The short version
- AI share of voice is your slice of all brand mentions across a set of category prompts. It is how rank-by-comparison works in AI answers.
- Calculate it from unbranded, buyer-intent prompts only. Branded prompts inflate it and make it meaningless.
- Build it from mention rates over a rolling window, and read it per engine before averaging.
- The most useful output is not your score, it is who gets cited instead of you, and which sources put them there.
Visibility tells you whether an AI mentions you. Share of voice tells you whether it mentions you instead of your competitors. In a world where a buyer often reads a single AI answer and stops, that relative position is the whole game. If the answer to "best [category] tool" names three rivals and not you, your visibility might still look fine on other prompts, but you are losing the comparison that decides the deal.
This is how to measure AI share of voice in a way you can actually trust, and the mistakes that turn it into a vanity number.
What share of voice means in AI answers
Share of voice is the proportion of brand mentions across a category that belong to you. If you track a set of buyer-intent prompts and, across all the answers, five brands get named, your share of voice is your mentions divided by the total brand mentions. One in four mentions is roughly 25 percent share.
It answers a sharper question than visibility: when an engine has to choose which brands to put in front of a buyer, how often does it choose you?
How to calculate it correctly
- Define a clean prompt set. Use unbranded, buyer-intent questions: category, comparison, alternatives and job-to-be-done prompts. These are the questions where the engine actually decides who to name.
- Name your competitor set. Decide which brands count, so a mention is attributed consistently. Watch for name collisions and aliases.
- Measure mention rates, not single hits. Run each prompt on each engine repeatedly and use the mention rate, because a single answer is non-deterministic noise.
- Compute the share. Your share of voice is your mentions divided by all brand mentions across the same prompt set and rolling window.
- Read it per engine first. You can lead share of voice on one engine and trail badly on another. Averaging too early hides exactly the gap you need to fix.
The mistake that quietly inflates the number
The fastest way to ruin share of voice is to let branded prompts into the set. A prompt that names you in the question ("is [brand] worth it") mentions you almost every time, which pads your mention count and lifts your share for no real reason. The metric is only honest when every brand in it has to earn its mention from an unbranded question. This is the same flaw that breaks visibility scores, covered in why we keep branded prompts out of the headline.
Two more traps to avoid:
- Sampling noise. Share computed from one scan per prompt swings day to day. Use a rolling window and treat the figure as an estimate with a confidence range.
- Unbalanced prompt sets. If half your prompts are about one narrow sub-topic where you happen to win, your overall share is flattered. Spread prompts across the real buying journey.
The output that actually matters
Your share number is a scoreboard. The decisions come from what sits underneath it:
- Who is named instead of you. The competitors winning the mentions you want are your real benchmark. Track the gap to them, per prompt, over time.
- Which sources put them there. AI answers are built from cited pages. If a rival keeps appearing, look at the sources the engine cited to name them. Those pages are your roadmap: earn a place in them, publish a better answer, or correct the record.
- Where the gap is closing or widening. A single share number hides movement. The win is spotting a competitor pulling ahead on a high-value topic while there is still time to respond.
For agencies
Share of voice is one of the cleanest client deliverables in GEO, because it is comparative and easy to explain: here is your slice of the AI conversation in your category, here is who is taking the rest, here is the trend. Two cautions make it credible. Build it on earned, unbranded prompts so the baseline is honest, and report it with the sources behind it so the client gets a route to act, not just a chart. A monitor per client brand turns this into a repeatable monthly report.
How llemmy does it
llemmy computes share of voice across every tracked brand from the same earned, unbranded prompt set it uses for visibility, so the two numbers reconcile. It separates branded prompts out automatically, measures over a rolling window across ChatGPT, Claude, Gemini, Perplexity and Google AI, and records which sources each engine cited when it named a competitor. For agencies, multi-project workspaces let you run a benchmark per client without per-seat or per-credit anxiety. Run a free GEO audit or start free to see your share against your rivals.
FAQ
What is share of voice in AI search?
It is the proportion of brand mentions in AI answers that belong to you, across a set of category prompts. If five brands are named across the set and you account for one in four mentions, your share of voice is about 25 percent. It shows how often engines pick you over rivals.
How do you calculate AI share of voice?
Pick unbranded, buyer-intent prompts, run them across the engines repeatedly, and count brand mentions. Your share is your mentions divided by all brand mentions across the same set and window. Compute it from rates over a rolling window, not a single scan, and read it per engine before averaging.
Why do branded prompts ruin share of voice?
Branded prompts name a brand in the question, so that brand is mentioned almost every time by construction. Including your own branded prompts inflates your share. The metric only stays honest when it is built from unbranded prompts where every brand has to earn its mention.
What is a good AI share of voice?
There is no universal benchmark, because it depends on how many credible brands exist in your category. The useful read is relative and directional: your share versus named competitors, on the prompts that matter, trending up or down over time.
By the llemmy team, June 2026. Related reading: How to track your brand across AI engines, Why we don't track branded prompts, and how llemmy compares.