X's Algorithm Weights Are Public — a Copied Link Share Is Worth 40 Likes

Last week X published new insights into how its feed algorithm works, aiming to add transparency to how the system decides what each user sees. The platform listed various positive and negative signals defining post reach, and Business Insider reporter Truman Dickerson dug into them to clarify what users should and shouldn't be posting.

The basic structure

X's Phoenix algorithm displays posts to each user based on predictions about what they'll engage with, drawn from past behavior.

On weighting, Phoenix is primarily focused on driving shares, with follows and likes now carrying less weight as indicators of perceived value.

Positive signal weights

According to X's source code:

• Copying the post URL to share: about 20 points — 40 times the value of a like
• Replies, quotes and shares via internal DMs: about 5 points each
• A follow: about 4 points — eight times a like
• A repost: about 1 point
• A like: 0.5 — one of the lower positive engagement indicators

Combined with X's other mathematical factors — who a user follows, recency and so on — the algorithm produces a ranking number for every candidate post each time a user opens the app. That number dictates the feed.

The direction is unambiguous. X wants to incentivize re-sharing of posts in other formats, via users copying and re-sharing links. The best way to maximize reach is to optimize for re-sharing.

Negative signal weights

Phoenix penalizes posts it predicts people will report, mute, or mark "not interested." Notably, the penalties for these are far larger than for a block.

• A report: -234
• A mute: -59
• Not interested: -43
• A block: -31

So posting divisive comments to spur response still works, but it's a risky bet — once users start reporting or blocking an account, significant reach penalties follow.

One important caveat: the algorithm is personalized to each user. These penalties apply only to that viewer and don't reflect broader reach and engagement. Reports and blocks do carry a negative signal, but X has seemingly lessened the impact on other users, making it more isolated per viewer.

What this means for marketers

The disclosure doesn't reveal a major change in approach. It does offer pointers for strategy.

First, demote likes as a performance metric. Treating a 0.5-point signal as a campaign KPI misaligns with what the algorithm actually rewards. Keep likes in the report if you must, but optimize for shares.

Second, define what makes a post worth copying. The 20-point action isn't hitting repost inside the app — it's copying the URL and carrying it somewhere else. Content worth pasting into a group chat or a Slack channel — specific numbers, a useful list, evidence for an argument someone is having — fits that condition. Content that only triggers an emotional reaction earns likes without producing that behavior.

Third, the math on provocation changed. A -234 report penalty is difficult to offset with any positive signal. Trading reports for reactions is especially dangerous on a brand account.

Ultimately, when building strategy from algorithm data, the main focus for brands should be audience research: understanding what target audiences are more likely to respond to and share with their connections, both on and off the platform.

This isn't X's first transparency move — as covered in X Open-Sources Its Reach Code, a tool for checking whether your own posts are restricted also shipped. But disclosure doesn't equal recovered platform value. As noted in X Reworks Its DM Composer, product updates keep arriving, yet resource allocation is safer when based on observed audience behavior than on announced features.