TikTok 90-day violation window: a label ledger for AI batches

Keep one ledger row per clip, and fill the label column before the clip is scheduled. AuditSocials reports that two to three violations in 90 days can bring a 7-day posting ban (third-party summary, read 2026-10-02). If that is accurate, a batch of 40 clips posted without the label is not one mistake, it is forty.

Sume can give you the first columns of that ledger. It cannot fill in the last two, because posting happens on TikTok.

What does the ledger need to hold?

Five fields cover the question a reviewer will ask: which clip, how it was made, when it went out, whether it was labeled, and who checked.

How do you get job ids for a whole batch?

A bulk run queues up to 100 Format runs with a concurrency window, and each child is an ordinary run you read at GET /v1/format-runs/{run_id}. For model jobs, GET /v1/jobs lists recent jobs for the workspace. Send an Idempotency-Key on submit so a retry does not create a second clip that never lands in the ledger.

curl https://api.sume.com/v1/jobs \
  -H "Authorization: Bearer $SUME_API_KEY"

What should block a post?

Make the ledger a gate: no row with Label set = yes, no scheduling. A spreadsheet formula or one line in your scheduler is enough. The point of the 90-day figure is that the cost of a miss is counted across clips, so the check has to be per clip.

Re-check the figure against TikTok's own policy pages before you rely on it. This post cites one summary and makes no claim about how TikTok counts violations.

How long should you keep the ledger?

At least as long as the reported window, which is 90 days, and longer if you run paid campaigns that may be audited. The ledger is small: a few hundred rows is a plain spreadsheet. Keep the Sume job ids even after the clips are posted, since the job record is the only first-party proof of how a clip was made.

If you generate with several tools, add a column for the tool. A ledger that only covers one tool will miss the clip that caused the problem.