Your AI share of voice is the proportion of answers that cite or name you when someone asks an AI the questions your customers ask. To measure it: freeze a list of typical questions, count how often each brand appears in the answers, then run the calculation again every month. A business cited in one answer out of four has a 25% share of voice. Most small businesses never look at this number, even though it increasingly decides who gets the call.
A number that traffic does not show
When an AI answers, most people no longer click. A Pew Research Center study found that only 8% of users click a traditional link when an AI summary appears, versus 15% without one (Pew Research, 2025). Tracking only your AI traffic, the visits, therefore shows only a small part of the story: you can be the recommendation without getting a single click.
Share of voice measures something else: your presence inside the answer itself. It also differs from an AI visibility audit, which is a snapshot at one moment. Share of voice is a number you follow over time, like a poll you repeat. And the stakes are real: Google's AI Overviews already reach 2 billion monthly users, and half of the links ChatGPT cites point to business and service sites (Semrush, 2026). The answer layer is exactly where a small business gets chosen.
Why a single measurement proves nothing
AI models are not deterministic. Ask the same question twice and you will often get different sources, sometimes different brands. On top of that, there is month-to-month drift: models update, the web moves, citations change. So a single run, a single screenshot, proves nothing at all.
The takeaway is simple. Measure like a poll, not like a photo: a frozen list of questions, several runs per question, and a regular cadence. That discipline turns an impression ("it feels like ChatGPT never cites me") into a number you can defend.
A five-point protocol
Here is the method you can run yourself, with no particular budget:
- Freeze 20 to 30 real customer questions. Not your brand name: generic queries such as "best X for Y", "how to do Z", "who do you recommend for W". These brandless questions are the ones that make your business appear, or not.
- List the brands to track. Yours and four to seven competitors. Without a comparison baseline, a percentage on its own means nothing.
- Ask each question several times. At least three times, on the same day, on each engine that matters to you: ChatGPT, Perplexity, Google's AI overview. For each answer, note who is named and who is cited as a source.
- Calculate the share of voice. Number of answers where you appear, divided by the total number of answers, engine by engine. Keep "named in the text" and "cited as a source" separate: they are not the same thing.
- Log it again every month with exactly the same questions. It is the trend, not the one-day number, that tells you whether you are progressing.
Each engine rewards something different
Why measure engine by engine, rather than one global score? Because an analysis of 118,000 AI answers found that only 11% of cited domains appear on more than one platform (Whitehat SEO / Qwairy, 2026). Being visible in Perplexity says almost nothing about your presence in ChatGPT. Each engine draws from a different pool of sources and rewards different signals.
| Engine | Sources cited per answer | What it rewards | What it means for a small business |
|---|---|---|---|
| ChatGPT | Around 8 (Bing index) | Business and service sites (half of the links cited), clear question-and-answer content | Submit your sitemap to Bing, sharpen your service pages |
| Perplexity | Around 22 | Freshness, primary sources, cited numbers | Update content often and date it |
| Google AI overview | Variable (2 billion monthly users) | Informational intent, authority, consistent local profile | Publish content that explains, keep one consistent identity everywhere |
The consequence is counter-intuitive: a brand can dominate one engine and be nearly absent from another. A single, global "AI visibility score" hides that reality. Measure each engine that matters to your customers, separately.
Moving the number: target the citation gaps
Once the measurement is in place, improving it is not a rebuild of your whole site. It is a targeted hunt for citation gaps:
- Spot the questions where you are absent (share of voice at zero on that query).
- Look at what the engine cited instead. Often the same pages keep coming back: a local directory, a "best of" article, a review site, a well-answered FAQ.
- Get into those exact sources. Earn a listing in that directory, a mention in that article, a review on that platform, or publish the missing answer better than the page currently cited.
- Measure again next month on the same questions.
As an illustration, suppose a wellness studio in Greater Vancouver tracks 25 questions. On ChatGPT it appears in 2 answers out of 25, an 8% share. Looking at the answers where it is absent, it sees two sources recur: a local directory and a "best studios" article. It earns a listing in the directory and a mention in the article, then measures again. Nothing magic: it closed two specific citation gaps. Going from absent to cited on three or four high-intent questions moves share of voice far more than a month of generic blog posts.
Frequently asked questions
AI share of voice vs AI traffic: what is the difference?
AI traffic counts the visits that reach your website from ChatGPT or Perplexity. Share of voice counts your presence inside the answer itself, even when nobody clicks. Because most answers generate no click at all, share of voice captures an influence that traffic alone cannot see.
How many queries do you need for a reliable measurement?
Use 20 to 30 real customer questions, asked at least three times each on every engine, on the same day. Below that, the natural variation in answers (two runs return different sources) makes the number too unstable to track over time.
Can you track your share of voice without a paid tool?
Yes. A fixed list of questions, a few runs per engine and a spreadsheet are enough to set a first measurement and its trend. Dedicated tools automate the counting and add engines, but they do not replace choosing the right questions.
Want to know the place AI gives you, month after month?
Book a discovery call