How to Measure AI Visibility
Classic SEO has rank trackers and impression counts. AI search has neither in the same form — answers are generated on the fly and don't expose a public ranking. That doesn't make visibility unmeasurable; it just means you measure it differently, and honestly, with sampling rather than certainty.
Sample a fixed set of queries repeatedly and track presence and share over time.
Why the old metrics don't transfer
A traditional rank tracker works because a search results page is a stable, ordered list you can check. AI answers break both assumptions. They are generated, so two people — or the same person twice — can get different wording, different sources, or a different decision about whether to answer at all. And they don't publish a ranked list of who was considered. That is not a flaw to route around; it is the nature of the medium. The right response is to measure with humility: sample broadly, track trends, and resist the temptation to treat any single answer as a verdict.
Four practical methods
Prompt sampling
Ask the AI systems the real questions your customers ask, and record whether you're mentioned or cited. Repeat on a schedule to spot trends.
Citation tracking
For engines that show sources (like Perplexity and AI Overviews), note which of your URLs appear and for which queries.
Referral analytics
Watch your analytics for referral traffic from AI products. It's an imperfect signal — attribution is incomplete — but a rising trend is meaningful.
Crawler logs
Check server logs for known AI crawlers (GPTBot, ClaudeBot, PerplexityBot). Being crawled is a precondition for being retrieved, though not proof of citation.
Building a repeatable check
The discipline that makes any of this useful is consistency. Define a fixed set of questions that represent how real people describe what you do, then run them the same way on a regular cadence. A simple tracking sheet is enough to start:
Query | Engine | Mentioned? | Cited? | Notes -----------------------------------|------------|------------|--------|------ "best roofer in Santa Rosa" | ChatGPT | yes | yes | #2 source "roof repair near me" | Perplexity | yes | no | named, no link "how much does a new roof cost" | AI Overview| no | no | competitor cited "emergency roofer sonoma county" | ChatGPT | no | no | not surfaced
Over a few cycles, the sheet tells a story: which questions you own, which you're losing, and whether your changes moved anything.
Turning measurement into action
Measurement only pays off if it feeds work. When a question you should own isn't surfacing you, ask why: are you even crawlable? Do you answer that question clearly on a page? Is a competitor's content simply more direct? The fixes usually live in the fundamentals — accessible pages, direct answers, clean structure, and a curated llms.txt. For the concept behind the metric, see what is AI visibility.
Frequently Asked Questions
Is there a rank tracker for AI answers?
Not in the way there is for classic search. AI answers are generated, can vary between users and sessions, and don't expose a public ranked list. You can sample and estimate visibility, but no tool sees the full, private selection process.
Why do results differ each time I ask?
Generated answers can vary with phrasing, timing, personalization, and ongoing model changes. That's why single checks are unreliable — you want repeated sampling across representative queries to see a pattern.
What's a realistic metric to track?
Think in terms of presence and share: for a defined set of important questions, how often are you mentioned, and how often are you cited versus competitors? Trend over time matters more than any single snapshot.
Does crawler activity mean I'll be cited?
No. Seeing GPTBot or PerplexityBot in your logs means your content is reachable, which is necessary but not sufficient. Retrieval and citation are separate decisions made by the AI system.
Can a tool measure this for me?
Tools can automate prompt sampling and citation logging at scale, which is far more practical than doing it by hand. Just understand they're sampling and estimating, not reading a hidden leaderboard.
Start measuring your AI visibility
Run a visibility check and generate an llms.txt so AI systems can find and understand your site.