According to the README of xai-org/x-algorithm, which we read on 10-11-2026, the For You feed gathers posts from accounts you follow and from outside your network, filters them, and ranks them with a model that predicts how likely you are to act on each one, for example reply, repost, share or mute the author.
Does X publish its algorithm?
Yes. X published the code of its For You feed on GitHub as xai-org/x-algorithm, announced on 20-01-2026 according to TechCrunch. It shows the structure of the system. The README says X runs experiments on part of the traffic and uses scheduled scripts to keep the defaults in the code in line with the main production values, so read it as the best official description, not a guarantee of every setting.
Do links hurt my reach on X?
The published code lists link clicks among the actions it predicts, and we found no rule in the README that penalises links. We cannot confirm any link penalty from official sources, so we do not claim one. What we can say is that a post people reply to and share is a better bet than one they scroll past.
How often should I post on X?
Pick a rhythm you can keep. The README describes a step that lowers the score of an author's repeated posts in the same feed, so flooding it is unlikely to help. Clautia allows up to 20 posts per X account every 24 hours, a limit of the publishing provider, not of X.
Is there a different algorithm for the Following feed?
The repository describes the For You feed. We did not find a statement about the Following tab in the README, so we do not describe it here.
Can Clautia tell me how my X posts perform?
No. Clautia does not measure results today, so read impressions and engagement in the statistics X shows you. Clautia helps you write, adapt and schedule posts for X.
The X algorithm, according to the code X published, builds your For You feed in a few steps: it gathers posts from accounts you follow and from outside your network, removes the ones that do not qualify, scores each remaining post by how likely you are to act on it, adjusts that score for repeated authors and out-of-network posts, and then removes what the visibility rules say to drop. This guide explains what the public repository says, what it does not say and what to do with it when you write posts. Clautia, the agentic social media manager, then schedules those posts for X. Everything below comes from the README of xai-org/x-algorithm, which we read on 10-11-2026, and from X's own announcement as reported by TechCrunch.
What X has published
X open-sourced the algorithm behind its For You feed on GitHub (xai-org/x-algorithm), announced on 20-01-2026 according to TechCrunch. The repository is licensed under Apache-2.0 and describes itself as the "Algorithm powering the For You feed on X". At launch, TechCrunch reported that the write-up said the system relies entirely on a "Grok-based transformer" and uses no "manual feature engineering for content relevance". X has kept updating the README since then; the version we read on 10-11-2026 says the feed "ranks posts using a transformer model" and lists notable updates up to 26-10-2026.
Two limits apply. First, it covers the For You feed only. Second, the README itself says that many tunable values live in a configuration system and that X runs experiments on a small share of traffic, so we treat the code as the best official source, not a rulebook.
The pipeline in plain words
Step
What the README says happens
What it means for you
Candidates
Recent posts from accounts you follow (in-network), plus out-of-network posts found through retrieval, SimClusters and a shared list of popular recent posts; both are ranked by the same model
Posts can reach people who do not follow you
Hydration
Post text and media, author details, quoted post, language, engagement counts and subscription status are attached
Clear text and good media travel with the post
Pre-scoring filters
Duplicates, posts older than 48 hours, the viewer's own posts, blocked or muted accounts, muted keywords and posts already seen or served are dropped
Fresh posts count, and a muted word can remove you
Scoring
A model predicts the probability of many actions; a weighted sum gives the score, then three adjustments apply
The model scores likely reactions, not raw counts
Post-selection filters
After the order is fixed, posts that visibility filtering says to drop are removed and branches of one conversation are collapsed
Spam or abuse labels can keep a post out of the feed
Blending
Ads, Who to Follow and prompts are added around the ranked posts
Not everything in the feed is ranked on the same terms
The actions the model predicts
The README lists the actions the scorer estimates. Grouped:
Engagement: favorite, reply, repost, quote, share, share by direct message, share by copied link.
Clicks: on the post, the profile, a link, a photo to expand, a video to open, a quoted post.
Attention: quality video views, time spent on the post, time spent after a click, seconds of video watched, seconds of profile visit.
Follow: following the author.
Negative: not interested, mute author, block author, report, and not dwelling on the post.
The score is a weighted sum of these probabilities: positive actions carry positive weights and negative ones negative weights. The weights live in the code, but the README warns against a common misreading: they multiply your own predicted probability of each action, not raw counts, so "one report cancels out a set number of likes" is wrong. That is why we do not turn the weights into a price per like or per reply.
Three adjustments after the score
Author diversity. Each post after an author's first in the same feed gets a decaying multiplier, down to a floor.
Out-of-network discount. Posts from accounts the viewer does not follow, and replies and reposts from accounts the viewer does follow, are multiplied by a factor below one.
New-author boost. Posts from authors whose impressions are below a threshold are lifted toward a target position.
What to do when you post on X
None of this is a trick. It points to habits:
Write posts people answer. A question, a clear opinion or a specific tip invites replies, a predicted action.
Make the post worth sharing. A shareable post is one a reader would send to a colleague.
Do not post the same thing twice. Duplicates are filtered, and a run of posts from one author is discounted.
Post when your followers are around. Old posts drop out after 48 hours, so timing matters. Best time to post on X helps you test slots.
Stay out of the negative actions. People mute, block and report accounts that feel like spam or bait.
Reply to those who reply. Conversation is the signal you can create yourself.
North Ridge Running, an online running shop, posts on X on Tuesday morning. Version A: "New trail shoes online." Version B: "Which would you pick for a Sunday long run: the cushioned road shoe or the grippy trail one? Orders before 3 pm ship today." Version B is a question that is easy to answer, so it invites the reply the model is trying to predict. This is an illustration of the logic, not a measured result.
How the other algorithms compare
Each network publishes different material. X stands out because it publishes code; LinkedIn, for instance, explains its feed in engineering posts, as we cover in how the LinkedIn algorithm works. If you are weighing X against another text network, read Threads vs X.
What Clautia does today with X
Clautia connects X next to nine other networks, including Google Business Profile. You write the post for X with its own preview, AI writing covers X with monthly credits, and you publish it or give it a date in the shared calendar. Posts stay as drafts until you or a teammate schedules or publishes them, and an AI agent such as Claude Code, Codex or Cursor can do it through MCP with your API key. Clautia allows up to 20 posts per X account every 24 hours, a limit of the publishing provider rather than of X. Clautia does not measure results today, so read them in X itself. See how to schedule tweets for the routine and the X scheduler page for the workflow.
What to remember
The official source is the xai-org/x-algorithm repository, read on 10-11-2026.
The feed ranks the actions people are likely to take: replies, shares, follows and also mutes.
Fresh, varied, answerable posts fit the logic of the code; tricks do not.