TikTok Shop's AI rule on instant-effect edits: check clips by frame

Before you post an AI product clip on TikTok Shop, pull still frames and the transcript and compare them with the listing page, because the August 2026 Policy Pulse names "instant-effect edits, fictional product functions, or exaggerated results" as the AI content problems it is refreshing guidance on. Sume can produce the evidence quickly, but it cannot decide for you what counts as exaggerated.

The facts about TikTok come from its own Policy Pulse page. The facts about Sume come from the video inspect docs.

The same page lists other August updates, such as minors as the sole or primary on-camera participants in commerce content and a prohibited-behavior policy. They are outside this post, but they are a reminder to read the whole Policy Pulse before a campaign rather than only the line you were told about.

What does the Policy Pulse say about AI content?

The August 2026 entry says refreshed resources address instant-effect edits, fictional product functions and exaggerated results. It lists the categories it applies to: cleaning tools, eyewear, knives, shoes, lights, fitness products, jewelry and home goods. It also says claims must match the product detail pages and that effects cannot exceed actual capabilities.

That is a rule about the match between video and listing, so the check is a comparison, not a keyword scan.

What would a frame-by-frame check look like?

Ask for eight stills across the clip and the transcript in one call. Inspect returns durable image artifacts and, with transcribe: true, the words spoken. The clip must already be on media.sume.com, since inspect does not fetch the open internet; import it first.

Then read the frames next to the product page. For a cleaning product, does the stain disappear in one wipe while the page says it takes repeated passes? For shoes, does the clip show a feature the listing does not?

Read the frames at the same size a shopper would see them, on a phone, not zoomed in on a monitor. A subtle before-and-after often looks bolder at thumb scale, and that is the version a reviewer or a viewer remembers.

curl -X POST https://api.sume.com/v1/video-inspect \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: tts-claim-check-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/clip.mp4",
    "frames": { "fps": 1 },
    "transcribe": true,
    "language_code": "en"
  }'

Which claims need the closest look?

The table maps the policy's wording to something you can check in a frame or a transcript. Anything you cannot verify against the listing should be cut or reworded.

What does Sume not do here?

Sume does not judge compliance and does not know TikTok's enforcement thresholds. Inspect returns probe facts, stills and an optional transcript, and that is all; the docs say it is not a typed-scene or semantic analysis. A person still reads the evidence.

If the generated clip overpromises, fix the prompt. Describe what the product actually does at its real pace, then regenerate; see the accuracy rules post for the earlier policy.

How do I write the prompt so the clip stays honest?

Describe the product doing what the listing says, at the pace the listing implies. If the page says a lint roller picks up pet hair in a few passes, ask for a few passes. Avoid prompts like "stain vanishes instantly" or "transforms in one second," which push the generator toward the exact jump cut the policy is concerned about.

Then check the output the same way every time, with the same inspect settings, so you can compare clips across a batch. The video frames route returns exact source-size stills at times you name, up to 24 per call, if you want a closer look at one moment such as the instant a result first appears.