Meta recently shared a behind-the-scenes look at how it’s enhancing the way Reels are recommended on its platforms, especially on Facebook. Instead of relying solely on traditional engagement metrics — like likes, shares, and watch time — Meta is now directly asking users how relevant or enjoyable a Reel was right after they watch it. These in-feed feedback surveys have been rolled out widely, giving Meta real, moment-to-moment user insights instead of just guesses based on behavior alone.
To make sense of this survey data, Meta applies statistical weighting to adjust for sampling differences and response bias, aiming to build a dataset that accurately reflects what users truly want to see. This allows the recommendation system to go beyond simple interaction signals and incorporate users’ own perceptions of relevance and interest. By doing so, Meta can fine-tune which content appears in Reels feeds, helping tailor the experience to individual tastes in a more nuanced way.
The impact of this shift has been notable. Meta reports that before implementing survey-informed insights, its recommendation systems only aligned with true user preferences about 48.3% of the time. After integrating survey feedback into its models, that alignment rate has risen to over 70%, suggesting users are now more likely to see videos that match their interests. This blended approach — combining machine learning with direct sentiment feedback — is designed to make Reels feel more personalized and engaging over time.
Despite the progress, Meta acknowledges that there’s still work to do. The company highlights ongoing challenges like better serving people with limited interaction histories, reducing biases in who sees surveys and whose feedback gets weighed most heavily, and increasing the diversity of recommendations so users aren’t limited to overly narrow content bubbles. Meta is using these insights to continually refine the system, aiming for a smoother balance between relevance and variety in users’ Reels feeds.
Even with these improvements, Meta still trails competitors like TikTok when it comes to recommendations. TikTok’s “For You” algorithm has become known for its deep engagement and capacity to keep users scrolling by identifying granular visual and context cues within videos — something Meta’s public systems don’t currently replicate at the same level. Still, Meta’s move toward more direct user feedback represents a significant step in evolving its recommendation tech.
