4 simple steps to fix bad YouTube recommendations
Engadget offers four practical steps to retrain YouTube's algorithm and see more relevant suggestions.
The first step is ensuring watch history is enabled. According to the guide, YouTube needs permission to track viewing history to make personalized suggestions. Users can enable this by clicking their profile icon, selecting Google Account, then navigating to Data & privacy, Personalization settings, and YouTube History. The guide also suggests checking the Auto-delete option, which allows users to clear data automatically every 3, 18, or 36 months, giving the algorithm a more focused dataset.
The second and most important step is providing positive and negative feedback. The guide advises users to like videos they enjoy, watch them fully, subscribe to preferred creators, enable notifications, and leave comments. On the other hand, users should hit Dislike on uninteresting videos, select "Not interested" from the three-dot menu on unwanted recommendations, and choose "Don't recommend channel" for entire channels they dislike. Negative signals can take time to affect the algorithm, but combining both types of feedback helps improve suggestions.
The third step involves minimizing incorrect watching. Small amounts of unwanted activity can be erased manually by removing items from search history or watch history. However, the guide recommends preventing this altogether, especially when multiple family members use the same account on a TV or iPad. Adding separate accounts, using the account lock feature, or using private browsing and Incognito mode on mobile can help avoid sending the wrong signals to YouTube.
The fourth step is managing watch history selectively. Through the Manage YouTube History link in the personalization settings, users can see a full list of their platform activity and remove individual entries by clicking the X next to them. The guide also mentions using the calendar icon to filter entries by date. Following these steps may help users regain control over their recommendations, though the algorithm may take time to adjust.