# I asked AI to Crack the X Algorithm (It Did)

Source: https://www.youtube.com/watch?v=xAsvOdZYk4I
Recap page: https://rapidrecap.app/video/xAsvOdZYk4I
Generated: 2026-01-20T20:03:48.897+00:00

---
## Quick Overview

The speaker successfully cracked the X (Twitter) algorithm by reverse-engineering its scoring logic, revealing that the algorithm predicts 18 potential user actions (positive and negative signals) on every post and weights them to calculate a final score, with key factors including in-network distribution, author diversity decay, and dwell time being crucial for reach.

**Key Points:**
- The X algorithm scores posts based on 18 potential actions, categorized into positive signals (like, reply, retweet, share, dwell time, follow) and negative signals (not interested, block, mute, report).
- The speaker verified the algorithm's success by achieving 2.13K profile visits from one viral post, netting $576.41 in revenue from various sources like Amazon and Mediavine.
- Key modifiers include the OON_WEIGHT_FACTOR penalizing out-of-network posts (multiplier < 1.0) and AUTHOR_DIVERSITY_DECAY, which exponentially reduces scores for the 2nd, 3rd, and 4th posts in the same batch.
- For creators with over 50K followers (Scenario 1), the focus is on high-weight actions like replies to big accounts and prioritizing in-network distribution to avoid the OON penalty.
- For new accounts starting from zero (Scenario 2), the main challenge is the OON penalty, requiring significantly higher raw engagement rates, focusing on Phoenix Retrieval via user/post embedding similarity, and becoming a niche expert.
- Dwell time is considered underrated, with longer posts getting fully read providing a strong dwell time signal, and the formula for the final score being the sum of (weight of engagement * probability of engagement).
- The speaker outlines a 3-month practical playbook for growth: Week 1-4 (Build Signal via replies/original posts), Week 5-8 (Compound via consistent embedding), and Month 3+ (Scale).

![Screenshot at 03:00: Visual diagram illustrating the X Algorithm's core recommendation system, showing inputs like 'Thunder \(In-Network\)' and 'Phoenix Retrieval \(Out-of-Network\)' feeding into the 'Phoenix Scorer \(Grok-based Transformer\)' before outputting content to the 'For You Feed' on the mobile app.](https://ss.rapidrecap.app/screens/xAsvOdZYk4I/00-03-00.jpg)

**Context:** The video features the creator detailing the proprietary algorithm used by X (formerly Twitter) for ranking content, which they cracked using AI and reverse-engineering leaked internal scoring data. The creator, Jacky Chou, presents this information as a detailed Standard Operating Procedure (SOP) for content growth, breaking down the scoring system into positive and negative signals, key modifiers, and specific growth scenarios based on follower count.

## Detailed Analysis

The creator reveals the inner workings of the X algorithm, which operates by calculating a score based on 18 distinct user engagement signals. These signals are divided into positive (boost ranking) and negative (tank ranking) categories. Positive signals include likes, replies, retweets, shares, profile clicks, dwell time, and following an author. Negative signals include 'Not Interested' clicks, blocking, muting, or reporting an author. The scoring is further modified by factors like OON_WEIGHT_FACTOR (penalizing out-of-network posts), AUTHOR_DIVERSITY_DECAY (penalizing rapid posting on the same topic), and MAX_POST_AGE (filtering old posts). For users with over 50K followers (Scenario 1), the strategy focuses on maximizing high-weight actions like replies to large accounts and leveraging in-network distribution to avoid penalties. For new users (Scenario 2), the challenge is overcoming the OON penalty by focusing on Phoenix Retrieval—matching post embedding space with user engagement history—and establishing a clear niche. The speaker emphasizes dwell time as an underrated factor, noting that longer content leading to full reads provides a strong signal. The overall goal is to build a consistent author embedding through deliberate actions like replying to valuable accounts and quoting tweets with analysis, following a phased growth playbook across the first few months.

### X Algorithm Scoring

- Positive Signals include Favorite (like), Reply, Retweet, Quote Tweet, Share, Share via DM, Share via Copy Link, Click, Profile Click, Photo Expand, Video Quality View (for videos > min duration), Dwell Time, and Follow Author
- Negative Signals include Not Interested clicks, Block Author, Mute Author, and Report

### Key Modifiers

- OON_WEIGHT_FACTOR penalizes out-of-network posts (multiplier < 1.0)
- AUTHOR_DIVERSITY_DECAY exponentially reduces scores for the 2nd, 3rd, and 4th post in the same batch
- MAX_POST_AGE filters posts older than a threshold

### Scenario 1 (50K+ Followers)

- Advantage is in-network distribution (avoiding OON penalty)
- Content Strategy focuses on optimizing high-weight actions like replies, shares, dwell time, and profile clicks, writing posts that trigger these actions

### Scenario 2 (Starting From Zero)

- Challenge is the OON_WEIGHT_FACTOR penalty due to no followers
- Path to Distribution involves understanding this disadvantage, leveraging Phoenix Retrieval (User embedding + Post embedding similarity, Author embedding, Engagement history pattern matching), and niching down hard to build author embedding

### Practical Playbook (Zero to 10K)

- Week 1-4 (Build Signal: Reply to 10-20 accounts, 1-2 original posts/day, quote tweets with analysis)
- Week 5-8 (Compound: Author embedding forms, Phoenix retrieval starts finding posts, double down on dwell time + replies)
- Month 3+ (Scale: Get in-network distribution, game changes to Scenario 1)

![Screenshot at 00:00: The creator, Jacky Chou, wearing a LocalRank hat and speaking into a microphone, introduces the video topic about cracking the X algorithm.](https://ss.rapidrecap.app/screens/xAsvOdZYk4I/00-00-00.jpg)
![Screenshot at 00:52: The creator consuming a scoop of powder directly from a container, illustrating the behind-the-scenes nature of the content.](https://ss.rapidrecap.app/screens/xAsvOdZYk4I/00-00-52.jpg)
![Screenshot at 01:19: A slide displaying the 'LLM Listicle Guest Post Packs' pricing structure, highlighting the 'Most Popular' option for LLM Guest Posts ranging from $99–$899.](https://ss.rapidrecap.app/screens/xAsvOdZYk4I/00-01-19.jpg)
![Screenshot at 02:53: A detailed diagram illustrating the X Algorithm's core recommendation system, showing data flow through components like Thunder, Phoenix Retrieval, Home Mixer, Filtering, Phoenix Scorer, and Selection & Filtering leading to the 'For You Feed'.](https://ss.rapidrecap.app/screens/xAsvOdZYk4I/00-02-53.jpg)
![Screenshot at 05:00: A screenshot of the creator's revenue breakdown from various ventures in December 2025, showing total revenue of $576.4K.](https://ss.rapidrecap.app/screens/xAsvOdZYk4I/00-05-00.jpg)