How to Measure Your AI Share of Voice
Measuring your AI share of voice means tracking, across a fixed set of real customer prompts, how often your brand is mentioned or cited by AI answer engines compared with your competitors — repeated over time with enough runs to be statistically reliable. The four steps below explain the method end to end; each one links to how CiteLens automates it. For benchmark numbers by vertical, see the AI Share of Voice Index, built from 500 prompts and 2000 AI model calls.
- 1
Define your brand and the prompts your customers actually ask
Start with a list of 20–50 real buyer-intent prompts — the questions a prospective customer would type into ChatGPT or ask Perplexity before they know your brand name, e.g. “best [category] for [use case]” or “[competitor] vs alternatives”. Include your exact brand name and at least 2–3 direct competitors so you have something to compare against.
CiteLens ships with a prompt library built from real search-demand data, so you don't have to guess which questions to track — start from a ready-made prompt set.
- 2
Run each prompt across the AI engines your buyers actually use, multiple times
Query each prompt across ChatGPT, Perplexity, Claude and Google AI Overviews — answers are non-deterministic, so a single run is noise. Run each prompt at least 5–10 times per engine before you trust the result; our own sample used 2000 model calls across 500 prompts to get a stable read.
Doing this by hand across four engines and dozens of prompts does not scale past a handful of checks a week. CiteLens automates the repeated runs and logs every raw answer — automate the runs.
- 3
Count mentions vs. citations, and compute your share with a confidence interval
For every answer, record two separate things: was your brand mentioned by name, and was your brand's URL citedas a source. These are different signals — a brand can be named without being linked, or linked without being named. Divide your mention count by total runs to get a share, then compute a Wilson confidence interval around that share so a 3-in-10 result on a tiny sample doesn't get reported as a solid 30%.
Across our sample, the average AI answer cited 9.6sources — so “share of voice” is really a competition for a handful of citation slots, not a single winner-take-all answer. CiteLens computes mention rate, citation rate and the Wilson interval automatically for every prompt — see your share with confidence intervals.
- 4
Benchmark against competitors and track it weekly
A single share-of-voice number means little in isolation — compute the same mention and citation rates for your top 2–3 competitors on the same prompt set, then track all of them weekly. AI answers shift as models update and as new pages get indexed and cited, so a one-time snapshot goes stale within weeks.
CiteLens runs this benchmark on a recurring schedule and alerts you when a competitor gains ground on a prompt you care about — set up weekly competitor tracking.
CiteLens runs this entire method for you — prompt library, multi-engine runs, Wilson-confidence-interval share of voice, and weekly competitor tracking.
Start measuring your AI share of voice