How does reply quality affect ranking?

Two ways, both in the code. First, "reply" is one of the positive predicted actions in the main scorer — a post the system predicts will earn replies ranks higher, so writing things worth replying to helps your post. Second, replies themselves are ranked: Grok includes a reply-ranking classifier that orders the replies under a post, so a low-effort or spammy reply can be buried while a substantive one rises. Replying well is a growth channel; replying badly wastes the slot.

Replies do double duty in the algorithm: they're a signal that lifts the original post, and they're independently ranked as content in their own right. Both are visible in the released code.

Replies lift the post they're on

"Reply" is one of the weighted positive actions in the main scorer — the system predicts how likely a viewer is to reply, and a higher predicted reply score raises the post's rank. This is why "ask a question" and "write something worth responding to" is real advice: replies are a first-class reward signal, not a vanity metric.

CODE-CURRENT0bfc279verified 2026-06-12
The current WeightedScorer combines predicted probabilities for: favorite, reply, retweet, photo expand, click, profile click, video quality view, share, share via DM, share via copied link, dwell, quote, quoted click, continuous dwell time, follow author, not interested, block author, mute author, and report.
xai-org/x-algorithm — home-mixer/scorers/weighted_scorer.rs, lines 44–68 (compute_weighted_score)as of the May 15, 2026 release

Your replies are themselves ranked

Replying to big accounts as a growth tactic works only if your reply surfaces. The Grox layer includes a reply-ranking classifier that orders the replies under a post — so a generic "great post!" or an obvious engagement-bait reply can be ranked low and buried, while a substantive reply rises to where people see it.

CODE-CURRENT0bfc279verified 2026-06-12
Grox's base ContentClassifier takes a list of content categories and an LLM, and exposes classify(post) returning per-category results — every Grox content classifier is an LLM judging a post against named categories.
xai-org/x-algorithm — grox/classifiers/content/classifier.py, ContentClassifier base (L19-L28)as of the May 15, 2026 release

Why low-effort replies fail as a tactic

The reply guy strategy — flooding popular posts with quick replies — runs straight into the reply-ranking model. Volume doesn't help if the replies rank at the bottom. A smaller number of genuinely good replies, which the classifier surfaces, reaches far more people than a flood of filler that gets sorted below the fold.

What the code doesn't say

▲ What the code doesn't say

Exactly how reply-ranking scores a reply, and how heavily reply prediction weighs in the main scorer. Both depend on withheld prompts and parameters. We can show replies matter twice — as a lift signal and as ranked content — but not the precise magnitudes.

UNKNOWN0bfc279verified 2026-06-12
The numeric values of the current weights are not included in the open-source release: weighted_scorer.rs references a params module (e.g. p::FAVORITE_WEIGHT, p::REPLY_WEIGHT) whose values are not present anywhere in the published repository.
xai-org/x-algorithm (verified by direct inspection of the full repository tree at the pinned SHA) — home-mixer/scorers/weighted_scorer.rs references crate::params; no params definitions with weight values exist in the releaseabsence verified at the pinned SHA; values may be published in a future release

What to do with this

Make posts that invite replies, and when you reply to others, make it good enough to rank. Both are leverage the code rewards. The reply-guy shortcut is a trap precisely because the reply-ranking model exists — quality is the only version of the tactic that survives it.

← Growth mechanics