Watch time

Instagram’s total watch time includes all the time a reel was played, including time spent replaying it, according to Meta’s Newsroom.1 Average watch time divides that viewing time by the total number of plays.1 An average watch time of 17 seconds therefore means the reel was watched for an average of 17 seconds per play, not that every viewer watched for that long.1

Instagram uses several details about viewing duration when deciding which content to show: Meta’s Transparency Center lists Reels inputs such as three-second watches, skips within two seconds and authors whose reels a person watched for less than three seconds.4 For Feed, Meta’s Transparency Center lists time spent on a post, its media type when estimating expected viewing time, and the likelihood of skipping it compared with nearby posts.5 For Stories, Meta’s Transparency Center lists a person’s average and maximum viewing times alongside their history of viewing and interacting with an author’s stories.6 Instagram for Creators also identifies whether people keep watching as information used to predict which reels someone might like.7 A Meta patent describes a related possibility for a personalized stream: predicting whether someone will watch an item longer than their usual viewing duration, without confirming that Instagram currently uses this method.8 Another Meta patent describes ordering collections of temporary content first by the likelihood of selection and then by the likelihood of spending time viewing them, again without establishing that Instagram currently uses that method.9

Instagram considers both seconds watched and the proportion of a video watched, Mosseri explained in February 2025.10 His example was that watching ten seconds of a one-minute video produces the same number of watched seconds as watching all of a ten-second video.10 A longer reel therefore is not penalized simply because the viewer watched a smaller percentage of it.10

In April 2023, Meta’s Newsroom announced creator-visible total and average watch-time measurements for Instagram Reels, which Mosseri also announced that month.111 Improving a reel’s watch time starts with its opening: Mosseri said the first three seconds should get attention and establish what the video will discuss, without necessarily using bait or a cliffhanger.12 He recommended using text when possible because viewers may initially have the sound off.12 Instagram for Creators described how Warren Abercrombie paired interesting openings with payoffs that encouraged viewers to finish his reels.13 When evaluating watch time for trial reels—reels initially shown only to non-followers—Mosseri recommended comparing them with other trials at similar amounts of time after posting.14 His broader guidance was to look for patterns in insights to identify what works while staying true to yourself.3

To extend the Instagram measurement discussion, Meta’s research on Facebook Reels illustrates how viewing duration can become a training target—a result a recommendation model learns to predict—without documenting Instagram’s current system.15 In Meta’s research, a viewing score depended on video length: clips under ten seconds required more than one completed viewing, clips from ten to under twenty seconds required completion, and longer clips required at least twenty seconds watched.15 Meta’s research also used a separate seconds-viewed target, with values increasing in steps from less than ten seconds to ninety seconds or more.15 Meta’s engineering team explained that Facebook Reels models trained only on watch time and other engagement measures can favor short-term responses while missing deeper interests and long-term satisfaction.16 To address that limitation, Meta’s engineering team described using direct survey feedback about whether Facebook reels matched viewers’ interests, including dimensions such as audio, production style, mood and motivation.16