X Algorithm Leaked (Jan 2026): The New “Rust” Update & Why Retweets Are Worth 20x

For years, we guessed how the algorithm worked. We assumed likes helped. We assumed links hurt. But we were just guessing. Yesterday, that changed. X officially open-sourced their new recommendation algorithm (xai-org/x-algorithm), confirming that the platform migrated from the old "Heavy Ranker" to a new engine powered by Grok, that part is real and confirmed. What I got wrong the same day, in the tweet below: treating the widely-shared engagement ratios as numbers I'd personally verified in the live code. They weren't, xAI didn't publish the exact production weights. Here's the original post, kept for the record, with the correction that follows it. If you're an Amazon Influencer or Associate using X for traffic, this shift matters for you specifically: strategies built around keyword stuffing or link-first posts are working against you now, not for you.

---

1. The Code Proof: Retweets Are King (20x)

The new scoring engine uses a cleaner structure than the old neural nets. I found the default engagement weights located in scoring/weights.rs. Correction (added July 20, 2026): when this post first went up, I presented the struct below as hard-coded numbers I pulled straight from the live repo. That overstated it. xAI's actual January 2026 release did not ship a weights.rs file with these literal coefficients, the real crate::params module holding the production weights was redacted, and xAI has said publicly the exact ratios aren't released. The numbers people (including me) have been citing since January trace back to the 2023 "the-algorithm" open-source release, and analysts treat them as directionally accurate for the new Phoenix system, meaning retweets and replies still outweigh likes by a wide margin, but not as a literal quote from the current codebase. Here's the struct as it's been widely reconstructed and cited, with that caveat attached:

// Reconstructed / cited widely, NOT a confirmed excerpt of the 2026 x-algorithm repo
// Original coefficients trace to the 2023 Twitter "the-algorithm" release

pub struct EngagementWeights {
    pub like_score: f32 = 1.0,          // Baseline vanity metric
    pub reply_score: f32 = 13.5,        // Conversation signal
    pub repost_score: f32 = 20.0,       // The Viral Gold Standard
    pub native_video_score: f32 = 10.0, // Native media boost
    pub external_link_penalty: f32 = -5.0, // The "Link Tax"
}

What that translates to, if the ratios still hold:

• If you get 100 Likes, your score is 100.
• If you get 5 Retweets, your score is 100.

A Retweet is roughly 20 times more powerful than a Like under the widely-cited (if unconfirmed) ratios above, and the direction only got stronger since: xAI's May 15, 2026 update shipped a runnable "mini Phoenix" model, and outside analysis of that release found a single author-to-replier exchange can carry roughly as much ranking weight as 150 likes. Whatever the exact numbers turn out to be, replies and reposts beating likes by a wide margin is the one part of this that keeps holding up. If you aren't explicitly creating content that demands a reply or a share, you're leaving ranking weight on the table.

---

2. The "Link Tax" Is Real (-5.0)

For us affiliates, this is the most painful part of the update. The code confirms an explicit external_link_penalty. If you put your Amazon Associate link in the main tweet, you start with a negative score. The algorithm categorizes you as "Spam" before anyone even sees the post. The Fix:

1. Write a high-value thread.
2. Wait 2 minutes (this bypasses the instant-spam filter).
3. Post your link in the Reply.

---

3. The Amazon Strategy: How to Make Money With This

Knowing the code is cool. Making money with it is better. Here is how I am pivoting my content strategy immediately for the Amazon Influencer Program.

For Amazon Influencers (Onsite Video Focus)

The code shows a native_video_score of 10.0. This is huge.

• The Play: Stop posting links to your storefront. Start posting Native Video Teasers.
• Example: Upload the first 15 seconds of your "Desk Setup" review directly to X as a native video.
• The Hook: "I just found the best cable organizer on Amazon. Watch how it cleaned my setup in 10 seconds. (Link in reply)."

For Amazon Associates (Offsite Traffic Focus)

Your goal is "Velocity." The new TimeDecay function in the code shows that the first 15 minutes are critical.

• The Play: You need Conversation, not just clicks.
• Strategy: Don't just post a deal. Post a Comparison.
• Example: "Unpopular Opinion: The DJI Osmo Pocket 3 kills the iPhone for vlogging. Here is the side-by-side footage. Which one looks better to you?"
• Why it works: This forces Replies, the signal that consistently ranks well above Likes under every version of these numbers, cited or confirmed. As people argue in the comments, your visibility should climb. Once the post is flying, drop the affiliate link in the comments rather than the original post.

---

4. Unconfirmed: An AI-Writing "Slop" Filter

This one I can't back up as firmly as the rest. There's a circulating claim of a grok_filter.rs module that flags low-effort AI writing, including the classic ChatGPT tells (words like "unleash," "delve," "tapestry," "elevate"). I could not independently confirm this module or that specific word list against the public repo or any outlet reporting on it, so treat it as an unverified rumor, not a confirmed feature. The underlying advice is worth following regardless of whether this specific filter exists: obviously AI-written posts get less engagement on X than posts that sound like a real person, that pattern holds with or without a named filter behind it. Write like a human. Be punchy, be specific, be real.

---

Conclusion

The rules of the game have changed. X is no longer a "link dump" platform. It is a "Velocity Engine." If you can get real retweets and real replies in the first 15 minutes, under every version of these ratios that's worth more than the same number of likes, you can plausibly drive more traffic to Amazon in one day than you used to in a month. Treat the exact multipliers as directional, not guaranteed. Audit your last 5 tweets. If they have links in the main text or zero replies, delete them and start over with the Phoenix Protocol.