How many visits do AI crawlers send back per scrape? Microsoft Clarity now puts a per-operator number on it

What Microsoft shipped

Per the announcement, the update adds four things to the Bot Analytics section of Clarity's AI Visibility dashboard: an AI Scrape-to-Referral Ratio card that "compare[s] AI scrape activity against referral traffic in a single metric"; an operator-level breakdown — "a ranked list of referring operators" showing "which AI sources are sending traffic back" and which have high scrape activity with little referral return; direct links into session recordings pre-filtered to the selected AI referral source, so you can watch whether AI-referred visitors "scroll, engage, convert, or leave quickly"; and mapped-domain safeguards so ratios are only computed where bot data and referral data cover the same domains. Search Engine Land covered the release the same day. It is the latest in a fast series of Clarity AI-visibility releases — Citations went GA in May, branded/non-branded segmentation landed 2026-08-03.

The catch: it reads server logs, not the tracking tag

AI crawlers do not execute JavaScript, so no tag-based analytics tool — Clarity's own snippet included — can see a scrape. Microsoft's Bot Activity documentation is explicit that the dashboard "is based on real server-side logs collected through supported CDN and server integrations": you connect a supported CDN or server in Settings → AI Visibility before any bot data flows (WordPress sites get it by default via the latest Clarity plugin), and Microsoft notes the log collection "might result in additional costs" billed by your CDN provider, even though Clarity itself is free. The docs also carry a caveat worth quoting verbatim, because it is the honest limit of the instrument: "Bot activity represents requests made to your site and doesn't indicate that content was retrieved, grounded, cited, or surfaced in AI generated responses."

Why this matters if you care about AI visibility

• The scrape side of the ledger was invisible to almost everyone. Ecosystem-wide crawl-to-referral ratios existed (Cloudflare's 50,000:1), but a small business had no free way to compute its own. A per-operator ratio on your own domain turns "AI bots are hammering us" from a feeling into a number — and makes allow/block decisions specific: the operator sending real visitors and the one only extracting are no longer the same line item.
• Blocking decisions get evidence — but remember what a block does and doesn't do. Our robots.txt parsing across the domains AI engines actually cite found naming an AI agent is frequently not blocking it, and blocked domains still get cited. The ratio tells you what an operator takes versus sends; it does not tell you what blocking that operator would cost you inside answers.
• A brutal ratio is not proof the scraping produced nothing. A panel study we covered yesterday found only 1.1% of post-AI-conversation visits arrive with an AI referrer — the influence overwhelmingly lands as direct navigation. A scrape-to-referral ratio undercounts AI's value for exactly the same reason referral rows do. The number that closes the loop is whether engines actually name you in answers — check what AI engines say when asked about your business the way a customer would ask.