Dissecting X's Retrieval Algorithm—Hidden Rules of Content Display

With Elon Musk open-sourcing Twitter’s algorithm, users like @0xTodd have begun sharing detailed analyses on X. This transparency has revealed the complex mechanisms behind how content is curated and displayed to users. Notably, there are two parallel content extraction systems known as the retrieval mechanisms. Understanding these systems enables users to develop more effective strategies to reach their target audiences.

Thunder vs. Phoenix Retrieval—Two Content Flow Systems

At the core of X’s algorithm are two critical retrieval systems. The first is “Thunder,” which retrieves only the content from accounts you follow. The second is “Phoenix Retrieval,” which extracts candidate content from the entire platform, including accounts you do not follow.

These retrieval systems are adjusted by an invisible assistant called “Grock” for each user. This system predicts the engagement potential of posts and dynamically decides whether to prioritize content from Thunder or to display a broader range of content from Phoenix Retrieval. In other words, your timeline is not just a chronological feed; this algorithm constantly calculates the optimal content distribution.

Reputation Score and the Hidden “Ranking” System

Each user is assigned an unseen reputation score ranging from -128 to +100. This score plays a crucial role in the algorithm, directly affecting how likely your content is to be extracted by Phoenix Retrieval and thus how often it appears to new users.

New accounts start with a low score, which can only be increased through active engagement with users who have high reputation scores. Conversely, interactions with low-quality accounts may lower your score, and associating with malicious content should be avoided. Interestingly, negative feedback is weighted much more heavily than positive. For example, a single block can have a greater impact on your score than multiple “likes.”

Video Completion Rate, Topic Fatigue, and Verification Status—Fine-Tuning Elements of the Algorithm

X’s algorithm is not based on simple rule application but has evolved to be quite flexible. In 2023, video content was heavily favored, and posts containing links were suppressed. Currently, the algorithm has become more sophisticated, shifting toward dynamic scoring based on individual user preferences.

The “Topic Fatigue” mechanism is also important. It lowers the ranking of multiple posts from the same author within a short period and penalizes repetitive content. This encourages timely posting and maintains content diversity across the platform.

Verified accounts (blue check) tend to be prioritized by Phoenix Retrieval. Unverified accounts need rapid engagement growth to reach the retrieval candidate pool. Additionally, the “dwell time”—the time users spend viewing a post even without direct interaction—is recorded and contributes points to the algorithm.

The system evaluates each tweet as an independent candidate, avoiding recommendations of duplicate or outdated content, and highly values video completion rates. These multiple factors interact to shape your timeline.

Elon Musk’s efforts to increase transparency have revealed these details, marking a significant turning point for the platform. A deeper understanding of retrieval mechanisms and scoring systems allows content creators to strategize more effectively, while users can better understand the logic behind the content they see.