# How the X algorithm actually works: the voice-first reading of the weights

> The X algorithm filters 500M daily posts to 1,500 ranked candidates using a weighted engagement score. The standard advice is to game the weights. The voice-first reading is that the weights amplify whatever voice you bring. Here's what the published weights mean for voice-first creators, whether the 2026 Phoenix rewrite changed them, whether the algorithm punishes AI content, what moves distribution for a small account, and how to optimize without letting the weights edit what you say.

URL: https://voicemoat.com/blog/understanding-the-x-algorithm

## Key facts
- Published: 2026-05-12
- Category: Growth
- Read time: 13 minutes
- Author: VoiceMoat team
- Publisher: VoiceMoat
- Topics: X algorithm ranking; Twitter algorithm 2026; Engagement weights; Voice-first growth; Content distribution

## Key points
- Did the 2026 Phoenix rewrite change these weights?
- Does the X algorithm punish AI-generated content?
- What actually moves distribution for a small account?
- Can you game the algorithm without hurting your voice?
- Should you rework your strategy every time the algorithm updates?

## Related articles
- Twitter reply strategy: why fewer voice-rich replies beat the 30-a-day playbook: https://voicemoat.com/blog/twitter-reply-strategy
- Twitter bookmarks as voice-research infrastructure: how to study voice without flattening yours: https://voicemoat.com/blog/twitter-bookmarks-voice-research
- How to increase Twitter reach: what compounds and what looks like it but doesn't: https://voicemoat.com/blog/twitter-reach-what-actually-compounds
- Twitter Blue vs X Premium in 2026: which tier, and is it worth it at all?: https://voicemoat.com/blog/twitter-blue-x-premium-decision
- The case against reply-bot automation at scale: https://voicemoat.com/blog/the-case-against-reply-bots
- What is audenAI? The brain inside VoiceMoat: https://voicemoat.com/blog/what-is-auden
- How to grow on X in 2026 without buying followers or running engagement pods: https://voicemoat.com/blog/grow-on-x-without-buying-followers-2026

## FAQ
Q: How does the Twitter algorithm actually rank posts?
A: It filters roughly 500 million daily posts down to a personalized 1,500 per user and ranks them by a weighted engagement-probability score. Replies count about 13.5x and rise to 75x when the author engages back, reposts around 20x, and likes and bookmarks near 30x. The penalties dwarf the boosts: reports cost 369x and negative feedback 74x, and every post loses about half its weight every 6 hours.

Q: Does the Twitter algorithm punish AI-generated content?
A: There is no weight for looking AI-generated, and the 2026 ranker runs no detector that demotes machine-drafted posts. But it punishes what AI-shaped content tends to produce: it earns fewer bookmarks and author-engaged replies, clears the dwell-time threshold less often, and accumulates the not-interested signals the audience gives interchangeable posts. Writing in a recognizable voice is not a trick on the ranker; it earns the engagement the ranker already rewards.

Q: What moves distribution for a small Twitter account?
A: Replies, not original posts. A small account's original posts reach almost no one because For You distribution is gated on early engagement it cannot yet generate, so a voice-rich reply on a large account borrows that account's distribution and the 13.5x reply weight does work an original post at zero base cannot. The sequence is replies-first until you have enough recognized-voice followers, then a shift toward original evergreen.

Q: Did the 2026 Phoenix rewrite change how the algorithm works?
A: The architecture changed more than the conclusion did. The rewrite moved to a transformer-based model that scores engagement actions jointly rather than summing a fixed weight table, and it leans harder on out-of-network retrieval by embedding similarity. What did not change is the direction of incentives: replies still dominate, negative feedback and reports still cost far more than any boost, and recency still decays fast. A coherent recognizable voice is rewarded even more under the rewrite.

Source: https://voicemoat.com/blog/understanding-the-x-algorithm
Site overview: https://voicemoat.com/llms.txt. AI usage policy: https://voicemoat.com/ai.txt.