// AI Search Performance Measurement

You Can't Measure What Google Won't Show You. You Can Still Measure Something Real.

No platform hands you a clean click count for AI Overviews, and no tool can promise the same precision as a rank tracker. That's not a reason to guess. It's a reason to measure the parts that are actually measurable, honestly, instead of buying a confident-looking number that isn't.

Talk to us about your AI search data

The Problem

Every Confident Number You're Being Sold Is Someone's Best Guess

Google's own Generative AI performance report in Search Console shows impressions only, no clicks, no click-through rate, and no visibility into the actual prompt or query that triggered the impression. That's not a rollout gap that will close later; it's Google's own documented boundary on what the report shows.

Referral traffic from AI platforms is harder still. ChatGPT, Perplexity, and Claude frequently strip the referrer header entirely, mobile in-app browsers drop it by default, and a user copying a link out of a chat transcript leaves no trace at all. Google Analytics' own native AI Assistant channel, launched May 13, 2026, doesn't cover Perplexity or Claude, and excludes Google's own AI Overviews and AI Mode entirely, since those pass through as ordinary Google Organic traffic. The gap isn't a missing feature. It's a structural blind spot every tool in this category has to work around.

The Folklore

Citation Tracking Tools Promise Rank-Tracker Precision. The Evidence Says Otherwise.

A traditional rank tracker measures something genuinely stable, a URL sits at position 2 for a given query until the index changes. AI citations don't work that way. Independent research (Kevin Indig, with AirOps, 815,000 prompt-page pairs, published on Search Engine Land, June 2026) found that running the exact same prompt three times against ChatGPT produced consistent citations only 2.2% of the time. A separate, larger study (SISTRIX, 82,619 prompts, over 1.5 million snapshots, 17 weeks across six countries) found weekly citation churn ranging from roughly 56% to as high as 74% depending on the platform and market.

A tool that runs a prompt once and reports the result as your "AI visibility score" is reporting a single, effectively random sample, not a measurement. That doesn't make citation tracking worthless. It means a single-run snapshot is close to meaningless, and a tool worth paying for needs to say plainly how many times it re-samples and how it handles the variation, not just hand you one clean-looking number.

What Actually Works

What's Genuinely Measurable, and What Isn't

Google Search Console's Generative AI performance report is real and useful within its actual limits: impressions, pages, countries, and devices for AI Overviews and AI Mode. Used honestly, that's a trend line for whether your visibility in these surfaces is growing or shrinking, not a click report and not a query report, because it was never built to be either.

On the traffic-attribution side, a custom GA4 channel group, matching hostnames for ChatGPT, Perplexity, Gemini, Copilot, and Claude, ordered above the default Referral rule, recovers sessions Google's native channel misses. It still can't see sessions that arrive with no referrer at all, which is a real and permanent floor, not a configuration problem to solve away.

On citation tracking, the fix isn't a better single-shot tool, it's repeated sampling treated as the baseline requirement, not an upgrade: the same prompt run multiple times across a period, reported as a rate with a stated sample size, not a static rank.

Breaking It Down

Two Different Blind Spots, Not One

What Google Shows You

Impressions, pages, countries, and device breakdowns for AI Overviews and AI Mode, directly from Search Console. Real data, genuinely useful as a trend line.

What Google Doesn't

Clicks, click-through rate, and the actual prompt or query behind any impression. Not a rollout gap, a documented boundary that isn't changing on its own.

This is one piece of what Answerability audits, alongside entity and brand consistency, technical crawlability, and whether your content itself answers a real question.

See the Full Answerability Approach

Questions

Frequently Asked

Can you get us exact click numbers for AI Overviews?

No, and any tool that claims to is either estimating and calling it exact, or measuring something else and calling it clicks. Google's own report doesn't expose click data for these surfaces at all. What we can build is an honest impressions trend plus a defensible estimate of referred traffic, clearly labelled as an estimate.

Are AI citation tracking tools worth paying for?

Some are, if they re-sample the same prompts repeatedly and report a rate rather than a single-run snapshot. Ask any vendor directly how many times they run a prompt and over what period before trusting the number they show you.

How much of our traffic actually comes from AI search right now?

Honestly, nobody has a precise answer, including us. Industry estimates range from roughly 1% to 6% of sessions depending on whose panel you trust, and the spread itself is the real finding, no neutral, audited measurement of this exists yet, only vendor panels with different compositions and different incentives. We'll give you your own site's best estimate, not someone else's industry average.

When is this not the right fit?

If you just need a rough sense of whether AI search visibility is trending up or down, Search Console's own report may already tell you that without any additional tooling. This matters most once you need to justify budget or prove impact, where "trending up" isn't a specific enough answer anymore.

Find Out What Your Site's AI Search Data Actually Shows

AI Search Performance Measurement is one piece of a larger picture, alongside entity and brand optimisation, technical AI search optimisation, and content architecture. See the full Answerability approach.