A new study from Emplifi, based on analysis of 10,110 Facebook Reels published by 704 brand pages, has identified the specific creative elements that most reliably improve Reels performance. The findings are straightforward: show a person, start talking immediately, and if your clip is short, make it loop.
The most significant finding involves audio. Videos that include human speech within the first three seconds drive substantially better retention than those that open with music or silence. Emplifi’s data shows that human speech increases 10-second retention by nearly 25% compared to music-led openings. Sound-on rates are also higher for speech-driven Reels, suggesting that viewers are more likely to engage with audio when they hear a human voice rather than a backing track.
Human presence matters almost as much as speech. When a person appears on screen for at least one second within the first three seconds of a Reel, 10-second view retention improves by approximately 10%. That early-face advantage does diminish over longer viewing windows, with 30-second views actually declining by 2.4% on videos featuring a person, suggesting that a human face hooks viewers initially but does not guarantee sustained watch time on its own.
For very short content, looping is a powerful lever. On micro-length videos of up to seven seconds, seamless loops boosted replay rates by 18.7% and engagement rates by 16.1%. The technique works because short clips that restart without a visible cut encourage repeated watches, which the algorithm interprets as strong interest.
The study also found that vertical video format produces 20.9% higher reach than non-vertical alternatives, and that text overlays drive modest improvements in engagement, though the effect is smaller than voice or face presence.
On average, users share 3.5 billion Reels daily across Facebook and Instagram, and roughly half of all time spent on Instagram is dedicated to Reels consumption. Meta recently converted all Facebook video uploads into Reels and has been investing heavily in recommendation algorithms that use survey feedback to improve personalization.
