How to rank on the X algorithm: what the For You feed actually pays for
The X ranking pipeline explained from the code side: retrieval channels, hard gates, engagement weights, the mutual-follow boost, cold start for small accounts, and repost mechanics — applied to marketing listings and local expertise.
Key takeaways
- Originals reach strangers; replies and reposts do not
- Mutual follows quadruple your reply value
- Reply, quote and share predictions outweigh likes 10 to 40x
- One report erases dozens of likes
- Reposts by others put you back in-network at full strength
How a post reaches the For You feed
Every request gathers candidates three ways: in-network retrieval from people who follow you (up to 1,200 candidates), embedding retrieval that matches your post to strangers’ engagement history (1,000), and cluster retrieval for posts similar to what a viewer just engaged with, under 48 hours old (800). Everything is scored together with no per-source quotas, only the top 50 survive, and roughly 35 organic posts get served. Being similar to what people are engaging with right now is the ticket into stranger feeds.
Hard gates: fail one and you get zero
Posts older than 48 hours are dropped from every channel. Already-seen posts are filtered twice over, so repeating content buys nothing. Mutes, blocks and muted keywords drop you before scoring. Reposts and replies never reach non-followers at all. Threads collapse to their single best-scoring post. And if an original trips a visibility rule, every repost and quote of it drops too.
What the scorer pays for
The ranker predicts, per viewer, the probability of each action and takes a weighted sum — predicted probabilities, not raw counts, so a like from a picky lurker outweighs a hundred reflexive likes. Approximate weights: copy-link share 20, reply to a mutual’s original 20, quote 5, DM share 5, reply 5, new follow 4, favorite 0.5, click 0.4, external-link click 0.2, dwell time 0.004 per second. Profile clicks score zero, and bookmarks have no serving weight at all.
Negative feedback is a different scale
A report weighs about minus 234, a mute minus 59, “not interested” minus 43, a block minus 31. One “show fewer posts like this” erases what dozens of likes earn. Engagement-bait, misleading “wait for it” framing, and spammy prompts are net-negative even when they harvest likes.
Mutual follows: the biggest coded advantage
If a viewer follows you back, the reply weight on your original posts jumps from 5 to 20 in their feed — a boost that was A/B tested before shipping. Converting followers into mutuals multiplies your reply value fourfold. Following back engaged clients, neighbors and local businesses is directly monetizable reach, not etiquette.
Multipliers working against you
Out-of-network candidates carry a 0.75 handicap, and even in-network replies and reposts eat it — only true originals from followed authors run at full strength. A second post by the same author in one feed batch runs at roughly 0.63, a third at 0.44, so posting five times an hour makes your posts cannibalize each other. A diversity reranker also demotes near-duplicate variants. One strong post per window beats three mediocre ones.
Cold start for small accounts
New accounts are not doomed. An original post under 24 hours old with fewer than 1,000 impressions, from an author with at most 1,000 followers, gets lifted to a guaranteed mid-feed slot if it is already ranking near the boundary. Under exploration, your early favorite-to-view ratio decides whether the system keeps trying you — front-load the hook, because slow-burn intros die before the reaction happens.
Playbook for listing and market content
Lead with originals that engineer replies, quotes and shares: hyper-local market data people can argue with, single-sentence claims worth quoting, and reference material worth sending to a friend. Use video at least 10 seconds when it earns its place. Post when your local audience is freshly active — the 48-hour clock starts immediately and the first hours decide retrieval. Niche consistency compounds: retrieval routes your future posts to people who engage like your past engagers, so drifting topics reset the association.
Reposts: yours versus getting reposted
Your reposts reach only your followers, score at 0.75, and get deduplicated — never a growth strategy. Getting reposted is how originals scale: the reposter’s followers see your post as in-network, at full strength, carrying your original’s predicted engagement. Make self-contained, screenshot-worthy artifacts that make the reposter look smart. Quoting your own earlier post is the only self-propagation that keeps original status.
Account hygiene
Non-follower delivery adds roughly two dozen extra drop rules on top of the base set. An NSFW avatar or banner alone removes all your media posts from recommendations. Spam labels kill a post everywhere it appears, and abuse ratios are computed against out-of-network favorites — reply-spamming strangers generates exactly the pattern those classifiers punish. Bot detectors read posting cadence, so automation-shaped activity risks suspension.
Myths the code debunks
One report does not cancel hundreds of likes — weights multiply predicted probabilities and reports are thousands of times rarer. Coordinated engagement produces nothing unless the post was served in the app; link-dropping into group chats scores zero. More posts means sublinear reach. Bookmarks have no serving weight. Profile clicks score zero. Hashtags appear nowhere in the pipeline.
More guides | Tools | Pricing