Which Engagements Matter Most on X?
Not all engagement signals are treated as the same thing by X's recommendation system. The current open-source X algorithm shows that the ranking system predicts multiple actions—including likes, replies, reposts, clicks, profile clicks, shares, dwell, follows and negative feedback—and combines those predictions using separate weights. (GitHub)
That means the simple idea of “one like = one point” is not an accurate model of how the For You ranking system works.
The Short Answer: Scored Signals
| Engagement Signal | In the Scoring Model? |
|---|---|
| ❤️ Like | Yes |
| 💬 Reply | Yes |
| 🔁 Repost | Yes |
| 💬 Quote | Yes |
| 🔗 Click | Yes |
| 👤 Profile click | Yes |
| 📤 Share | Yes |
| ✉️ Share via DM | Yes |
| 🔗 Copy link | Yes |
| ⏱️ Dwell | Yes |
| 👀 Video quality view | Yes, conditionally |
| ➕ Follow author | Yes |
| 🙈 Not Interested | Yes, negative |
| 🚫 Block author | Yes, negative |
| 🔇 Mute author | Yes, negative |
| ⚠️ Report | Yes, negative |
The current ranking_scorer explicitly has separate weights for these signals. (GitHub)
1 to 13. Individual Signals and Weighted Combination
1. There Isn't One "Most Important" Engagement: The system predicts probabilities for multiple actions and applies separate weights, rather than using a universal hierarchy like Repost > Reply > Like.
2 to 9. Likes, Replies, Reposts, Clicks, Dwell, Profile Clicks, Shares & Follows: Each action has dedicated scoring parameters (`favorite_score`, `reply_score`, `retweet_score`, `click_score`, `dwell_score`, `profile_click_score`, `follow_author_score`), allowing the system to weigh distinct user intents separately. (GitHub)
10 to 13. Negative Engagement & User Signals: Negative predictions ("Not Interested", "Block", "Mute", "Report") act as penalties in the weighted scoring stage, ensuring content that triggers negative friction is demoted.
14 to 21. Retrieval vs. Ranking and Creator Takeaways
14. Retrieval vs Ranking: Engagement signals influence both what candidates are retrieved into the pool and how candidates are scored during ranking.
15 to 21. Creator Strategy: Rather than optimizing for a single metric, successful recommendation content encourages strong hooks, meaningful discussions, shareable insights, and avoids triggering negative user feedback.
How X Ranks and Recommends Posts
Official Open-Source References
- • X / xAI — For You Algorithm repository — current architecture, Phoenix, Thunder, filtering, scoring and candidate pipeline. (GitHub)