FTC: your company answers for ad claims, so log every AI avatar batch

The FTC's endorsement Q&A says "Your company is ultimately responsible for what others do on your behalf," and expects reasonable programs to train and monitor the people in your network (FTC Q&A, read 2026-10-02). If you generate dozens of avatar ads a day, the monitoring program is a log: one row per job with the script, the disclosure and who approved it.

What the FTC page asks for

The page is written about influencer networks, but the logic carries to automated production: you cannot point at a tool or a freelancer when a claim is wrong. It also says disclosures must be hard to miss, and that one in the middle or at the end of a post is easier to miss.

What a log row holds

Sume jobs give you stable ids to hang a review on. A job has an id, a status and a result you can fetch again later, and you can set a webhook_url so each finished clip posts to your own system (see webhooks). That is enough to build the log without a spreadsheet of screenshots.

• Job id and the artifact URL of the file you actually published.
• The exact script and the disclosure line burned into it.
• Reviewer name, review date and the claim substantiation on file.
• The platform label or paid-partnership flag you set at upload.

A review step that uses a transcript

For a finished clip, video inspect can return probe facts and stills, and with transcribe: true a speech-to-text pass. A reviewer can read the transcript instead of scrubbing the video, and compare it with the approved script to catch drift. Probe and stills are unbilled per the docs; transcription is a separate step, so check the catalog price before running it on every file.

Limits

Sume does not review claims, check substantiation or decide whether a testimonial is truthful. A log shows that you looked; it does not make a false claim true. Use it alongside legal review, and keep it for as long as the ad can still be seen.