Replies Beat Likes by 13x: What X's Algorithm Actually Rewards

Most people on X are optimizing for the wrong metric. Likes feel good — they're the number everyone watches — but according to multiple independent breakdowns of X's open-sourced ranking code, a reply is worth somewhere around 13 times more than a like in the algorithm's actual scoring. A retweet is worth even more than that. And the single most valuable thing that can happen to your post isn't a viral share — it's a real back-and-forth conversation in the replies.

Where this number comes from

X has open-sourced parts of its recommendation algorithm twice now — first in 2023, then a much more complete release in 2026 built on a Grok-powered transformer model. People who've gone through the code have pulled out a simplified version of the scoring formula that keeps showing up across multiple independent write-ups: roughly, likes count for 1x, replies for around 13.5x, retweets for about 20x, and profile clicks and bookmarks somewhere in between.

It's worth flagging clearly: the exact production weights X runs live aren't published — only the architecture and an older reference formula are public. So treat "13x" as a widely-repeated, code-informed estimate rather than a number stamped with X's official seal. Still, the direction every analysis agrees on is hard to ignore: conversation outweighs passive approval, by a lot.

Why this actually makes sense

Likes are nearly frictionless — a single tap, no real signal of how much someone cared. A reply means someone stopped scrolling, formed a thought, and typed it out. From a platform's perspective trying to maximize how long people stay and how engaged they feel, that's a far stronger signal of quality than a like ever could be.

This also explains a pattern a lot of creators have noticed without being able to name it: posts that are technically "well written" but airtight — no question, no opening for disagreement, nothing left hanging — often underperform posts that are rougher but invite a response. The algorithm isn't rewarding polish. It's rewarding posts that make someone want to talk back.

What this looks like in practice

A few patterns that tend to produce replies, based on what performs well under this kind of system:

Ask something specific, not generic. "What's one tool you can't work without?" gets replies. "Thoughts?" mostly doesn't.

Take a position someone might disagree with, without being needlessly combative. Disagreement drives replies; outright hostility tends to get penalized separately for tone.

Reply to your own replies. Several breakdowns of the code suggest a two-way reply chain — where the original poster replies back — gets scored even higher than a one-way reply. If someone responds to you, responding back isn't just polite, it's apparently doing real algorithmic work.

The likes habit is hard to break

None of this means likes are worthless — they're still a real, if smaller, signal. But if you've been chasing likes and wondering why your reach has flattened, this is probably part of the answer: you might be winning a metric that barely matters anymore, while a smaller, more conversational post quietly outperforms you on the one that does.

---

*TweetGem's "Reply Bait" style is built specifically to write tweets designed to start exactly this kind of back-and-forth — give it a try on your next idea.*