YouTube AI Detection: How Kurzgesagt Lost Recommendations

You spend months making something original. Then the platform mistakes it for the cheap AI copies flooding the feed and quietly turns off the tap.

That's reportedly what happened to Kurzgesagt, one of YouTube's biggest educational channels. Twenty-five million subscribers didn't make the algorithm less wrong. It only made the mistake easier to spot.

The numbers made no sense

Kurzgesagt had already noticed odd swings in views. Then its animation about microscopic superpredators landed with a thud: the channel's weakest upload since 2013.

Except viewers weren't rejecting it. Click-through rate was above normal. People watched longer. Audience feedback was strong. The video should've performed somewhere around average. Instead, recommendations dried up.

In an account published by the studio¹, Kurzgesagt said YouTube investigated and found that automated systems had misclassified its human-produced animation as low-quality AI content.

YouTube helped address the problem, according to the channel. Kurzgesagt removed the affected upload and plans to revise and release it again within weeks.

The scary bit isn't that automation made a mistake. Automation always makes mistakes. The scary bit is discovering that mistake through missing revenue.

Your dashboard won't explain this

YouTube began rolling out new AI-detection signals in May 2026. Its public announcement² focused on identifying significantly AI-generated material and said an AI label alone wouldn't affect recommendations or monetization.

Kurzgesagt's case appears different: not simply a visible label, but an automated judgment that affected distribution. YouTube hasn't published a technical breakdown of the incident.

That gap matters. When a video fails, creators normally blame the topic, thumbnail or opening hook. Fair enough. But if CTR and watch time improve while impressions suddenly collapse, you may be fixing creative work that wasn't broken.

Smaller creators reacted with the obvious concern: Kurzgesagt had contacts who could investigate. Several commenters claimed they'd experienced similar drops without meaningful support. Those stories aren't proof of a wider failure. They are, however, one hell of a smoke alarm.

Build your paper trail

For your next upload, capture screenshots after the first hour, six hours and 24 hours. Track impressions alongside CTR, average view duration and the share coming from Browse and Suggested. Views alone tell you almost nothing.

Keep your production evidence too: scripts, project files, raw recordings, drafts and dated exports. Not because you should have to prove you're human. Because arguing with a machine is easier when you arrive with receipts.

If the metrics contradict the reach, appeal with specifics. "My views are down" is weak. "Retention rose 12%, CTR held, but Browse impressions fell 80%" gives support something usable.

And please, build an audience you can reach without permission. Email list, Discord, Patreon, whatever fits. YouTube can remain the main stage. It shouldn't also own every door into the building.