Meta explains how AI curates your feeds

If you have ever stared at a totally bizarre Instagram Reel and wondered how the algorithm got there, Meta is finally showing its work.

Meta just pulled back the curtain on its recommendation engines, releasing 22 detailed "system cards" that explain how AI decides what billions of people see across Facebook and Instagram.

The math behind your doomscroll

Instead of treating recommendations as a mystical black box, Meta broke down the predictive machinery driving Feeds, Stories, Reels, and content from accounts you do not even follow.

In plain English: AI systems generate dozens of predictions about your future behavior every second. They calculate the odds that you will share a post, leave a comment, or hit like, combining behavioral signals with survey feedback to estimate how "valuable" a post is before putting it at the top.

Meta also published the vast majority of signal categories used in Facebook Feed ranking in its Transparency Center, alongside how frequently those signals are factored in. But there is a catch: Meta is intentionally withholding signals used to catch bad actors to keep people from gaming abuse filters.

On the user side, the company is rolling out its "Why Am I Seeing This?" tool to Instagram Reels, Explore, and Facebook Reels. Instagram is also testing an explicit "Interested" button on Reels to complement the "Not Interested" button introduced in 2021.

For researchers, Meta is launching a new Content Library and API via partners like the University of Michigan, offering access to public posts, pages, and groups across both platforms.

Why it matters

For creators and everyday users, knowing the explicit signals means fewer guessing games. You now have a better playbook for why your reach fluctuates, along with clearer controls to actively steer your algorithm if your feed starts going sideways.

For big tech, this is classic preemptive compliance. Regulators around the world are bearing down on black-box social feeds, and Meta is laying down a transparency benchmark that puts pressure on competitors like TikTok and ByteDance.

We think this move is clever, but people should not confuse transparency with total surrender of power. Explaining what signals exist is very different from revealing the exact weights that shift traffic, and the researcher tools remain strictly gated. It is a welcome step forward, but the house still controls the table.