Instagram says it is testing letting users reset the algorithmic suggestions that power Feed, Reels, and Explore, and the update “will soon roll out globally”
The app will offer users a chance for a “fresh start” with the algorithm. — If your Instagram recommendations …
Context & Ripple Effects
Instagram’s move from a chronological feed to algorithmic personalization made ranking central to how people encounter content across the app. It later added narrower controls, including tests of an “Interested” signal for recommended posts, but a reset offers a broader intervention when recommendations no longer fit.
This test is an early step in an arc toward making recommendation systems more legible and configurable: later coverage describes topic-level visibility into the activity shaping Reels recommendations and those controls extending to the main feed.
First-order effects
- Users in the test can clear the recommendation profile influencing Feed, Reels, and Explore, creating a new starting point for what Instagram suggests.
- Instagram gains a direct user-control mechanism for cases where its current recommendations feel stale or mismatched, ahead of a stated global rollout.
Second-order effects
- Creators and businesses dependent on recommendation distribution may see their content reach newly reset audiences less predictably while the system rebuilds signals.
- The reset raises the practical value of explicit preference controls: subsequent tools that show or let users select topics can help users steer recommendations after a fresh start.
Third-order effects
- If reset and customization controls become standard, recommendation feeds may shift from opaque, persistent profiles toward systems users can periodically inspect and revise.
- That would preserve Instagram’s role as the ranking gatekeeper while making the quality of user controls—not merely the existence of personalization—a more important competitive distinction.
The trend: Consumer platforms are gradually adding user-facing controls around algorithmic recommendations without giving up algorithmic distribution as the default interface.