Ad interactions

Instagram’s Help Center lists ad views and engagement among the factors informing suggested products, alongside location, interest in product categories and activity across Instagram, Facebook and Threads.2 Meta’s engineering team also identifies clicks, views and other interactions as recorded information used in its advertising models.3 A recorded view need not mean consciously remembered engagement: Mosseri said in October 2025 that people sometimes scroll past ads quickly without later realizing they have already seen them.4 When selecting ads, Instagram’s Help Center lists followed accounts, liked posts and activity on third-party websites and apps, so explicit feedback on an advertisement is not its only input.1

For ad interactions, Meta’s engineering team’s November 2024 account describes learning from the order of engagement and conversion events—recorded outcomes—rather than relying mainly on manually assembled summaries of past activity.5 In its November 2025 GEM account, Meta’s engineering team describes processing histories of up to thousands of events involving ads and unpaid content to identify patterns along a person’s path toward a purchase.3 This distinction between observed actions and anticipated behavior also appears in Meta’s August 2020 patent, which proposes using past actions to predict how much someone will interact with content associated with an advertisement, but does not establish Instagram’s implementation.6 The Meta patent proposes using that prediction to estimate an ad’s value and determine its position among ads considered for display, with possible measures including money spent and the number of times someone visits the associated content.6 The Meta patent also proposes requiring a minimum predicted amount of interaction before someone qualifies to see an ad.6