Platform AI labels — Instagram's "AI info," TikTok's "AI-generated," Facebook's ad disclosures — are triggered by metadata read at upload, not by a human review. Remove the C2PA manifest and IPTC digital-source-type field before you post, and the platform has nothing to read.
Every major platform that auto-labels AI content does so by reading the same two underlying signals: a C2PA content-credential manifest, or an IPTC DigitalSourceType field. Neither is visible when you look at the image — both are embedded structured data. See exactly how each platform reads these signals: Instagram, Facebook, TikTok.
Removing the metadata addresses the most common automated trigger — it doesn't address visual AI-detection classifiers some platforms run independently of metadata, and it doesn't change what the content actually is. If you're subject to a legal disclosure duty (several jurisdictions now require labeling certain AI content, independent of any platform's own system), metadata removal doesn't satisfy or override that obligation.