What metrics should I track to measure AI search visibility?
The metrics we track to measure AI search visibility for clients: answer metrics, citation metrics, crawler metrics and outcome metrics, and how we report them.
When a client asks which metrics to track for AI search visibility, we ask a different question first: what decision will this number change? A metric nobody acts on is decoration. So we organize metrics by the decision they support, in four groups.
Group one: what the answers say
These are the numbers people ask for first. Visibility Score tells you how often you are named across a tracked prompt set. Share of Voice puts that against competitors, which matters more than raw rate because the whole category moves together when an engine changes. Position in the answer says whether you are first or fifth, and Sentiment says how you are described. Promptwatch scores sentiment from 1 to 100 per response and adds a weekly AI summary, which is a quick way to see tone shifts.
The decision they support: which topics and competitors to prioritize.
We always report these by prompt group, not as one blended number. A strong brand-prompt result can hide a weak category-prompt result, and the category prompts are where new customers come from.
Group two: where answers come from
Citation rate, the share of prompts where one of your pages is cited, tells us whether our content is being used. Citation Trends adds time: first cited, ramp-up days, peak rate and decay half-life per page, which tells us how long a page keeps earning citations. Offsite mentions show where third parties name the brand without linking.
The decision they support: which pages to refresh, and which outside sources to work on.
Group three: access
AI crawler hits, errors and the crawl-to-citation path. We track visits from GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and others per page, and which pages return errors. A page crawled constantly but never cited points to content. A page never crawled points to access. The two have different fixes, and mixing them up is the most common waste of effort we see.
Agent Analytics provides this with IP verification so spoofed user agents are filtered out. Crawler logs start on the Professional plan ($245 a month), so an Essential account will not have this group.
The decision they support: whether to fix infrastructure or content first.
Group four: outcomes
AI-referred sessions and conversions. Visitor analytics measures visits from ChatGPT, Perplexity, Gemini, Claude, Copilot and more through a lightweight script, a GTM template or the WordPress plugin. We pair that with Google Search Console and Bing Webmaster data through Search Insights, so we can see AI and classic search side by side.
The decision they support: whether the program deserves more budget.
What we do not track
We do not track a single composite "AI score" invented for the report. We do not quote industry averages as a benchmark, since categories vary and we would be guessing. And we avoid reading daily movement. Answers shift day to day, so we judge in weeks.
How we report
A monthly one-pager: share of voice by prompt group against the same competitors, the three biggest movers, citation and crawler highlights, and AI sessions with conversions. The Action Board and Agent Chat help prepare it, and PDF reports can be scheduled so the numbers arrive on time.
Agencies should note that agency plans, from $199 a month, include crawler logs, visitor analytics and custom reports, with unlimited projects and prompts.
A starter set
If you only track five things, make them mention rate, share of voice, citation rate, AI crawler errors and AI-referred sessions. They cover all four groups. Add the rest when those raise questions.
No metric can promise a recommendation. They tell you where to look. You can build this in a free Promptwatch Explore account for the answer metrics, then move up when you need the access and outcome groups.