What X's Open-Sourced Recommendation Algorithm Reveals About What Gets Amplified

Why X's Open-Source Code Matters

In March 2023, X (formerly Twitter) released portions of its recommendation algorithm on GitHub. Unlike vague platform statements about "meaningful interactions," this code includes actual numerical weights—the multipliers that determine whether your post reaches ten people or ten thousand.

For creators, this transparency removes guesswork. You can see that a reply is worth 27× more than a like in the ranking model, or that spending two minutes reading a post signals far stronger interest than a quick scroll-past. The code doesn't tell you how to go viral, but it does explain why certain content types consistently outperform others.

This article breaks down the most actionable findings from X's public algorithm code and what they mean for your posting strategy.

The Core Ranking Signals: What X Actually Measures

X's recommendation system evaluates every post using a weighted sum of engagement signals. Here are the primary factors and their relative importance:

Signal	Weight (approximate)	What it means
Reply	27×	Someone took time to write a response
Retweet	20×	User endorsed your content to their audience
Like	1×	Baseline engagement signal
Video view (≥50%)	0.005× per second	Sustained attention on video content
Profile click	12×	User wanted to learn more about you
Dwell time	Variable	Time spent reading without scrolling away
Negative feedback	-74× to -369×	Blocks, mutes, reports, "show less"

The ratio between replies (27×) and likes (1×) explains why a post with 50 replies and 200 likes often outperforms one with 2,000 likes and five replies. X interprets replies as proof that your content sparked conversation, not just passive acknowledgment.

Why Replies Dominate the Algorithm

The 27× multiplier for replies isn't arbitrary—it reflects X's design goal of surfacing content that generates discussion. A reply requires cognitive effort: reading your post, formulating a thought, and typing a response. That investment signals genuine interest.

Practical takeaway: Posts that end with a clear question or invite disagreement tend to generate more replies. Compare "Here's my take on the new iOS update" with "The new iOS update breaks my workflow—anyone else seeing this?" The second version gives readers a specific prompt to respond to.

However, reply quality matters. The algorithm down-weights one-word replies or responses from accounts that reply to everything. A thoughtful, multi-sentence reply from an engaged follower carries more weight than "lol" from a bot.

Dwell Time: The Hidden Multiplier

Dwell time—how long someone spends looking at your post—acts as a quality filter. If users consistently scroll past your content in under a second, X interprets that as low relevance, even if the post eventually accumulates likes.

The code doesn't publish an exact dwell-time multiplier, but internal documentation references "time spent" as a key signal. Posts that hold attention for 10+ seconds receive a boost, while those dismissed in under two seconds get suppressed.

Why this matters: Long-form text posts (the kind X now supports with extended character limits) benefit when readers actually finish them. A 500-word post that keeps someone engaged for 90 seconds signals higher value than a one-liner that gets a quick like and scroll.

Thread format exploits this mechanic effectively. Each click to "Show this thread" registers as continued engagement, and readers who expand all five tweets in a thread accumulate significant dwell time across the sequence.

Negative Feedback: The Amplification Killer

Negative signals carry disproportionate weight. A single "Show less of this" action can offset dozens of likes. The code assigns these penalties:

• Block: -369×
• Mute: -74×
• Report: -369×
• "Show less often": -74×

One block effectively erases the positive signal from 13 replies. This asymmetry means that polarizing content—posts designed to provoke strong reactions—faces a steep algorithmic penalty if even a small percentage of viewers actively reject it.

Practical takeaway: Clickbait headlines or deliberately inflammatory takes might generate short-term engagement, but the negative feedback accumulates quickly. X's algorithm prioritizes content that your existing audience wants to see more of, not content that angers people into engaging.

Video and Media: Seconds Watched, Not Just Views

X counts video views, but the weight per view is minimal (0.005× per second watched). A 30-second video watched to completion adds roughly 0.15× to your ranking score—far less than a single like.

However, video dwell time compounds with the general dwell-time signal. If someone watches your 60-second video and then reads the caption, you've held their attention for 70+ seconds, which triggers the dwell-time boost separately.

Why this matters: Short, looping videos (under 10 seconds) rarely accumulate enough watch time to compete with text posts that generate replies. Longer videos (2–3 minutes) only help if viewers actually watch them—posting a five-minute video that most people skip after three seconds hurts more than it helps.

Profile Clicks: The Underrated Signal

A profile click (12× weight) signals that your post made someone curious about who you are. This metric correlates strongly with follower conversion—people who click your profile are evaluating whether to follow you.

Posts that introduce a novel idea, share a surprising stat, or showcase expertise tend to drive profile clicks. Generic observations ("Monday vibes") rarely do.

Practical takeaway: If your goal is audience growth, optimize for profile clicks by including a credibility marker in your posts. "After analyzing 500 accounts…" or "I've spent six years building…" gives readers a reason to investigate further.

What the Algorithm Doesn't Care About

Several metrics that creators obsess over carry little to no weight in X's recommendation system:

• Follower count: The algorithm evaluates each post independently. A 50-follower account can outrank a 50,000-follower account if the smaller account's post generates stronger engagement signals.
• Posting frequency: There's no bonus for daily posting. Consistency helps audience retention, but the algorithm doesn't reward volume.
• Hashtags: X's code shows minimal weighting for hashtag matching. Hashtags help with search discoverability but don't boost algorithmic distribution.
• Time of day: The algorithm prioritizes recency but doesn't encode specific "best times to post." A strong post from 3 a.m. will surface in morning feeds if it's still generating engagement.

Comparing X's Approach to Other Platforms

Platform	Primary signal	Transparency
X	Replies (27×)	Open-source code available
Instagram	Shares to DMs	Disclosed in creator resources
TikTok	Watch time + replays	Disclosed in creator portal
LinkedIn	Comments + dwell time	Partially disclosed

X's decision to open-source its algorithm is unique among major platforms. Instagram and TikTok publish high-level explanations, but no other platform has released actual weighting multipliers.

This transparency benefits creators who want to reverse-engineer what works, but it also means that engagement-bait tactics (reply-begging, rage-bait) get identified and countered faster. X has already updated its code multiple times to down-weight low-quality replies and repetitive engagement patterns.

How to Apply These Insights

Based on the public algorithm weights, here's a prioritized checklist for X creators:

1. Write posts that invite replies. Ask specific questions, share contrarian takes that invite respectful debate, or post incomplete information that prompts corrections.
2. Optimize for dwell time. Use line breaks, bullet points, and clear structure to make long posts readable. Threads work well if each tweet adds value.
3. Avoid negative feedback. Test controversial topics with smaller audiences first. If a post type consistently gets "show less" clicks, stop posting it.
4. Use video strategically. Only post video if you're confident viewers will watch at least 50%. A skipped video hurts more than no video at all.
5. Include credibility signals. Give readers a reason to click your profile by demonstrating expertise or unique access.

For creators focused on building reach on X, understanding these concrete ranking weights—especially the 27× multiplier for replies and the severe penalties for negative feedback—offers a measurable advantage over those still guessing at what the platform prioritizes. If you're looking to accelerate visibility while applying these algorithmic insights, Fanovera's X campaigns can help amplify posts that already demonstrate strong engagement signals.

The Limits of Algorithmic Transparency

X's open-source code reveals how the algorithm weighs signals, but not why certain content resonates with human audiences. A post can check every algorithmic box—high dwell time, lots of replies, zero negative feedback—and still fail to grow your audience if it doesn't deliver genuine value.

The algorithm is a distribution mechanism, not a creativity engine. It amplifies content that people already want to engage with. Your job as a creator is to produce that content; the algorithm's job is to show it to more people.

Treat these weights as guardrails, not a formula. The 27× reply multiplier tells you that conversation matters, but it doesn't tell you what topics your audience wants to discuss. Use the data to avoid algorithmic penalties (like negative feedback) and to double down on formats that work (like reply-generating questions), but let audience feedback—not code—guide your creative direction.