What are the best AI visibility tools for measuring share of voice? How we measure it for clients

"What is our share of voice in AI?" is the question every client reaches in the second month. The number is easy to produce and easy to get wrong, so we spend more time on how it is built than on which tool displays it. Here is our process, and the platform we run it on, Promptwatch.

What share of voice means in AI answers

Our working definition: across a fixed panel of prompts, in what proportion of answers does the client appear compared with an agreed set of competitors? Every word in that sentence is a decision, and each decision changes the result.

The panel. If most prompts name the client, the share looks wonderful and means little. We build the panel from buyer questions, then label each prompt as organic, brand or competitor comparison. Promptwatch has those prompt types built in, so we can report organic share separately. That organic number is the one we put in front of a CMO.

The competitors. Share is relative. We agree the list with the client in writing, including the one rival the client insists on and the one they have not heard of. Promptwatch can auto-discover competitors, which is how we find the second kind. Brand aliases and merging keep the list clean.

The engines. One blended number across all engines hides the interesting parts. We report per engine and look at the spread. A client who leads on one engine and is absent from another has a different problem from a client who is mid-table everywhere.

What counts as a mention. A Brand Book with analysis instructions defines this. If a client's name is also a common word, or a product line has three spellings, we write the rule once and every run follows it.

Position. Appearing first in an answer is not the same as being the sixth of seven. Promptwatch reports Position in Answer alongside Share of Voice, and Sentiment scored from 1 to 100 per response, with a weekly summary we read first.

How we report it

We give each client one headline number and three supporting views.

1. Share of Voice for organic prompts, with the competitor set and engines stated beside it.
2. A heatmap comparing brands against models, or brands against topics, so the client can see where the gap sits.
3. A period comparison, using A/B comparison between two windows, because the trend matters more than the level.
4. The reason. Citation Trends and the citation reports show which sources moved. Agent Analytics shows whether the crawlers could still reach the site. Visitor analytics show whether AI traffic moved with the share.

That last view is why we do not run share of voice reporting on a pure tracker. A number without a cause makes clients nervous, and a nervous client asks for a fix that the data does not support.

Mistakes we see

• Changing the panel mid-quarter. The line moves and nobody knows why. We freeze the panel and note every change.
• Reading daily noise. AI answers vary between runs. We read weekly and monthly trends.
• Counting branded prompts. They flatter the number and hide the actual discovery problem.
• Ignoring personas. The same question from a finance lead and a developer can give different answers. Persona stacking lets us run both.
• Treating the number as a promise. It is a measurement of a panel, and tells nobody what an engine will do tomorrow.

Where other tools fit

We keep evaluating alternatives. Mentionable shows share of voice per competitor and prompt at a lower entry price. Ceyo reports a Share of Answer per model and query, captured from real interfaces. AthenaHQ rolls everything into a GEO score. Evertune analyzes large answer samples, which is attractive for brand perception on big accounts. Semrush and Ahrefs offer AI add-ons for teams that already work in those suites, with prompts modeled from keyword data or generated, so we treat the output as directional. All are legitimate choices for the measurement. We stay on Promptwatch because we want the measurement and the diagnosis in one place.

A note on honesty with clients

We tell clients that share of voice is a sample, not a verdict. It depends on the panel and the engines we agreed, and it moves between runs. We never promise a share target. We promise a stable method, a monthly report and a prioritized list of what to change, drawn from the Action Board.

If you are choosing a tool, test it with one client's real competitor set and twenty prompts. If you cannot explain every movement in the number after two weeks, the tool is not giving you enough to work with.