How to Measure Your AI Share of Voice vs Competitors
AI share of voice is the percentage of buyer-intent prompts where your brand gets named or cited across engines, relative to rivals. Measure it with a fixed prompt set, run the same prompts on each engine, and read the trend over weeks, not one snapshot.
What share of voice means in AI answers
In classic marketing, share of voice meant your slice of paid impressions or media spend. In AI answers it means something more specific: out of a defined set of buyer-intent questions, in how many does an engine name or cite your brand, and how often does it name a competitor instead? It is a relative measure. Being mentioned in 4 of 20 prompts is meaningless until you know rivals appear in 15 of 20.
Two things make AI share of voice different from a keyword ranking. First, there is no single result list; ChatGPT and Perplexity may each compose a different answer to the same question, so coverage is per-engine. Second, presence is not binary. Being named as the recommended option carries more weight than appearing in a long list, and a cited source link is stronger again. A useful score weights these tiers rather than counting any mention equally.
Decide upfront what counts as a win: a direct recommendation, a neutral mention, or a source citation. Keep that definition fixed, because a moving definition is the most common reason share-of-voice numbers drift for no real reason.
Sampling buyer-intent prompts consistently
Your number is only as good as your prompt set. Build a fixed panel of 20 to 50 prompts that mirror how real buyers ask: comparison questions (best X for Y), category questions (tools for Z), problem questions (how do I solve W), and alternative questions (alternatives to a named rival). Write them in the language and market your buyers actually use, since German and English answers, and DACH versus US contexts, return different brands.
Consistency beats volume. Run the exact same prompts, in the same wording, across each engine you care about, on a regular cadence, and log who was named and who was cited. Small wording changes shift results, so freeze the panel and version it; only add prompts deliberately. Account for variability too: answers are non-deterministic, so a single run can mislead. Sampling each prompt a few times and aggregating gives a steadier signal than one shot.
Doing this by hand across ChatGPT, Perplexity, Gemini, Claude, Copilot and Google AI Overviews is tedious and error-prone, which is where a tracker like CitePeak helps keep the panel and cadence identical run to run.
Reading the trend, not the snapshot
A single measurement tells you almost nothing, because AI answers wobble between runs. The signal lives in the trend. Plot your share of voice per engine over weeks and look for sustained direction: a slow climb after you published comparison content, or a steady drop while a competitor accumulated citations. One bad week inside a rising line is noise; three falling weeks in a row is a pattern worth acting on.
Segment the trend so it is actionable. Separate engines, because a gain in Perplexity and a loss in Google AI Overviews net to zero but mean very different things. Separate prompt types too: you might lead on alternative-to queries yet trail on broad category questions, which points to exactly where content is missing. Tie each visible shift to something you changed or something a rival did, so the chart becomes a feedback loop rather than a vanity graph.
Set a realistic cadence, weekly or biweekly, and judge progress against your own baseline and against named-but-anonymous rival coverage. The goal is not a perfect snapshot; it is knowing whether the line is moving the right way and why.
FAQ
How many prompts do I need to measure AI share of voice?+
A fixed panel of 20 to 50 buyer-intent prompts is usually enough to be stable without becoming unmanageable. What matters more than the count is keeping the exact same wording across every engine and every run, so changes reflect reality rather than a different question being asked.
Why does my AI share of voice change when nothing changed on my site?+
AI answers are non-deterministic, so the same prompt can return different brands on different runs, and engines update their models and sources continuously. That is why one snapshot is unreliable. Sample each prompt a few times and read the multi-week trend instead of reacting to a single result.
Can I track share of voice across all AI engines the same way?+
The method is identical, but each engine answers separately, so you measure coverage per engine and never blend them into one number. A gain in Perplexity and a loss in Google AI Overviews tell different stories, so keep ChatGPT, Gemini, Claude, Copilot and the rest on separate lines.
See how AI describes your brand — free.
Check visibility