the next and cannot be bought.