Spoke
AI share of voice
How often you are named across a fixed set of questions, relative to the people you compete with. It is noisy, it is the closest thing available, and it breaks the moment anyone reports it as a single number.
A practitioner method. No platform publishes citation share or endorses any methodology for computing it, and the limitations described here follow from how generated answers vary between runs.
Rank tracking worked because position was stable, ordinal and observable. None of those hold for generated answers, so the substitute has to be relative and statistical rather than absolute.
What is AI share of voice?
It is the proportion of a fixed set of questions on which you are mentioned or cited, measured against the competitors who appear on the same questions. It is a relative measure by design, because an absolute count has no meaning without knowing which prompts were asked and how often.
It is the closest available substitute for rank tracking in generated answers, and it is substantially noisier. Treat it as a trend across many prompts over time, never as a position.
Why share rather than count
"We were mentioned eleven times" is uninterpretable. Out of how many prompts? Which ones? Compared with whom?
Share fixes the denominator. If you appear on a third of your panel and your closest competitor appears on two thirds, that is a fact you can act on. If both numbers move together next quarter, that is the engines changing rather than you.
What to record per run
Four things, kept separate, as the prompt panel spoke sets out:
- Mentioned — named at all.
- Cited — named with a link, meaning retrieval found your page specifically.
- Competitors named — the denominator that makes share meaningful.
- Characterization accurate — whether what was said about you is true.
The fourth is the one teams skip and the one that most often produces urgent work. A confident wrong description is worse than absence, and it is invisible in any metric that counts appearances.
The noise problem
Generated answers vary between runs for the same prompt. That is not a defect in your measurement; it is a property of the thing being measured.
The consequences are practical. Run each prompt several times. Report ranges rather than points. Treat a single run as an anecdote and a single-period change as noise until it persists. A panel re-run monthly produces a usable trend in a quarter — not next week.
What cannot be done
Market-wide share. No engine publishes citation data, so nobody can compute your share across all conversations. A vendor reporting your "AI visibility score" is reporting their panel, which is a sample of their choosing — ask what the prompts were.
Cross-engine comparison as one number. Engines differ in whether they retrieve, cite and link. Averaging across them produces a figure describing no system that exists.
Causal attribution. Share going up alongside a content program does not establish that the program caused it. Engines change underneath you constantly.
Reporting it honestly
A defensible slide says: these prompts, these engines, this cadence, this many runs, here is our share and here are the competitors, here is the direction over three periods, and here is what we cannot see.
An indefensible one says: our AI visibility is sixty-two percent. The number looks more rigorous and is the one that cannot survive a follow-up question.
More in Measurement
Prompt panel tracking
Asking an assistant about yourself once tells you nothing. A fixed set of prompts, re-run on a schedule and recorded, tells you a direction — which is the most that is available.
AI visibility reporting
Google and Microsoft now report AI visibility directly, and OpenAI tags its referrals. That beats a third-party citation tracker — and each one measures something narrower than its name suggests.
The attribution gap
Being cited without being clicked is a real outcome that last-click reporting cannot see. It usually resurfaces later as direct or branded traffic, credited to the wrong thing.
AI referrers
The visitors who do click from an AI answer are identifiable in ordinary analytics. It is the easiest measurement win available here, and it counts the smallest part of the channel.
Self-reported attribution
One required field on every intake. It is biased, imprecise, and the only place a human tells you directly that a machine recommended you.
KPIs that survive
Sessions as a headline metric tells a story that gets less true every quarter. These are the measures that keep meaning what they meant when the traffic mix changes underneath them.