Decoding X’s algorithm: Here’s how Grok open-sources and curates the “For You” feed

Elon Musk’s X is finally lifting the lid on how its platform works, particularly how the “For You” feed is shaped for users worldwide. The company has made public the code on GitHub that explains how its algorithm functions, in a move aimed at increasing transparency.

X has now outlined how its newly open-sourced algorithm operates, breaking the process down into 10 summarised steps.

Reportedly, the feed is built through several stages using the company’s own Grok AI model, marking a shift from manually set rules to artificial intelligence. In simpler terms, X’s system scans more than 100 million posts, narrows them down to about 1,500 highly relevant posts, and then displays them on users’ feeds.

How Grok open-sources system works to curate “For You” feed

The first step explains that the system collects a user’s personal data, such as the accounts they follow and the posts they interact with through likes, reshares, and comments. This helps the system create a digital map of the user’s interests.

The next step involves collecting posts and interactions from users’ in-network accounts (people or pages they follow), which are fetched through a high-speed system called “Thunder.” It also pulls posts from outside the user’s network by identifying content from strangers that aligns with the user’s interests using AI.

Next, X’s system adds additional information to the shortlisted posts, such as whether a user subscribes to X Premium, whether a post is paywalled, video quality, and other metadata. It then removes duplicate posts, outdated content, and posts containing keywords that the user has muted. This ensures the feed excludes content the user has already scrolled past or chosen to avoid.

Grok then plays a major role in the ranking process by assigning importance scores to each post. Using an advanced AI model, the system predicts how likely a user is to engage with a post through likes, replies, reposts, or clicks. Based on these predictions, each post is scored, boosting high-quality content and reducing the visibility of less relevant posts.

Finally, the “For You” feed is created using the top-ranked posts. A final safety check ensures that no deleted posts, spam, or violent content appear in the feed.