AI Share of Voice: What It Is and How to Measure It
By Vouch8 min read
Picture a buyer who asks ChatGPT to recommend something in your category, weighs three names, pushes back on price and settles on one. Whether your brand was part of that exchange, and how often compared with your rivals, is what AI share of voice measures.
This post is about that one number: how it differs from the advertising metric of the same name, how to calculate it so the result means something, and what to do once you have it. For the wider set of visibility metrics, start with AI search visibility; for the hands-on logging method, see how to track brand mentions in AI search.
From advertising weight to answer presence
In media planning, share of voice is your advertising weight as a percentage of the category's: your spend or impressions divided by everyone's. It worked because ad slots could be counted.
Buying research has moved onto a surface that is harder to count. Google describes AI Overviews and AI Mode as features of Search in its AI features guide, and the results page increasingly is the destination. SparkToro's 2026 clickstream study, built on Similarweb's US panel, put the share of Google searches that ended without any click at 68.01 percent for January to April 2026; zero-click search covers the study and its caveats. The data does not prove that AI answers caused the trend, but it does show where buyers now stop.
AI share of voice carries the old idea onto that surface. A definition you can apply consistently:
- Fix a set of buyer prompts and a set of assistants.
- Run them on a schedule.
- In each answer, count each tracked brand once, however many times it is named and wherever it sits.
- Divide your count by the total count for every tracked brand across the same answers.
The glossary entry for share of voice gives the short version. Its partner is mention rate: the share of answers that name you at all. Mention rate is absolute, measured against your own prompt set; share of voice is relative, measured against the rivals you chose to track.
Why the counting rule matters more than the formula
Any reasonable formula works if it stays fixed. What breaks share of voice is drift: a prompt reworded between runs, an assistant added halfway through the quarter, or one run counting repeated mentions while the next does not. Each of those moves the number without anything changing in the market.
Three habits keep the figure honest:
- Sample repeatedly. Assistants write each answer afresh, so the same question can name different brands an hour later. One run is an anecdote.
- Report each assistant separately as well as combined. ChatGPT, Perplexity, Gemini and Google's AI features retrieve from different sources, and a combined figure can hide a collapse on one of them.
- Look past the opening answer. Buyers ask follow-ups about price, reliability and alternatives. A brand that dominates the first answer and vanishes by the third question has a flattering share of voice and a weak position.
A measurement routine you can run
This is a recommended routine, not an industry standard. Its value comes from running it the same way every time.
Choose prompts that mirror real buying
Write 15 to 30 short scripts. Each opens with a discovery question and continues through the follow-ups a real buyer would ask: comparing the leaders, raising a doubt, checking reliability, questioning value and asking for a final pick. Use the words customers use in sales calls and reviews, not your positioning statement. The buyer conversation prompts are a ready starting set.
Pick assistants and a cadence, then freeze both
Cover the assistants your buyers use. For most categories that means ChatGPT, Gemini, Perplexity and Copilot, plus Google's AI Overviews and AI Mode. Weekly is a sensible starting cadence for a manual routine; daily runs earn their cost when you are shipping changes and want to see their effect sooner. When you do change the prompt set, do it in one deliberate revision and mark the break on your chart.
Record enough to explain the number
For each answer, keep the date, the assistant, the prompt, its stage, the full answer text and the sources cited. From those records you can derive four figures alongside share of voice:
| Figure | The question it answers |
|---|---|
| Mention rate | How often are you named at all? |
| Citation rate | How often is your own site a source? |
| Stage survival | Which follow-up question loses you? |
| Final-pick rate | How often are you the recommendation at the end? |
Share of voice tells you where you stand against rivals. The other four tell you why.
Benchmark against the same rivals every time
The denominator is the set of brands you track, so choose two or three real competitors and keep them. Scoring them in the same runs also separates your own movement from a shift across the whole category, such as a model update that reshuffles every answer.
A worked example
Take two invented bike brands, Demo Cycles and Rival Bikes, followed through 20 buyer conversations. In the opening answers Demo Cycles is named 9 times and Rival Bikes 13 times. That gives Demo a 45 percent mention rate against Rival's 65 percent, and an opening-answer share of voice of 9 out of 22, roughly 41 percent.
The later stages change the picture. When buyers ask about warranty and returns, Demo is still named in 3 of its 9 conversations, while Rival holds on in 9 of its 13. When the buyer asks for a final pick, Rival is chosen six times and Demo once.
Share of voice says Demo is behind but competitive. The stage records say the real gap opens at the trust question. The records still do not say why, so the next job is reading those answers: which sources the assistants quote about Rival, what they say about Demo, and whether Demo publishes its warranty and returns terms anywhere an assistant can find them.
What tends to move the number
Nobody outside the companies that build these assistants knows how their answers are put together, so treat the following as hypotheses to test against your baseline.
- Reachable text. If AI crawlers are blocked, or key facts live only in images and downloads, assistants have less of your own material to draw on. You can test your robots.txt with the free AI crawler checker.
- Facts for the hard questions. Prices, comparisons, warranty terms and integration lists answer the objection and trust stages, which is where share of voice often leaks.
- Coverage elsewhere. Reviews, forums and comparison articles give assistants something to say about you beyond your own pages.
- One consistent story. When your site, listings and reviews agree on prices and policies, there is less contradictory material for an answer to repeat.
Judge any change over several weeks of runs, not from one good morning.
Turning share of voice into a plan
A falling share of voice is a prompt to look, not a diagnosis. Find the stage where you lose ground, read the answers at that stage, and compare the sources quoted for you with those quoted for the rival who replaces you. If your citation rate is low, check access and whether the facts you want repeated exist on your site at all. Then report a monthly trend of share of voice and final-pick rate against the same two rivals; that gives leadership a line to watch and your content work a before and after.
How Vouch measures it
Every day, Vouch asks your tracked prompts on ChatGPT, Claude, Gemini, Perplexity, Copilot, Google AI Overviews and Google AI Mode, seven assistants in all. Its dashboard shows share of voice for the week, your share of the mentions among the brands you track, next to a ranking of those brands.
For the follow-up stages, Vouch runs Threads: six-question buyer conversations through Discovery, Shortlist, Objection, Trust, Value and Decision, three times a week on ChatGPT, Claude, Gemini and Perplexity. Each conversation is scored on the stages the brand survives and whether it is chosen at the end, and the week's average becomes the Vouch Rate, from 0 to 100. Sentiment adjusts each conversation's score by at most a tenth, which is never more than 10 points on that scale. The methodology shows the formula. Share of voice tells you how often you are present; the Vouch Rate tells you how far into the conversation you last.
Plans cost between $39 and $599 a month. Vouch is currently invite-only, every invite comes with a 14-day free trial, and the waitlist is open.
