TikTok Algorithm Safety Experiment Exposes the Cost of Watch Time

Creators love an algorithm when it delivers views. Less fun: learning what the machine may be willing to trade for another minute of watch time.

A newly revealed TikTok experiment is dragging that uncomfortable question into daylight. And on August 21, the platform got another expensive reminder that "trust us" isn't a safety policy.

The experiment

In 2021, TikTok developed a system meant to interrupt runs of repetitive, potentially harmful videos. Think depression, extreme dieting, loneliness and self-harm - subjects that may become dangerous when the feed serves them again and again.

But roughly 10% of American users, estimated at 15 million people, reportedly remained on the old recommendation system as a control group. Minors weren't excluded.

One was Chase Nasca, a 16-year-old placed in the test in January 2022. He died by suicide the following month. An internal review reportedly found that, during his final two weeks, he watched 7,563 videos. Seventy-three percent involved sadness or personal struggles, while nearly 10% violated TikTok's own rules.

Senators Marsha Blackburn and Richard Blumenthal have now sent TikTok executives 13 detailed questions¹, including who approved the experiment, why children were included and whether engagement metrics influenced safety decisions. TikTok has until September 1, 2026, to answer.

Separately, the Justice Department announced a $400 million settlement² with TikTok and ByteDance today over alleged children's privacy violations. TikTok will pay $300 million immediately, with another $100 million tied to the removal of an earlier consent decree. The settlement contains no finding of liability, and the DOJ says TikTok has since strengthened age controls and parental oversight.

Your content isn't alone

Creators tend to judge a post as one object: helpful or harmful, good or rubbish. Recommendation systems don't work that way. They build sequences.

A vulnerable viewer may not receive one video about heartbreak. They may receive fifty. A perfectly policy-compliant clip can become another brick in a very dark tunnel.

The algorithm doesn't understand your good intentions. It understands that somebody paused, replayed and stayed.

This matters for distribution, too. Platforms are tightening what teens can discover. Instagram, for example, now restricts younger users from following accounts that repeatedly publish age-inappropriate material and can remove teens from those follower lists. That means certain growth tactics won't merely look grubby. They may quietly shrink reach.

Do this now

• Audit sequences, not posts. Watch your last ten videos together. What emotional loop do they create?
• Stop baiting vulnerable viewers. Don't turn anxiety, body image or despair into an open-ended cliffhanger factory.
• Add exits. Resource links, context and a clear change of tone can help break repetition.
• Build direct distribution. Email lists and communities matter when platform safety changes suddenly redraw your audience.

Views are numbers. Viewers aren't. Easy sentence to nod at. Harder one to build around.