I finally found a way to clean up my YouTube feed from those
Weedout is a Safari extension designed for macOS that targets one specific thing: it strips away any video that YouTube has explicitly tagged with a "Made with AI" label. Instead of trying to build a complex, heavy-duty AI detection model itself—which would likely be inaccurate and resource-intensive—it simply listens to what YouTube is already telling us. If YouTube puts that label on a video, Weedout hides it.
What I appreciate about this setup is how it handles the different parts of the YouTube ecosystem. It doesn't just work on the homepage; it cleans up:
- The main discovery feed
- Search results
- The sidebar of related videos
- Your existing playlists
- YouTube Shorts
Because it relies on YouTube's own metadata, the extension is incredibly lightweight. It runs locally on your machine, meaning your browsing data isn't being sent to some third-party server to be analyzed. There is also a built-in safety mechanism: it won't accidentally hide videos that haven't been labeled. If a video is real human content but just looks a bit strange, it stays. You only lose the stuff that the platform has officially flagged.
For anyone interested in the technical side or looking to build something similar, the developer has actually open-sourced the code. It’s a great example of a focused, single-purpose AI workflow tool. It’s not a massive project, but it solves a very specific pain point that a lot of us are feeling right now.
If you're a developer, you can check out the repository here:
https://github.com/masteranza/weedout-for-youtubeJust a heads-up for those looking to contribute: the original developer isn't accepting pull requests. The repo is essentially a foundation for people to fork and create their own independent versions or adapt it for other browsers.
It’s a small, $1.99 extension, but for anyone who feels like their YouTube algorithm is being hijacked by synthetic content, it's a practical tutorial in how to take control of your own digital environment. It's refreshing to see a developer tackle a niche problem with such a streamlined, "no-nonsense" deployment.