Social media
X (Twitter) publishes its recommendation algorithm
In brief
X (formerly Twitter) open-sources the code of its "For You" feed recommendation algorithm, a first for a major social platform.
Key points
- X (formerly Twitter) publishes on GitHub the source code of its For You feed recommendation algorithm, unprecedented transparency.
- The code reveals the importance of replies, retweets and likes, with different weights (a reply counts more than a like).
- Negative signals are documented: external links and blocks/mutes can reduce reach.
- The time spent on a tweet and opening the author's profile are among the ranking signals.
Analysis
The publication of the code offers rare visibility into the real mechanics of an algorithmic feed: conversational engagements, in particular replies, are strongly valued, while content that pushes the user off the platform is disadvantaged.
For a brand present on X, this steers the strategy toward posts that trigger replies and reading time, rather than posts with a single outbound link. It is a direct source for calibrating format and the call to conversation.
What to do
- Favor native formats that elicit replies (questions, threads, polls) rather than posts with a simple outbound link.
- Place external links in a reply or at the end of a thread to limit the reach penalty.
- Measure the replies/impressions ratio as an indicator of affinity with the algorithm, beyond likes alone.
Impact
Unprecedented visibility into ranking signals (replies, retweets, time spent). Helps optimize organic reach.