How to Run an AI Visibility Audit, Step by Step
A practical AI visibility audit method: pick buyer questions, ask ChatGPT, Gemini and Google AI Overviews more than once, then score mentions and citations.
Read the guideWhat AI share of voice means, how to calculate it from ChatGPT, Gemini and Google AI Overviews answers, a worked example, and the traps that skew the number.
AI share of voice is the portion of brand mentions in AI-generated answers that go to your brand, compared with a set of competitors you choose. You ask ChatGPT, Gemini and Google AI Overviews a fixed list of buyer questions, count the answers that name each brand, and divide your count by the total across all the brands. If answers name you 30 times and your competitors 70 times between them, your AI share of voice is 30%.
Share of voice is one of the main numbers in an AI visibility audit. This guide covers the formula, a worked example and the mistakes that most often make the number misleading.
Share of voice is an old advertising measure: a brand’s portion of all advertising, or all media mentions, in its category. Applied to AI search, the “voice” is the written answer an assistant gives when someone asks about the category. Buyers often read that answer instead of clicking through to results; Pew Research found Google users clicked a regular result in 8% of visits when an AI summary appeared, against 15% when one did not. If the answer is where the shortlist forms, your share of the names in it matters.
AI share of voice (brand) =
answers that name the brand
÷ sum of answers that name each brand in the comparison set
× 100
Three definitions make the formula usable:
Some teams weight mentions by position or count every repetition. Both are defensible, but they make the number harder to explain and, given how much order varies between runs, position weighting adds noise. Counting answers is the simplest version that holds up.
These two numbers answer different questions, and reports should show both.
| Measure | Formula | Question it answers |
|---|---|---|
| Mention rate | Answers naming the brand ÷ answers returned | How often does the brand make it into an answer at all? |
| Share of voice | Answers naming the brand ÷ all brand mentions in the set | How much of the conversation does the brand get against these competitors? |
Mention rate is absolute; share of voice is relative. A brand can improve its mention rate while its share of voice falls, if competitors improved faster.
The numbers below are illustrative, invented for this example, not from a real audit.
An agency audits a project-management tool, “Brand A”, against four competitors. It uses 20 buyer questions, each asked twice on ChatGPT, twice on Gemini and once on Google: 100 checks in all. Eight checks return nothing (a few Gemini requests never come back, and several Google searches show no AI Overview), leaving 92 answers.
Counting the answers that name each brand:
| Brand | Answers naming it | Share of voice |
|---|---|---|
| Competitor B | 61 | 61 ÷ 180 = 33.9% |
| Competitor C | 45 | 45 ÷ 180 = 25.0% |
| Brand A (client) | 38 | 38 ÷ 180 = 21.1% |
| Competitor D | 22 | 22 ÷ 180 = 12.2% |
| Competitor E | 14 | 14 ÷ 180 = 7.8% |
| Total mentions | 180 | 100% |
The total is higher than 92 because one answer usually names several brands.
Brand A’s mention rate is 38 ÷ 92 = 41%: it appears in about four answers out of ten. Its share of voice is 21%, third of five. The two numbers together tell the story better than either alone. Brand A is reasonably visible, but Competitor B is named in two answers out of three and owns a third of the conversation.
Now watch what the comparison set does. Drop Competitor E from the list and the total falls to 166; Brand A’s share rises to 38 ÷ 166 = 22.9% without a single answer changing. Add a dominant brand to the list and the client’s share falls. That is why the list must stay fixed between audits.
One overall number hides most of the useful detail. Split it:
A text search for “Wave”, “Monday” or “Notion” will count sentences that have nothing to do with the brand. One common-word competitor can take a share of voice it never earned and push everyone else’s down. Read the matches for these brands, correct the false ones by hand, and add the brand’s other names (product names, abbreviations, the domain) so real mentions are not missed either.
With 20 answers, one answer is five percentage points of mention rate. A share of voice built on a handful of questions can swing by ten points between two audits for no reason at all. Use at least 20 buyer questions, ask them more than once, and be cautious about reading meaning into differences of a few points.
AI answers change between runs of the same question. A 2025 SparkToro and Gumshoe study, with volunteers running 12 prompts 2,961 times, found less than a 1 in 100 chance that ChatGPT or Google’s AI would return the same list of brands twice. The researchers concluded that how often a brand appears across many runs is the meaningful measure, which is what share of voice is built on, but only if there are enough runs. Ask each question at least twice on the chat assistants and report how many question-and-engine pairs gave different results, so readers know how settled the number is.
“Is Brand A any good?” will almost always produce an answer naming Brand A. Including such questions inflates both mention rate and share of voice. Ask them separately to hear what is said about the brand, and keep them out of the scores.
The denominator only includes brands in your comparison set. If the assistants keep naming a sixth brand you left out, your share of voice looks better than the real conversation. Check which other brands appear in the answers and decide whether to add them before you report.
An answer that says “Brand A is often criticised for its pricing” counts as a mention. Share of voice measures presence, not approval. Read the passages around each mention, or at least a sample, and report sentiment alongside the number.
Different questions, a different competitor list, different engines or a different location all change the number. To compare two audits, keep everything fixed except time.
AI Visibility Desk uses the method above. Each audit asks the buyer questions twice on ChatGPT, twice on Gemini and once on Google, then counts, for the client and up to five listed competitors, the answers that name each brand. Each brand counts once per answer, checks with no answer are left out, and the two branded questions every audit asks are reported separately and never scored.
The review screen shows each answer with a short passage so wrong matches can be corrected, lists other brands the assistants named so you can add or hide them, and reports how many question-and-engine pairs disagreed between runs. Share of voice sits in the report next to mention rate per engine, average position, head-to-head results, mention rate by question type and the most cited sources. With monitoring on, each re-run adds a point to the trend.
If you run AI visibility work for clients, the agency page explains the white-label report, and the pricing page lists the plans, including a free audit to try.
AI share of voice is a brand's portion of all brand mentions in AI-generated answers to a set of category questions, compared with a defined group of competitors. If answers name your brand 30 times and your competitors 70 times in total, your AI share of voice is 30%.
Ask a fixed set of buyer questions on each AI engine, count the answers that name each brand, add those counts across your brand and its listed competitors, then divide each brand's count by the total. Count a brand once per answer, however many times that answer repeats its name.
Mention rate is the share of answers that name your brand, measured on its own. Share of voice compares your mentions with competitors' mentions. A brand can appear in 40% of answers yet hold only 20% share of voice if competitors are named even more often.
Use at least 20 buyer questions, asked more than once on engines whose answers vary. With very few answers, a single answer moves the result by several points. Keep the questions and competitor list fixed between measurements, or the numbers cannot be compared.
A practical AI visibility audit method: pick buyer questions, ask ChatGPT, Gemini and Google AI Overviews more than once, then score mentions and citations.
Read the guideHow ChatGPT, Gemini and Google AI Overviews decide which brands to name and which sources to cite, and what actually moves AI search visibility for a brand.
Read the guideAnswer engine optimization (AEO) makes your content the answer AI assistants and Google give to buyer questions. What it is, how it works and how to measure it.
Read the guideEnter a client’s website, let the AI fill in competitors and buyer questions, and get a branded report across six AI engines. No card needed.
Start free