How the X Algorithm Works in 2026: Ranking Stages, Signals and What Changed

What is the X algorithm?

Metricool describes it as a series of rules and factors built on location, virality, personal interests, recency and whether a post contains media. Sprout Social goes further and says the system now uses artificial intelligence, specifically Grok, to choose what people see.

Three published guides supply the evidence for this page: Teract AI, Metricool and Sprout Social. None is X's own documentation, and the research behind this article found no primary source from X. Each figure below is therefore a named source's claim, and the weakest ones are flagged.

Some headlines in this space promise lessons from 100 million views. This page does not: it reports no original view or engagement data, only what the three guides say and where they disagree.

What are the key claims about the X algorithm in 2026?

The core claim is that X ranks posts by predicted engagement, favors conversation and early reaction, and is said to hold back posts that link elsewhere. The table lists the ten claims the rest of the page unpacks, with who made each and how it is supported.

Only two of the ten claims point to a named person at X. One is the mutuals tweak, quoted from X's head of product, and the other is the link penalty that Sprout Social attributes to Elon Musk without a link. Treat the rest as informed secondhand reports.

How does the X algorithm work, stage by stage?

It works in three stages: it gathers candidate posts, ranks them with a machine-learning model, then filters the ranked list. Teract AI and Metricool describe the same structure, with small differences in detail.

Stage 1: candidate sourcing

In the first stage, X pulls recent and relevant posts from two kinds of source. Metricool names them as in-network sources, meaning accounts you follow, and out-of-network sources, meaning accounts you do not.

The aim is a For You page that mixes both. Metricool notes that the out-of-network side is the harder job, since it means picking from millions of posts by accounts you have never followed.

Teract AI counts three sources instead of two: followed accounts, accounts that similar users engage with, and accounts in your interest graph. It puts the resulting pool at roughly 1,500 candidates. Its advice for reaching out-of-network feeds is to earn engagement from accounts with varied follower bases, for example by replying usefully to larger accounts.

Stage 2: ranking

In the second stage, a machine-learning model gives every candidate a score for how likely you are to engage with it. Metricool says the model scores a post the same way whether or not you follow its author, and a higher score means a higher position in For You.

Teract AI says the score draws on more than 1,000 signals and predicts whether you will like, reply, repost or click. Sprout Social calls the scorer the heavy ranker and says it looks at your past interactions, what similar users engage with and what is trending.

Stage 3: filtering

In the final stage, rules and product features remove or reorder posts after the model has scored them. Metricool says X filters out blocked accounts, adult content and posts you have already seen.

Teract AI adds spam, duplicates and posts with negative signals such as high block rates, spam reports and misleading-information flags. Metricool lists five of the rules in more detail.

Two of these rules matter for creators. Author diversity limits how often one account can appear in a row, and feedback-based fatigue means a post that readers dislike loses ground even if it started well.

How do For You, Following and Explore differ?

They differ in who picks the order: a ranking model runs For You, time runs Following, and Grok-assisted curation shapes Explore. Sprout Social describes all three surfaces and says each ranks content in its own way.

The practical consequence is in the Following row. Sprout Social says X opens on For You by default, so even your own followers are more likely to meet you through a ranked feed than a chronological one.

Ranking by relevance is not new. Sprout Social dates it to 2017, when X replaced While You Were Away with In Case You Missed It and began presenting posts by relevance instead of in time order.

What changed in the X algorithm in 2025 and 2026?

Three changes stand out: a late-2025 shift toward conversation quality, a stronger link between reach and X Premium, and a July 2026 boost for mutuals. The sources date only the last of these precisely.

Teract AI says the algorithm changed significantly in late 2025 and keeps evolving. Its list of updates covers replies weighted well above likes, a decisive first half hour, hashtags that no longer help, harsher treatment of external links and faster personalization that leans on your last seven days of activity.

Sprout Social frames the change differently. It says X has moved toward pay-for-reach and put Grok into its core ranking logic, and that Explore became heavily integrated with Grok in 2025.

The July 2026 mutuals tweak

In July 2026, X boosted the visibility of posts to mutuals, the people you follow who follow you back. Metricool reports that head of product Nikita Bier announced it on July 13, 2026.

According to Metricool's account of his post, the team noticed this data was missing from the algorithm, so friends appeared less often in replies. The result was a reply section that felt like a battleground with people you do not recognize.

Metricool reads the tweak as part of a broader pattern: X has spent 2026 pushing the platform toward a creator hub and away from a free-for-all comment section. That is the guide's interpretation, not X's statement, but it fits the direction of the other changes.

Which signals does the X algorithm weigh most?

Teract AI says engagement velocity weighs most by a wide margin, but its multipliers are one site's reading of the system and the excerpt does not show how they were derived. The table lists the nine weights visible in Teract's outline.

Read the numbers with care. They are not on an obvious common scale, since replies are said to count 27 times as much as likes while conversation depth carries only 5x, and the guide does not reconcile the two.

The three guides agree more on direction than on size. The next table sets their signal lists side by side.

Where all three overlap, in recency, engagement and relevance, you can be reasonably confident. Where only one speaks, as with account standing, you are relying on that one source.

How much do the first 30 minutes matter?

They matter a great deal if Teract AI is right, because it calls early engagement velocity the single most important signal. It says 10 or more engagements within the first 15 minutes sends a post to far more people, while fewer than 3 lets it stall.

Teract AI gives the window as 15 to 30 minutes in one place and the first 30 minutes in another, so treat the exact threshold as approximate. Its advice is to post when your audience is active and to seed early replies, and it names replying to others as one way to do that.

Sprout Social points the same way from a different angle. It says a post loses half its potential visibility score every six hours and tells readers to generate engagement immediately after posting.

One gap matters in practice. Neither source says whether the thresholds scale with account size, and 10 engagements means something very different for an account with 500 followers than for one with 500,000.

Do links, hashtags and media change reach on X?

Outbound links are the penalty the sources mention most, hashtags are described as nearly irrelevant by one guide, and rich media is described as an advantage. The evidence is thinner for each than the confident tone of the guides suggests.

Links

Teract AI says a link in the first post now cuts reach by about 50% and recommends putting the link in a reply. Sprout Social says Elon Musk confirmed that posts sending people off the platform often lose 50% to 90% of their reach.

The two figures overlap at 50% but differ at the top end, and neither guide links to the statement or test behind its number. The consistent message is simple: if a post exists to earn reach, keep the first post free of outbound links.

Hashtags

Teract AI says hashtags have minimal impact because X now categorizes content with semantic language processing, and that using three or more can trigger spam filters. Metricool still lists topics, hashtags and keywords among the cues used to match posts to your past activity.

These two views are compatible if hashtags are one weak cue among many. The safe reading is that one or none costs you nothing and a pile of them risks a spam flag.

Media and length

Metricool says X will reward posts that drive meaningful engagement and use rich media, and Teract AI weights visual content at 10x. Sprout Social adds that video watch time and bookmarks help train the model to place you in interest clusters it calls SimClusters.

Teract AI also lists tweet length at 6x without saying whether longer or shorter is favored in the excerpt reviewed. Do not guess at that one.

Does X Premium affect reach?

According to Sprout Social, yes: the algorithm favors accounts with X Premium, and organic reach for non-subscribers is significantly lower. It describes the model as meritocracy via subscription and the platform as pay-centric.

Sprout Social does not say small accounts are shut out. It says the algorithm does surface content from smaller accounts, but generally needs them to be verified subscribers to signal credibility to the ranking model.

This is the claim to hold most loosely. Sprout Social offers no figures for the gap, and neither Teract AI nor Metricool makes the claim in the excerpts reviewed, although Teract's author authority signal at 50x could be related.

Where do the sources disagree about the X algorithm?

They disagree on how many candidate sources exist, how big the link penalty is, what actually does the ranking, and whether hashtags or X Premium matter. The table sets the differences side by side.

Some of these gaps are differences of emphasis. A guide that does not mention X Premium has not denied it, and a guide that names two candidate sources rather than three may simply be summarizing.

Others are real conflicts. The link penalty range and the claim that Grok does the ranking cannot both be settled from these sources, and settling them would take X's own documentation or code.

What should you do with this information?

Act on the points at least two sources support, and test the rest on your own account. Replies and conversation, early engagement, rich media and avoiding negative feedback each appear in more than one guide.

1. Write for replies. Teract AI says conversation quality now outweighs raw likes, and Metricool says meaningful engagement is rewarded. End posts with a question or a position people can answer.
2. Be present in the first half hour. Post when your audience is online and reply to early comments, since Teract AI and Sprout Social both stress early engagement.
3. Keep the first post link-free. Put the link in a reply, as Teract AI advises, and compare reach against your link-in-post history.
4. Use one or no hashtags. The sources disagree on their value but agree that a pile of them is a risk.
5. Add media when it helps the point. Metricool and Teract AI both credit visual content, so test an image or short video against plain text.
6. Protect against negative feedback. Metricool says posts that draw negative feedback rank lower, and Teract AI lists block rates and spam reports among its filters.
7. Engage with your mutuals. After the July 2026 change, Metricool says your posts are boosted to people you follow back, so replying to the people who follow you back is worth the time.

Then test one variable at a time. Change only the link placement, the media or the posting time for a few weeks, record impressions for each post, and compare the averages.

A handful of posts proves nothing, and the guides give few sample sizes of their own beyond Teract's stated tests of 2,000 and 10,000 tweets. Your own account is the only data set you can fully trust.

Where does Kanso fit into posting on X?

Kanso's X Agent drafts posts for X from your company context and leaves the decision to publish with you. It does not change how X ranks anything.

• The X Agent drafts posts from company context, lets you edit, regenerate or dismiss each one, and publishes the approved ones through your connected X account while tracking each publish. It does not schedule or post on its own.
• The Agent Hub lists every live agent, and the pricing page compares the plans.

Use the checklist above when you review each draft: does it invite replies, is the first post free of links, and is it ready for the first 30 minutes?

How the X algorithm works: frequently asked questions

Sources

• X (Twitter) Algorithm 2026: How It Works, Latest Changes & Viral Secrets, Teract AI
• How the Twitter Algorithm Works, Metricool
• How the Twitter algorithm works in 2026 [6 strategies], Sprout Social