AI share of voice: how often AI names you vs competitors
Short answer AI share of voice is the share of all business mentions in AI answers that name you. Ask the same 20 to 30 customer questions in each assistant, two or three times each, and list every business named. Divide your mentions by the total. Report it next to how often you appear at all, and ignore small swings.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callThe formula: three numbers from one question set
Ask a fixed list of customer questions, write down every business each answer names, then work out three numbers. Share of voice is the headline; the other two explain why it moved.
| Number | Formula | What it tells you | Example |
|---|---|---|---|
| Mention rate | Answers that name you ÷ all answers | How often you show up at all | 12 of 60 = 20% |
| Share of voice | Times you're named ÷ times any business is named | Your slice of all recommendations | 12 of 240 = 5% |
| Citation share | Links to your site or listings ÷ all links shown | How often AI uses your pages as a source | 9 of 300 = 3% |
The bottom of the fraction decides the result. If every answer names you plus four rivals, you appear 100% of the time yet hold a 20% share of voice. So write your counting rules down once and never change them:
- One mention per business per answer, however often the answer repeats the name.
- Count every business the answer names, not only the rivals you had in mind. The AI decides who you compete with.
- Directories and review sites (Yelp, Angi, a Google listing) go in the citation count, not the business count.
- Answers that name nobody still count as answers. They lower your mention rate, not your share of voice. They aren't rare: 10% of ChatGPT's local answers and 11% of Google's named no business in BrightLocal's 2026 study.
- Ignore position. Being listed first doesn't score extra (see why below).
Which questions to ask
Ask what customers type, in their words, with the city in every question. Then freeze the list: change one question and the score changes with it.
- 20 to 30 questions. Mix "best [service] in [city]", specific services, specific situations, and price or comparison questions. The 20-question ChatGPT test shows the mix with a plumber's list.
- Put the location in the question. For local results, OpenAI's help center advises: "Include your city, neighborhood, or postal code in your question." Otherwise ChatGPT guesses your area from your IP address, and saved memories can change what it searches for.
- Use a clean session: a temporary or logged-out chat, so your own history doesn't steer the answer.
- Keep new questions in a separate list, out of the trend.
- Don't weight questions. If the job that pays best matters most, report its questions as their own group. A weighted total hides which question moved.
Score each AI assistant separately
Each assistant reads different sources, so a blended number hides where you're weak. Google grounds the Gemini app in its search index, Microsoft's Copilot fetches from Bing, and ChatGPT runs its own search plus partner search providers (which source feeds which assistant). Even Google's two AI features differ: Google says AI Mode and AI Overviews can run on different models, and the links they show vary.
Track the two to four assistants your customers use, each in its own row.
How many times to ask each question
At least twice, ideally three times. AI answers change from run to run, so only totals mean anything.
- BrightLocal's 2026 study tracked ChatGPT, AI Mode and AI Overviews for 1,342 locations over 60 days. Repeat searches shared only 20–33% of their results, and a business seen once came back in about half of later tries.
- SparkToro and Gumshoe had 600 volunteers run 12 prompts 2,961 times through ChatGPT, Claude and Google's AI. The same list of brands came back less than once in 100 runs, and the same order less than once in 1,000. Their verdict on what still works:
"visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric" (SparkToro, January 2026)
That's why position doesn't count: it's noise.
How big a change is real?
Small samples swing. The table uses the standard margin of error for a percentage (NIST's statistics handbook), at a 50% mention rate, where it's widest.
| Answers per assistant | One answer moves mention rate by | Treat a change as noise if under about |
|---|---|---|
| 40 (20 questions × 2) | 2.5 points | 15 points |
| 60 (20 questions × 3) | 1.7 points | 13 points |
| 90 (30 questions × 3) | 1.1 points | 10 points |
| 200 | 0.5 points | 7 points |
For share of voice, look at the count behind the percentage. Going from 12 to 14 mentions is two answers, not a trend.
The scoring sheet
One row per answer.
| Date | Assistant | Question | Run | Businesses named | You? | Links to you / all links |
|---|---|---|---|---|---|---|
| Nov 3 | ChatGPT | Best roofer in Tulsa | 1 | Rival A, Rival B, you, Rival C | Yes | 1 / 5 |
| Nov 3 | AI Mode | Roof leak repair Tulsa, older house | 1 | Rival A, Rival D | No | 0 / 4 |
Per assistant, total six columns: answers, answers naming you, your mentions, all mentions, your links, all links. The three formulas come straight from those.
What Google, Bing and ChatGPT report, and what they leave out
Their free reports count you in real customers' questions, but none names a competitor. Only Bing shows a bottom of the fraction, and only for links.
| Report | What it counts | Shows rivals? |
|---|---|---|
| Search Console: Generative AI performance (all sites since August 31, 2026) | Impressions of your pages in AI Overviews, AI Mode and Discover's AI features | No |
| Bing Webmaster Tools: AI Performance (public preview since February 2026) | Citations of your pages in Copilot and Bing's AI summaries, plus sample queries; since June 2026, your share of all citations per query | Totals only, no names |
| ChatGPT Business Hub insights (verified claimed listings) | Your mentions and website clicks over the past 28 days | No |
Local businesses can claim only existing US listings in ChatGPT, so a Canadian shop can't see ChatGPT's count at all. OpenAI also warns "Counts may be incomplete." Use these reports to check your hand count's direction: if Search Console's AI impressions climb while your AI Mode mention rate falls, look again.
If you pay a tracker for a share-of-voice figure, ask how it counts: answers or mentions as the denominator, which rivals, how many runs per question, and from which city.
How to read a change
| What you see | What it likely means |
|---|---|
| Mention rate up, share of voice flat | Answers name more businesses overall, you included |
| Share of voice up, mention rate flat | Answers name fewer businesses: the same appearances in a smaller crowd |
| Both up, beyond the noise | Real gain; find which questions moved |
| Citation share up, mentions flat | AI reads your pages but recommends others; pages may explain topics without selling your service |
| One assistant drops, the rest hold | That assistant's source changed: a listing, a blocked crawler |
No visibility number proves a lead came from it. Put it beside calls and bookings, and ask every new customer how they heard about you.
Example
Illustrative cases. Numbers are invented for the example.
A Tulsa roofer
20 questions, twice each, in ChatGPT and AI Mode. ChatGPT names him in 10 of 40 answers, 10 of 160 mentions: a 25% mention rate and 6% share of voice. His top rival holds 19%. AI Mode names him in 4 of 40. Reading: the bigger gap is on Google, so his Google Business Profile comes first.
A Calgary dentist
Mention rate went from 30% to 40% on 40 answers, the month her implant page went live. That's 4 more answers, under the 15-point noise line. Reading: don't credit the page yet; rerun next month.
An online bookkeeping firm
Named in 30 of 60 answers both months. Share of voice fell from 17% to 10% because answers went from about 3 firms each (180 mentions) to 5 (300). Reading: it didn't lose ground; the crowd grew.
FAQ
What is a good AI share of voice?
There's no benchmark; it depends on how many businesses each answer names. ChatGPT names about 4 per local answer (BrightLocal, 2026), so matching your strongest rival is a fair target.
Is AI share of voice the same as ranking in AI?
No. AI answers almost never repeat the same order, so a "rank" is noise. How often you're named across many runs is what holds up.
How often should I measure?
Monthly, with the same questions and assistants. At a small business's sample size, weekly counts mostly measure noise.
Can I see competitors' AI mentions in Search Console or Bing?
No. Both cover only your own site; Bing shows your citation share per query but never names rivals. Count by hand or pay a tracker.
What does 25% share of voice mean?
One in four business mentions across your questions named you. It says nothing about how many real customers saw those answers.
Different for you?
- Measuring ChatGPT for the first time: the 20-question test.
- Your score is low and you want the fixes: how to get ChatGPT to recommend your business.
- One assistant lags the rest: Perplexity, Gemini and Claude answers.
- You want the numbers that show leads, not mentions: is my SEO working?
- Someone is selling you GEO with a share-of-voice chart: is GEO worth paying for?
- You'd rather have it tracked every month: SEO & GEO Agent.