Share of voice
The share of answers in your question set where you get named. The population is all answers. Drawback: a very general question counts the same as one close to a purchase.
Metrics for AI visibility
Four numbers, four different questions.
There is no single number for AI visibility, and anyone offering one is aggregating things that do not belong together. This page sorts out which metric answers what.
Assignment
Which metric belongs to which discipline.
In detail
What each metric says.
Answer ownership
The share of questions where the cited passage is recognisably yours. Tighter and more meaningful than share of voice, and more work to record, because the text has to be traced back.
Citation frequency per model
How often a given system names you and which source it leans on. The distribution is what counts: if a model knows you only through one third-party page, everything hangs on that page.
Click loss
Clicks per position compared with earlier. Not an exact quantity, because no official report exists. The most solid approximation: position holds, click rate falls.
Limits
What no metric delivers.
None of the four measures revenue. The path from a mention to an enquiry cannot be traced end to end, because many contacts happen without a click.
What helps is a plain question in the contact form: how did you find us. That is not statistics, and it is the only link between mention and enquiry you get without speculating.
And none of the four replaces the question of whether the mention is even correct. High visibility with wrong details is worse than low visibility with right ones.
Common questions
About metrics.
Can the four metrics be combined into one?
Technically yes, sensibly no. They rest on different populations. A combined figure looks precise and loses exactly the information you measured for.
Which metric do I report to the board?
Usually share of voice, because it is the easiest to explain, with the number of competitors named next to it. Everything else belongs in the appendix.
How exactly is share of voice calculated?
Mentions divided by answers in the set, per system. We count a mention once even if it appears several times in the same answer, because repetition says nothing about visibility.
Is a high value in one system worth much?
Only if your audience uses that system. A top figure in a service nobody in your sector uses is a nice slide with no effect.
How do I recognise progress?
Through the same measurement at a later date. Without a repeat on an identical set, any statement about progress is an assertion.
What about sentiment?
The tone of a mention can be recorded and matters for brands with reputation issues. We add it on request, though not as a lead metric.
Does position inside the answer count?
We record it and weight it cautiously. Whether the first of three mentions is worth more than the third is plausible and unproven.
How can click loss be evidenced?
Only by approximation. Lay position and clicks per page against each other over time, subtract season. Anyone promising an exact figure does not have one.
Does this need a special tool?
No. A question set, a spreadsheet and discipline are enough to start. Tools save time and change nothing about the method.
How often do sensible metrics change?
The metrics themselves barely, the ways of collecting them frequently, because interfaces change. That is why we document the method and not only the result.
Next step
Let us find out where you stand in AI answers.
One conversation, 30 minutes, no sales pressure. You leave with an honest read on whether this is worth the effort for you.
