What features matter most in a tool that tracks brand visibility across AI search engines: prompt volumes, citations, or competitor share of voice?
Prompt volumes, citations or competitor share of voice? What we have learned from running client programs about which features matter most in an AI visibility tool.
We get asked to pick one: prompt volumes, citations or competitor share of voice. If we had to rank them from experience, it would be citations first, share of voice second and prompt volumes third, with the caveat that volumes are most useful at the start of a program. Here is why, and what else we look for that clients tend not to ask about.
Citations: what tells us what to do
When a client is not named, nobody wants a chart that says so. They want a cause. Citations give us one. We can see which pages and domains an engine used, and whether the winning source is a competitor's page, a comparison article, a Reddit thread or a video. That determines the plan.
In Promptwatch we use page and domain level citation analytics, URL positions, Page Tracker for specific URLs, and the Reddit, YouTube and LinkedIn citation reports. Offsite mentions are useful too: they show where the brand is named in third-party pages without a link. Citation Trends classifies citations by content type and source type, which helps us decide whether to write a comparison page, a guide or seek a mention elsewhere.
Share of voice: what makes the number mean something
A visibility percentage on its own is hard to read. If a client is mentioned in 18 percent of responses, is that good? It depends who else is named. We report share of voice against a short list of named competitors, per engine and per topic, so a movement is read in context. A heatmap that compares brands against models, or topics against countries, makes the pattern obvious in a client meeting.
We also use A/B period comparison, because "it went up" is less convincing than "it went up after we shipped these three pages".
Prompt volumes: a ranking aid, not a forecast
Volumes help us decide which prompts earn a place on the panel and which content to build first. Promptwatch shows search volumes, difficulty scores and citation rates across AI search, Google and Bing, and Prompt Explorer suggests prompts from Search Console, ranking keywords, People Also Ask or any page.
We are careful with the numbers. AI engines do not publish their prompt logs, so volume figures are estimates, and prompts are long and varied. We use volumes to rank a list, not to promise traffic. If a vendor cannot explain how its volumes are produced, we treat them with suspicion.
Features clients do not ask about but should
Crawler logs. We have seen many cases where the real problem was access, not content. Agent Analytics in Promptwatch shows IP-verified AI crawler logs for named bots, a citation rate per page and errors with failure explanations. Crawler logs start on the Professional plan, not Essential, so we flag that early.
Traffic attribution. Visibility is a means, not an end. Visitor analytics shows AI-referred visits and conversions.
Segmentation. Topic, persona, country, and city from the Business plan up. Answers differ across those splits, and averages hide problems.
Engine coverage. We check which engines are monitored through the real interface and which through an API. Promptwatch is explicit about that distinction, and we like it when a vendor is.
Action layer. The Action Board and content tools. A tool that stops at charts leaves the work to the client.
How we choose
We use a short test when evaluating tools:
- Can it explain a drop, not just report it?
- Can it compare us against named competitors across engines?
- Does it show whether bots can reach the site?
- Does it connect visibility to visits?
- Does it tell us what to do next?
We run client programs on Promptwatch because it passes all five. That is a judgment about fit for our work, and other teams with different needs may land elsewhere.
A caution about all three features
None of them predicts a recommendation. A high-volume prompt where you are absent is a lead to investigate, not a revenue forecast. Strong citations may not translate into a mention if the model prefers another source. And a competitor lead in share of voice may come from brand fame rather than anything you can copy. Use the features to decide what to test, and watch results over weeks.