How Instagram's recommendations keep users engaged

Instagram's Explore page and Reels feed don't look the same for any two people, because the platform is constantly predicting what will keep each individual user watching.

Meta's own engineering documentation describes a recommendation system built on machine learning that continually ranks content across Feed, Stories, Reels, Explore, and even notifications.

Understanding how that system actually works is central to why Instagram keeps coming up in the social media addiction litigation. Meta has said Instagram now runs more than 1,000 machine-learning models across its recommendation surfaces. These models retrieve billions of potential posts, then narrow that pool down through multiple ranking stages before a much smaller set reaches an individual user's feed. The system evaluates candidates and predicts specific actions a person might take, including clicks, likes, and even choosing to see less of something.

A recommendation system that learns from every interaction

Instagram's models don't just track a single like or follow, they continually update as a person's behavior changes. Likes, saves, and shares all feed into what the system predicts a user wants to see next. Meta describes using a person's interaction history to identify similar posts for future recommendations, creating a loop:

• A user interacts with content
• The system records that interaction as a behavioral signal
• Similar content is identified and ranked higher
• The user engages again, generating more signals for the next round

Reels works the same way but takes it further, since it's specifically designed to expose users to creators they don't already follow, meaning a person doesn't need to search for new content at all.

How this can shape what a teenager actually sees

A 2026 study looked at how liking and saving appearance-related content changed Instagram's Explore recommendations. Researchers found that engaging with appearance-focused posts affected the composition of a user's Explore feed more strongly than simply following certain accounts, meaning a handful of interactions could shift the type of content someone saw going forward.

A separate 2026 qualitative study of 27 young people ages 14 to 19 examined how adolescents experienced Instagram Explore and TikTok's recommendation feeds. Researchers identified algorithmic reinforcement of engagement as one of the central mechanisms sustaining exposure, alongside emotional feedback loops between a person's mood and the content being recommended. Participants described this exposure continuing even while they understood how the algorithm was shaping what they saw.

Why this system is central to the litigation

MDL 3047 alleges that Instagram's recommendation architecture is part of a broader design built to encourage compulsive engagement among adolescents. Plaintiffs point to several components of this system as working together, including:

• Behavioral data collected from likes, saves, shares, and viewing time
• Predicted engagement, where the system estimates what a user is likely to click or interact with next
• Continuous content delivery across Explore and Reels, which supplies new material without requiring a search

The relevant theory for this litigation isn't simply that teenagers spend a lot of time on Instagram. It's the interaction between these pieces that keeps a session going long after a user opened the app for one specific reason.

Talk to Atraxia Law about a social media addiction claim

If your child developed a diagnosed mental health condition tied to Instagram's recommendation system, Atraxia Law can look into whether the circumstances fit the current social media addiction litigation. We review the platform features involved alongside your child's medical history to determine eligibility, then connect qualifying families with an attorney handling claims in MDL 3047. Contact Atraxia Law today for a free, confidential case evaluation.