X has published new code and documentation for the For You feed algorithm through the xAI GitHub repository. The release gives creators, brands, publishers, and marketers a clearer view of how X now appears to retrieve, score, filter, and rank posts.

The short version: X is no longer a simple “post often and farm likes” platform. The current For You system is built around candidate retrieval, content understanding, predicted engagement, author diversity, negative feedback suppression, duplicate filtering, and out-of-network discovery. That means creators should optimize for strong individual posts, clear topical identity, high-value engagement, and fewer low-quality repetitions.

This article reviews the official xAI GitHub documentation, compares it against popular creator claims about the new algorithm, and turns the most defensible takeaways into a practical X posting strategy.

Influencer’s / Creator’s Claims about the New X Algorithm

We will be examining specifically the claims by 2 different creators on X that were getting attention and RT’s.

This post by @DeRonin_

BREAKING: X dropped the new ranking algo today. What's dead and what wins for bangers:

[ what just died ]:

– spam posting (4+ posts/day = author dilution penalty)
– reply farming (replies weighted by WHO replies, not how many)
– text-only posts (media gets 2x signal weight now)… https://t.co/lgdvRea4Bn pic.twitter.com/feW0by9lPV

— Ronin (@DeRonin_) May 15, 2026

And this post by @RyanMalin_

top10 tips for growing an account on X with the latest algorithm pic.twitter.com/7kv1EBVsT1

— Ryan (@ryanmalin_) May 15, 2026

What xAI Actually Says About the X For You Algorithm

According to the xAI GitHub repository for the X For You Feed Algorithm, the For You feed combines two major sources of posts: in-network content from accounts a user follows and out-of-network content discovered through machine-learning retrieval.

The repository says those candidates are ranked with Phoenix, a Grok-based transformer model that predicts the probability that a user will take several different actions on a post. These predicted actions include favorites, replies, reposts, quotes, clicks, profile clicks, video views, photo expands, shares, dwell time, follows, not-interested feedback, blocks, mutes, and reports.

That matters because the algorithm is not simply asking, “Will this get likes?” It is asking, “Given this user’s history, how likely is this post to produce several desirable or undesirable actions?”

The same xAI documentation says the weighted scorer combines those predicted actions into a final score. Positive actions such as likes, reposts, and shares have positive weights, while negative actions such as blocks, mutes, and reports have negative weights. In plain English: the best post is not the one that gets the most raw engagement. It is the one that is predicted to create the right mix of attention, interaction, clicks, shares, dwell, and low negative feedback for a specific user.

Viral Myths vs. Reality: What DeRonin_ Got Wrong About the X Algorithm

After xAI published the X For You algorithm repository, several viral posts claimed to explain exactly what changed. One of the most widely shared was from DeRonin_, who claimed that specific tactics were now “dead” while others had suddenly become ranking winners.

Some of those claims are directionally useful. Others go far beyond what the public xAI documentation actually proves. The safest way to optimize for X is to separate confirmed algorithmic mechanics from speculative creator-growth advice.

The More Accurate Takeaway

The new X algorithm is not a simple checklist where “media equals 2x,” “four posts equals penalty,” or “30-minute replies equal ranking gold.” According to the xAI repository, the For You feed is a prediction and filtering system. It retrieves candidate posts, predicts many possible user actions, applies positive and negative weights, filters low-quality or ineligible content, reduces duplicates, and tries to diversify what each user sees.

That means the real optimization strategy is more nuanced:

• Make each post strong enough to win on its own. The system scores individual candidates, not just accounts.
• Optimize for multiple valuable actions. Likes matter, but so do replies, reposts, shares, clicks, profile clicks, media expansions, video views, dwell time, and follows.
• Avoid negative feedback. Blocks, mutes, reports, and not-interested actions can hurt distribution.
• Do not duplicate yourself. Deduplication and previously-seen filters make repeated posts less useful.
• Build a clear topical identity. Retrieval systems need to understand who your content is for.
• Use proof and specificity. Specific claims, screenshots, data, names, and examples are more likely to earn meaningful engagement than generic takes.
• Use media strategically. Media can help when it creates more attention, understanding, proof, or interaction. It is not a magic ranking multiplier.

So DeRonin_ was directionally right that spam, repetition, shallow engagement, and generic content are bad bets. But many of the most viral claims in that post are presented with more certainty than the public xAI documentation supports. Creators and brands should treat those claims as hypotheses, not confirmed ranking rules.

The Big Strategic Shift: Optimize for Predicted Actions, Not Vanity Metrics

The most important takeaway from the xAI release is that creators and brands should stop treating X as a one-dimensional likes game.

The official repository says the model predicts many engagement probabilities, including replies, reposts, clicks, profile clicks, video views, photo expands, shares, dwell, and follow-author events. That supports the general creator advice in DeRonin_ post’s on X that creators should optimize for more than likes. However, some specific claims circulating on X go beyond what the public xAI code directly proves.

For example, it is reasonable to say that profile clicks, media expansions, video views, dwell time, replies, reposts, and shares matter because xAI lists them as predicted actions. It is not currently safe to say from the public repository alone that “media gets exactly 2x signal weight,” that “4+ posts per day triggers a hard author dilution penalty,” or that “replying to every comment in the first 30 minutes is ranking gold.” Those may be observations, tests, or creator heuristics, but they are not directly confirmed by the official GitHub documentation.

What the Viral X Algorithm Claims Get Right and Wrong

1. Post Less Often, But Make Each Post More Complete

The post by @RyanMalin_ recommends spacing out posts because of a diversity penalty. That is a reasonable interpretation of the official xAI documentation. The repository specifically names an Author Diversity Scorer that attenuates repeated author scores to ensure feed diversity.

For brands, this means volume alone is less attractive than it used to be. If a company posts eight generic updates per day, X may not need to show all eight to the same audience. Worse, weaker posts may generate low engagement, low dwell, muted keywords, not-interested clicks, or other negative patterns.

A practical cadence for many brands would be one to three strong posts per day, not a constant drip of low-value posts. The goal is not to disappear. The goal is to avoid training the system that your account produces skippable, repetitive, or low-signal posts.

Recommended action

• Publish fewer, stronger posts instead of many thin updates.
• Give each post one clear idea, one clear audience, and one reason to engage.
• Separate major posts by several hours instead of posting a rapid-fire batch.
• Watch account-level performance after high-volume days versus low-volume days.

2. Optimize for the Full Action Set

The public xAI documentation lists a broad set of predicted actions. This should change how creators write.

A post that only gets a few passive likes may not be as valuable as a post that earns replies, shares, profile clicks, dwell time, media expansion, and follows. A post that earns high engagement but also causes mutes, blocks, reports, or not-interested clicks may be riskier than it looks from public metrics.

That means creators should design posts around the actions they want to trigger.

3. Use Media, But Do Not Treat Media as a Magic Ranking Hack

The DeRonin_ post claims text-only posts are dead and that media now receives 2x signal weight. The official xAI documentation does not confirm that exact multiplier. However, it does confirm that the system uses media-related signals, including media detection, video duration, video views, and photo expands.

So the practical recommendation is simple: pair important posts with useful media when the media adds value. Do not add generic stock images just to satisfy a perceived algorithm requirement.

For creators, useful media can include screenshots, short videos, charts, diagrams, carousels, product clips, annotated examples, and proof images. For brands, useful media can include customer examples, product walkthroughs, data visualizations, founder videos, behind-the-scenes proof, and industry-specific charts.

Best media formats for the new X environment

• Screenshot proof: data, results, search results, dashboards, emails, product screens, or receipts.
• Short videos: under 90 seconds, focused on a specific action, explanation, or result.
• Image carousels: 3 to 7 slides, one clear claim or takeaway per slide.
• Charts and diagrams: simple visuals that make a complex point easier to understand.
• Annotated examples: show the reader exactly what to look at.

4. Build a Clear Topical Identity

The xAI documentation says Phoenix Retrieval uses a two-tower model. The user tower encodes user features and engagement history, while the candidate tower encodes posts into embeddings. The system then retrieves top candidates using dot product similarity.

In practical terms, this means X needs to understand which users are likely to care about your posts. A scattered account that posts about SEO, parenting, crypto, politics, fitness, cooking, and memes may make that matching harder. A focused account that repeatedly publishes strong content around a few related topics gives the system a clearer pattern.

This supports @RyanMalin_’s recommendation to build consistent niche embeddings. Creators should not interpret that as “never evolve.” But abrupt topic pivots, mixed-audience posting, and vague personal-brand content can make the account harder to match to a specific audience.

Recommended action

• Pick two to four recurring content pillars.
• Use consistent language around your niche, industry, audience, and use cases.
• Repeat themes, not duplicate posts.
• Make the first line of each post clearly signal the topic.

5. Do Not Duplicate Content

This is one of the most defensible optimization recommendations. The xAI documentation lists several filtering and deduplication systems: duplicate post filters, repost deduplication, previously seen post filters, previously served post filters, and conversation deduplication.

That means creators should be careful with reposting the same promotional text, using identical hooks, recycling viral templates, repeating the same image, or posting the same link with minor wording changes.

Reusing a successful idea is fine. Reposting near-identical content is risky.

Better approach

• Turn one idea into several distinct formats: a data post, a short video, a thread, a chart, and a reply.
• Change the angle, audience, proof, and format instead of lightly rewriting the same post.
• Avoid posting identical promotional CTAs multiple times in a short period.
• Do not use the same image or link preview repeatedly with nearly identical copy.

6. Avoid Negative Feedback Traps

The official xAI documentation is clear that negative actions matter. The model predicts not-interested, block-author, mute-author, and report actions. The weighted scorer can use negative actions to push down content users are likely to dislike.

This is important because many “growth hacks” produce visible engagement while also creating hidden negative feedback. Rage bait, engagement bait, vague dunk posts, repetitive promos, polarizing claims without proof, and misleading hooks may get comments while also causing users to mute, block, report, or tap not interested.

For brands, this is especially important. A creator may survive being provocative. A brand account usually has less room to generate negative feedback without damaging its distribution and reputation.

Reduce negative feedback by avoiding:

• Misleading hooks that do not pay off.
• Generic engagement bait such as “Agree?” or “Thoughts?” with no real prompt.
• Repeated promotional posts.
• Low-effort AI-generated threads with no original point of view.
• Polarizing statements without evidence, examples, or nuance.
• Posts that attract the wrong audience for your account.

7. Write for Out-of-Network Discovery

The xAI repository says For You includes out-of-network content discovered through Phoenix Retrieval from a global corpus. This is a major opportunity for small and mid-sized accounts because distribution does not depend only on follower count.

However, out-of-network discovery depends on relevance. If your post is going to be shown to people who do not follow you, the first line has to make the topic obvious and the post has to create enough value without relying on prior familiarity with your account.

The DeRonin_ post claims original takes from small accounts won because out-of-network discovery increased. The exact “3x” number is not confirmed by the public xAI documentation, but the general direction is supported: out-of-network candidate retrieval is a core part of the system.

Recommended action

• Write posts that make sense to people who do not know you yet.
• Open with a clear topic, claim, result, or tension.
• Use proof early: numbers, screenshots, names, examples, or specific experiences.
• Make your expertise visible inside the post, not only in your bio.

8. Treat Replies as Quality Signals, Not Reply Farming

The viral tweet claims reply farming is dead because replies are weighted by who replies, not just how many replies a post gets. The public xAI documentation confirms that replies are one of the predicted actions, but it does not fully confirm the exact weighting of “who replies.”

Still, the broader strategy is sound. Meaningful replies are more valuable than low-quality engagement bait. A post that attracts thoughtful discussion from relevant people is more likely to be a healthy signal than a post that attracts one-word replies from engagement pods.

Creators should also reply to comments because replies extend the conversation, clarify the original idea, and may create additional useful content. But do not interpret “reply in the first 30 minutes” as an officially proven ranking rule. Treat it as a practical community management habit, not a confirmed algorithmic law.

Better reply strategy

• Reply quickly when the discussion is active.
• Add new context instead of saying “thanks” to every comment.
• Turn good objections into follow-up replies.
• Pin or highlight replies that deepen the conversation.
• Avoid engagement-pod behavior that creates unnatural low-value comments.

9. Use Long-Form Posts When the Topic Deserves It

The DeRonin_ post claims 4,000-character long-form posts get heavier signal weight. The public xAI documentation does not confirm a special boost for 4,000-character posts. However, the model does predict dwell time, clicks, replies, reposts, and other actions that long-form posts can generate when they are genuinely useful.

Long posts work when they create more value, not merely because they are long. A 2,500-character tactical breakdown with proof, examples, and a clear structure may outperform a short vague post. But a bloated long post with no payoff can still fail.

Good long-form formats

• Breakdown of a current industry change.
• Step-by-step tactical playbook.
• Contrarian argument backed by proof.
• Case study with numbers and lessons learned.
• “What we tested / what happened / what we changed” post.

10. Make Posts More Specific and Less Abstract

The tone-of-voice advice in the DeRonin_ post is mostly consistent with how a prediction-based recommendation system should behave. Specific posts give users and models more signals. Abstract posts are harder to match, easier to ignore, and more likely to look like generic content.

For example, “AI is changing marketing forever” is weak because it is generic. “We tested AI-generated service pages across 42 local SEO campaigns and found three patterns that hurt indexing” is stronger because it includes a specific subject, proof direction, and audience.

Use this pattern

• First person: “I tested,” “we shipped,” “we analyzed,” “I found.”
• Specific object: “42 local SEO campaigns,” “17 SaaS pricing pages,” “1,000 Reddit citations.”
• Concrete result: “rankings dropped,” “demo requests increased,” “profile clicks doubled.”
• Useful lesson: “here is what changed,” “here is what I would do differently.”

Optimization Techniques Not Mentioned Enough

Use Muted Keyword Awareness

xAI lists a MutedKeywordFilter as part of pre-scoring filters. That means some posts may be removed for specific users before scoring if they contain muted keywords. Creators cannot know every user’s muted keywords, but they can avoid unnecessary rage terms, spam language, and repetitive promotional phrasing.

Keep Content Fresh

xAI lists an AgeFilter that removes posts older than a threshold. This means X is still a freshness-sensitive environment. Evergreen content can work, but the packaging should feel timely, relevant, or newly useful.

Reduce “Previously Seen” Problems

xAI lists filters for previously seen and previously served posts. If a user has already seen a post or something too similar in the same session, the system has reasons not to show it again. This is another reason to vary your angles instead of reposting the same message repeatedly.

Think About Language and Audience Fit

The xAI update mentions candidate hydrators for language codes. Brands with multilingual audiences should avoid mixing languages randomly in the same account strategy unless that is intentional. Clear language targeting can help the system understand who the post is for.

Avoid Brand-Safety Problems

The xAI update mentions brand-safety signals and visibility filtering for spam, violence, gore, and similar issues. Even when a post is not technically policy-violating, brands should be careful with sensational imagery, extreme wording, and unsafe adjacency if they want consistent distribution.

Use Quote Posts Strategically

The xAI update mentions quote post expansion as a candidate hydrator. This suggests quote posts can provide context, but they should add real value. A lazy quote post that says “this” is weaker than a quote post that explains why the original post matters, what is missing, or how it applies to your audience.

A Practical Posting Framework for X in 2026

Based on the xAI repository, the post by @RyanMalin_, and the DeRonin_ post, the safest playbook is:

1. Post one to three strong posts per day. Respect author diversity and avoid flooding the feed.
2. Make every post action-oriented. Write for replies, shares, clicks, profile visits, follows, dwell time, and media expansion.
3. Use media when it improves the post. Add screenshots, charts, videos, carousels, or proof images.
4. Build topical consistency. Train the system and your audience to understand what you are about.
5. Avoid duplicates and recycled templates. Reuse ideas, not identical posts.
6. Reduce negative feedback. Avoid rage bait, low-value AI content, spammy CTAs, and misleading hooks.
7. Write for out-of-network readers. Assume many readers do not know who you are.
8. Use specific proof. Numbers, screenshots, names, timelines, and examples beat vague claims.
9. Reply with substance. Treat replies as an extension of the post, not a place to farm empty engagement.
10. Measure beyond likes. Track profile visits, follows, replies, reposts, media views, video views, and negative feedback patterns.

Example Post Templates That Fit the New Algorithm

1. Tactical Playbook Post

Hook: “We changed how we write X posts for B2B clients after reviewing the new xAI algorithm repo. Here is the 7-step framework we are testing.”

Structure: numbered steps, one action per step, clear examples, no fluff.

2. Proof-Driven Case Study

Hook: “We cut posting volume by 40% and saw better average engagement per post. Here is what changed.”

Structure: before/after numbers, screenshots, lessons, next test.

3. Contrarian Take With Evidence

Hook: “Posting more is probably hurting some brand accounts on X.”

Structure: explain author diversity, show examples, give practical cadence advice.

4. Visual Breakdown

Hook: “The new X algorithm is not a likes machine. It is a predicted-action machine.”

Structure: include a simple diagram showing retrieval, scoring, filtering, and selection.

5. Short Video Demonstration

Hook: “Here is how I would rewrite a generic brand post for the new X algorithm.”

Structure: show bad post, rewrite it live, explain why each change matters.

What Not to Do

• Do not spam four, six, or ten low-quality posts per day just to stay visible.
• Do not copy viral post templates without adding original proof or point of view.
• Do not rely on engagement pods to manufacture weak replies.
• Do not post generic AI roundups with no experience or original insight.
• Do not use misleading hooks that create negative feedback.
• Do not post the same promotion repeatedly with minor wording changes.
• Do not optimize only for likes while ignoring dwell, profile clicks, shares, follows, and negative signals.

Final Takeaway

The new X algorithm appears to reward posts that are individually strong, topically clear, likely to generate meaningful actions, and unlikely to create negative feedback. It also appears to punish or filter many of the behaviors that made old-school Twitter growth feel spammy: duplicate posts, engagement bait, rapid-fire posting, low-quality replies, recycled templates, and generic content.

For creators and brands, the best strategy is not to “hack” the algorithm. The best strategy is to publish fewer but better posts, use proof, build a recognizable niche, make content easy to match with the right audience, and optimize for the full range of valuable actions.

The accounts that win on X now are likely to be the accounts that give the ranking system a clear reason to predict: this post is relevant, this person will engage with it, and showing it will not make them mute, block, report, or leave.

Sources

• Open Sourced X Algorithm on GitHub by xAI: https://github.com/xai-org/x-algorithm
• Post detailing optimization concepts by @DeRonin_: https://x.com/DeRonin_/status/2055334050676814056
• Top 10 optimization tips by @@ryanmalin_: https://x.com/ryanmalin_/status/2055307561121776004

REQUEST A SOCIAL MEDIA MARKETING QUOTE