The three measurement surfaces
| Surface | What it shows | Honest limits |
|---|---|---|
| Referral segmentation | Sessions arriving with AI referrers (chatgpt.com, perplexity.ai and peers) segmented in your analytics | Engines do not always pass referrers; treat it as the floor, not the total |
| Crawler log analysis | Which AI bots fetch which pages, how often - retrieval demand for your content | Crawling is not citing; it is the leading indicator, not the KPI |
| Dark-traffic inference | Direct-traffic patterns consistent with answer-driven visits (branded landings on deep pages, post-answer timing) | Inference, and labeled as such - never reported as a measured count |
What we build
- An AI referral channel in your analytics - properly grouped, consent-aware, and stable across the engines' referrer quirks.
- Crawler analytics from your logs - our own log-analysis tooling, run in production on our estate first, reporting AI bot activity per page over time.
- The join to outcomes - AI-referred sessions carried through to leads and revenue inside your existing measurement boundary, so "AI traffic" becomes a line in the same report as every other channel, not a novelty screenshot.
- AI-platform tracking installs - as AI platforms ship their own pixels, tags, and conversion APIs, we install, consent-gate, and runtime-verify them exactly like any other vendor tag - no snippet enters the estate unreviewed.
- The disclosure of limits - every number in the layer carries its method, and inferred figures are labeled inferred. This report will be quoted upward; it has to survive scrutiny.
Why the crawler layer matters most right now
Referral traffic tells you about yesterday's answers; crawler activity tells you about tomorrow's. When an engine's bot starts fetching a page cluster repeatedly, that content is in the retrieval pool for generation. Watching crawl patterns per section is the earliest feedback loop GEO work has - and it lives in logs most teams never open.