The share of judgeable visits on which the client asked before downloading. Its denominator is never the total request count. Where there are no judgeable visits the rate is reported as unavailable, not as zero.
The total size of response bodies sent on blind return visits to pages the client had already been given. It excludes first visits entirely, and excludes responses that were not full pages.
This is the study's headline figure because it is not a ratio and cannot be inflated by how the sample was drawn. It answers a concrete question: how much content was sent to a client that already had it.
The elapsed time between successive visits by the same client to the same page, reported as shortest, typical and longest. Used to distinguish scheduled background crawling from fetches triggered by a user's query, which arrive within seconds of the query being submitted.
Crawlers are grouped by the name they declare in their user-agent string. This is self-reported and can be imitated, so identity claims are treated as claims. Where a stronger assertion is needed, declared identity is checked against the operator's published address ranges.
Part of an open study of AI crawler cache behaviour. Study reference RCP-2026-08. These pages are served by the study's own logging server; requests to them are recorded for the purpose described in the method. Content is stable by design and is not edited during a measurement run.