Which software measures a brand's visibility score in AI answers? What we use
How we measure a brand's visibility score in AI answers for clients: prompt sets, mention rules, position and sentiment, and the Promptwatch features behind the score.
Every client wants one number. "What's our AI visibility score?" We understand why, and we give them one. But a score is only worth reporting if you can explain it, so how it is measured matters more than the number. Here is how we measure a brand's visibility score in AI answers, and the software we run it on.
A score is a method, not a fact
Two tools can report 40% for the same brand and mean different things. One counts any mention across a handful of prompts. Another counts only first-position recommendations across hundreds. Without knowing the method, the number is decoration.
So before we report a score to a client, we define four things.
The prompt set. Which buyer questions, in which markets, grouped by topic and intent. We agree this with the client, then keep it stable, because changing prompts changes the score.
The mention rule. What counts. Official brand name, product names, common misspellings, domain mentions. And what does not: a similarly named competitor, an unrelated company with the same name.
Position and sentiment. Being named first for "best X" is not the same as being named fifth with a caveat. We track position and sentiment as well as presence.
The cadence. Answers vary run to run, so we run daily and look at trends.
How we configure it in Promptwatch
We run client programs on Promptwatch, and the scoring setup maps neatly to the four points.
Its dashboards report Visibility Score, Share of Voice, Sentiment (scored 1 to 100 per response, with a weekly AI summary) and Position in Answer. Monitors run prompt panels daily per model, country and language. The Brand Book carries analysis instructions that decide what counts as a mention, and brand aliases handle name variants, so we write the mention rule down once and it applies to every run. Topics, tags, prompt types (organic, brand, competitor comparison) and personas let us slice the score the way the client thinks about their market.
When the score moves, we open it. The heatmap compares any two dimensions, such as brands against models or topics against countries. A/B period comparison shows what changed between two windows. Competitors are auto-discovered, and brand merging helps keep the competitor list clean. This is where we stop saying "the score dropped" and start saying "it dropped on Perplexity for the comparison prompts in Germany".
The score is also not where our work ends. Citation analytics tells us which pages and sources the answers drew on, IP-verified crawler logs tell us whether AI bots reach the client's pages, visitor analytics counts AI-referred visits and conversions, and the Action Board turns the diagnosis into a prioritized to-do list. A score that connects to those is something we can improve, not just report.
Tracked models, and how we label them
Promptwatch monitors the live interfaces of ChatGPT (Search), Google AI Overviews, Google AI Mode, Microsoft Copilot, Gemini, Perplexity, Amazon Alexa and Meta AI, and tracks models such as Claude, DeepSeek, Mistral and Grok through APIs. In reports we say which are which, since an API response is not the same as what a consumer sees in the interface. A client asking about Claude visibility should hear that.
Plans and scoping
Brand plans run from the free Explore (10 prompts, ChatGPT only) through Essential $95/mo (50 prompts), Professional $245/mo (150 prompts) and Business $579/mo (350 prompts). Crawler logs start on Professional. Agency plans (Kick-off $199/mo, Growth $399/mo, Scale $799/mo) include unlimited prompts and projects. We pick the plan by prompt volume, number of markets and whether the client needs the crawler and conversion layers.
Other approaches we know about
Some tools sample each prompt many times to stabilize the score. Evertune takes that route, aimed at enterprise insights teams, and is sold through demos. Others, including SEO suite add-ons, report mentions beside keyword data. These are valid, and the right choice depends on the audience. We prefer a score that sits next to the evidence we need to improve it.
What we say to clients
We say the score is a measurement of what the models said across an agreed prompt set, on the dates we ran them. We say that trend matters more than any one day. We say that the score can be improved but not guaranteed, and that no software, ours included, can promise a model will recommend a brand. And we show the breakdown, because the number invites questions and we would rather answer them in the report than in a panicked email.
If you are choosing yourself
Ask whether you can see and control the prompt set, whether a mention rule is configurable, whether position and sentiment are included, whether the score breaks down by engine and country, and whether it links to the sources and traffic behind it. Promptwatch's free Explore plan lets you see the score on ChatGPT with 10 prompts before you commit to more.