TikTok tests AI-spam detection: politics, finance, medical explainers

On 2026-09-29 TikTok said it recently began testing improved detection for accounts dedicated to posting AI-generated spam, with politics, financial advice and medical content named as focus areas. It gave no thresholds, no penalties and no date for a rollout. If you publish explainers in those three areas with AI visuals or AI narration, the practical step is to make each clip show its own sourcing and its own script, and to check that before you post. Sume can do that check with video inspect.

This post is about preparation, not about what TikTok will decide.

What exactly did TikTok say?

The newsroom article, dated 2026-09-29 and written for Ireland, covers AI transparency. It says TikTok was the first video platform to implement C2PA Content Credentials and has joined the C2PA Steering Committee. It reports more than 3 billion videos labeled as AI-generated, and says TikTok is testing a Manage Topics control that lets people choose how much AI-generated content they see. In the same article it says it is testing improved detection for accounts dedicated to posting AI-generated spam, in politics, financial advice and medical content.

Why does this matter for a finished clip?

The signal is account-level: accounts dedicated to AI spam, not single videos. A channel that posts dozens of near-identical AI explainers about money or health is the pattern the sentence points at. A channel where each clip has its own script, named sources and a person who reviewed it is a different pattern. TikTok has not described how it tells them apart, so do not read more into it than that.

What can I check on Sume before posting?

Three things are checkable with the tools in the docs. First, that the script differs per clip: run video inspect with transcribe: true on each finished file and compare transcript.text across a batch. The docs list the public rate as $0.01 per audio minute. Second, that on-screen claims are visible and sourced: sample stills with frames.fps up to 2 and look at the source line you burned in. Third, that captions match speech: video captions takes script_text or STT, and burns cues for silent clips.

If you assemble a series from several clips, the timeline lets you place a source card clip at a fixed slot, so every episode ends on the same visible sourcing line. That is a layout choice. It does not make a clip pass any platform test.

What should I keep per video?

Keep the job ids of the generation, the inspect result with its transcript, and the source list you showed on screen. If an account is ever flagged, a per-video record shows the work that went into each clip. For a general log template, see the C2PA post linked below.

What is still unknown?

Whether TikTok's test affects reach, monetization or account standing is not in the article. Whether it applies outside Ireland is not stated. Treat this as a notice to review your own pipeline, and check TikTok's own policy pages for rules that apply to you.