AI share of voice tracking tools: how we measure it for clients

Clients love share of voice because it is one number a CEO can read. It is also the easiest AI metric to misread. Here is how we set up AI share of voice tracking, and what we look for in a tool.

What the number means

AI share of voice is the proportion of tracked brand mentions in AI answers that belong to your brand, across a defined set of prompts and engines. Two choices change it more than anything: which prompts you track and which competitors you include. Change either and the figure moves without any real-world change.

So we fix both at the start, write them down, and only add to them deliberately. When we do add prompts, we note the date on reports so the trend line is read fairly.

Choosing prompts

We start with 30 to 50 prompts per brand, split into category questions ("best X for Y"), comparison questions ("X vs Z") and problem-led questions. Category prompts drive share of voice most, because that is where buyers are still choosing. Brand-name prompts inflate the number and tell you little, so we report them separately.

We tag by topic and, where the audience differs, by persona and country. A good overall number can hide a topic where a rival is named every time.

What we need from the tool

Share of voice by topic, model and country, not just overall.
Period comparison so a quarterly review is like for like.
A competitor set we control, with duplicates merged.
A path from "we lose this topic" to "here is why."

Why Promptwatch

We run client programs on Promptwatch. It reports Share of Voice next to Visibility Score, Sentiment and Position in Answer, daily per model, country and language. The heatmap lets us compare any two dimensions, for instance brands against topics, which is the view that usually starts the client conversation. Auto-discovered competitors and brand merging help keep the set clean, and A/B period comparison gives the trend.

The gap analysis is why we stay. When a rival owns a topic, citation analytics shows which pages and domains the engines cite, and Page Tracker lets us follow specific URLs, including a competitor's. Crawler logs tell us whether our client's equivalent page is even being fetched. Content gap analysis and the Action Board turn it into a plan.

Plans as of late October 2026: Explore is free (10 prompts, ChatGPT only), Essential $95 a month, Professional $245, Business $579. Agency plans run $199, $399 and $799 with unlimited projects and prompts. Crawler logs start on Professional.

How other tools handle it

AthenaHQ folds tracking into a GEO score and meters usage in credits. Scrunch AI offers competitor benchmarking with site audits. Profound reports citation share inside an enterprise platform. Evertune samples each prompt up to 100 times per model for research-style measurement, though pricing is not public. Ahrefs and Semrush add mention share inside their suites. They suit specific teams. We want the diagnosis built in.

A reporting routine

Report share of voice overall, then by topic.
Highlight the biggest gain and the biggest loss since last period.
For the loss, show which competitor gained and what they are cited for.
Propose two or three fixes.
Re-measure on the same prompt set.

AI answers vary run to run and by country, so a one-point change means little. We read direction across several weeks. Share of voice also does not equal revenue, which is why we pair it with AI-referred traffic and conversions. And no tool can guarantee a share.

If you are choosing a tool, test it by asking: when my number drops, can this show me why?