TikTok's reported detection methods and what to check in your clip

Assume any of three signals may be read, and label the clip yourself. Audit Socials, a third-party summary and not TikTok's own page, reports that TikTok detects AI content through C2PA metadata, visual artifacts and audio fingerprinting (read 2026-10-02). It also reports that content flagged after posting may get a temporary distribution hold. This post takes those as reported, not confirmed.

What can you check before you post?

You cannot see what TikTok sees. You can check what you are publishing, and whether it is one you would label anyway.

How do you pull stills and the audio?

Video inspect takes up to 24 stills per call from one Sume-hosted clip; probe and stills are unbilled. Audio detach returns the audio as wav or mp3 at $0.01 per job (confirm in GET /v1/catalog). Run frames: false first to confirm probe.has_audio, because detaching a silent clip fails with detach_source_has_no_audio.

curl -X POST https://api.sume.com/v1/audio-detach \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: audio-check-001" \
  -d '{"video_url": "https://media.sume.com/artifacts/example/ad.mp4", "format": "mp3"}'

What is the plain decision?

If the clip has an AI face, an AI voice or a photoreal AI scene, label it in TikTok's own control whatever the detection might do. Keep the job id, the date and a screenshot of the label setting. The cost of labeling is a label; the reported cost of not labeling is a distribution hold or a penalty tier.