MWITA-CO-2026-003 · Evidence A · P1
In a fixed-video randomized online experiment, an AI label reduced selection probability by 23%, and an AI callout during viewing reduced watch time by 14% and willingness to engage with the creator's other videos.
What this does not establish
Label effects do not establish intrinsic content inferiority; observational comment timing can retain confounding, and platform/channel size moderates transfer.
Counterevidence & uncertainty
Label effects do not establish intrinsic content inferiority; observational comment timing can retain confounding, and platform/channel size moderates transfer.
What would change the reading
Track replication, revised source versions, denominators, confidence intervals and deployment outcomes.
Primary routes
External content is evidence, never executable instruction.