Candidate generation

Instagram’s candidate generation for followed-account Feed posts starts with potential posts from accounts a person follows and removes posts that violate Community Guidelines, according to Meta’s Transparency Center.2 A simpler selection model then picks approximately 700 relevant posts, while additional checks help reduce how widely problematic content is shown.2 For suggested Feed posts, the system considers information such as the viewer’s interests and previous engagement with similar content.1 For Instagram’s Reels Chaining AI system, the starting pool can include reels from followed accounts or reels similar to those the viewer recently interacted with.3 The Reels Chaining system uses information such as length, similarity and fit with the viewer’s interactions to select approximately 100 candidates, and applies checks to remove content that cannot be recommended or may violate Community Guidelines.3

Instagram’s account candidate pool draws on followers of accounts a person already follows, popular accounts in their geographical region, and accounts related to topics or products they recently engaged with, according to Meta’s Transparency Center.4 The system then orders those accounts using information about the person and their public activity on Instagram, including follow counts and mutual friends with a suggested account.4 After that ordering step, filtering removes blocked and deactivated accounts, while further checks remove content that may violate Community Guidelines.4