Lore

Content-Level AI Labeling vs. Account-Level Authenticity Signals

A policy framework that treats two questions as separate classification problems rather than one:

  1. Was this specific piece of content AI-made? — labeled directly on the content, with a stated confidence tier (definitely / probably / not sure / definitely not), since certainty about AI-generation varies.
  2. Is this account a real person? — exposed separately via account-level signals like profile-change history and account age, letting users judge authenticity for themselves rather than the platform asserting it outright.

Rationale: content-authorship and account-authenticity are different questions with different evidence and confidence levels, so collapsing them into one label loses information users need.

Feeds directly from AI Content Abundance Increases Demand for Authenticity — this is the concrete policy mechanism built in response to that pressure.

Labeling Real Content Instead, and Disclosure Over Detection

Adam Mosseri (Instagram) argues it may eventually be more practical to label camera-captured (non-AI) content as authentic rather than trying to label AI-generated content, since confidence in AI-detection is inherently limited and likely to erode further as generative models improve — flipping the labeling target from the harder-to-verify class to the easier one. He also frames enforcement around disclosure, not medium: a named abuse pattern he wants Instagram to crack down on is undisclosed 'AI monk'-style personas used to sell bogus supplements. His stated principle: using an AI creator persona isn't itself the offense — using non-disclosure to deceive is. He extends this to judgment generally: 'judge it based on the content, the point of view, the person behind the content,' not on whether AI was involved in production.