X algorithm May 2026.

What actually changed (and when)

The biggest structural shift happened in January 2026, not May. X replaced its legacy recommendation engine with a system built on the same architecture as Grok, their in-house AI. That change was real and significant. What has happened since, including in May, is the fine-tuning layer on top of that new base.

Think of it this way. The new Grok-based ranking engine is the operating system. Each monthly adjustment is a software update on top of it. May 2026 is running the same OS as January but the weights have been nudged in ways that are visible in the output.

The reply weighting shift

Here is the number that surprised us most. Third-party analysis of the ranking model, including regression work we have done on our own campaign data, is showing replies now carry roughly 27 times the scoring weight of a like. That is not a typo.

In the old model before 2024, retweets were the king signal and replies were maybe three to five times a like. The shift toward replies makes sense when you think about what X is trying to measure: genuine conversation and investment from a reader, not passive scrolling reactions. A like takes one tap. A reply takes actual thought, even if it is just "this." The platform has decided replies are the better quality signal, and the algorithm reflects that now.

Practically what this means for your content: any post that drives conversation in the first 30 minutes has a significantly higher shot at For You distribution than a post that gets a lot of likes but no discussion. If you are writing content and optimizing for likes, you are optimizing for the weaker signal. Write for replies.

External link suppression is still happening

This one has been documented since 2023 and it is still very much in effect as of May 2026. Posting an external link in your main tweet body reduces distribution by an estimated 40 to 50 percent compared to the equivalent post without a link.

The workaround everyone knows about: post the link in the first reply on your own thread. Does it work? Yes. But here is the thing people do not talk about: the engagement velocity you need in the first 30 minutes still applies to the main tweet, and a link-heavy main tweet gets that velocity from a smaller initial audience because the suppression hits before the engagement starts. So you are fighting an uphill battle from minute one.

Our recommendation: if the link is essential, put it in a reply. If the link is not essential for the main message to land, just remove it from the first post entirely. The metric difference is real and consistent across our test accounts in May.

Hashtag behavior in May 2026

Here is the thing about hashtags in 2026: X does not use them for ranking in the For You feed. They are indexed for search, meaning your post can appear when someone searches a hashtag. But they contribute nothing to For You distribution, and if you use three or more in a single post, there is now evidence they actively trigger a soft spam flag that suppresses distribution.

The spam filter looks for patterns associated with coordinated posting and low-quality promotional content. Posts with many hashtags often look like SMM panel outputs to the classifier. Whether or not your post is actually spam is irrelevant. The pattern match is what matters.

One or zero hashtags seems to be the safe zone in May 2026. Two is a gray area. Three or more and you are likely taking a meaningful distribution penalty. This is frustrating for creators who built habits around hashtag discovery, but the data is pretty unambiguous on this point.

Long-form versus thread format

Threads are not dead but they have lost their edge. The current ranking model scores individual posts in a thread separately, each getting its own freshness decay and engagement velocity window. A thread with five tweets means five separate freshness counters, five separate velocity checks, and the first post doing the heavy lifting for distribution while posts two through five ride along on whatever momentum post one generates.

Long-form single posts work differently. X changed how it calculates dwell time on longer native posts in early 2026, and now a single post over 500 characters that a user reads all the way through generates a dwell signal that threads rarely produce tweet-by-tweet. The full read is worth more to the algorithm than five quick tweet-to-tweet swipes.

So the format shift we are seeing in our May data: posts between 500 and 1,000 characters are outperforming short posts by 20 to 35 percent on equivalent accounts, and they are outperforming equivalent threads by a similar margin. Not every piece of content works as a long post, but if you have been defaulting to threads out of habit, test the single long post format for a few weeks and see what the numbers say for your audience.

The first 30 minutes: still the only window that matters

This has not changed and probably will not. The For You distribution window for a new post is still governed almost entirely by the engagement velocity in the first 30 minutes. Hit the threshold in that window and the algorithm pushes the post into expanded distribution, typically lasting 24 to 48 hours before temporal decay makes it effectively invisible to new viewers.

Miss the threshold in the first 30 minutes and the post is essentially done. It does not get a second window. It does not catch up. Later engagement might trickle in from search or from people following back to your profile, but the For You rocket either launches at minute 30 or it stays on the pad.

What is the threshold? Based on our regression analysis, for an account with under 5,000 followers, roughly 10 to 15 genuine engagements in the first 30 minutes is enough to trigger expanded testing. For accounts 5,000 to 50,000 followers, the model normalizes by follower count, so the absolute number required scales but the per-follower ratio stays similar. The key word in all of this is "genuine." Engagement from low-credibility or obviously suspicious accounts actually hurts rather than helps, because the quality modifier on the signal can turn negative.

Account credibility signals and what they mean in May 2026

The credibility score is probably the least visible part of the ranking model and the most important for long-term reach. Here is what we know it includes as of May 2026, based on X engineering disclosures and our own testing:

• Account age. Older accounts with posting history score higher in the credibility vector. New accounts get a grace period of about 90 days and then are scored at a lower credibility tier until posting consistency accumulates.
• Posting consistency. Accounts that post regularly on a predictable schedule score higher than accounts with massive gaps and bursts. You do not need to post every day, but you do need a discernible pattern.
• Profile completeness. Profile picture, banner image, bio, and pinned tweet all present produces a meaningfully higher credibility score than a sparse profile.
• Engagement rate variance. This is the one that catches people out. A sudden spike in engagement without corresponding growth in reach or follower count signals a purchased or coordinated pattern to the classifier. Gradual improvement scores better than sudden jumps.
• Penalty history. Past warnings, temporary restrictions, or content removals lower the credibility floor. These penalties have a half-life that appears to be about six months, after which their scoring impact decreases.
• Premium status. X Premium gives a modest credibility bonus, probably in the five to fifteen percent range based on our estimates. Not a game-changer by itself but it compounds with other signals.
• Follow graph quality. Who follows you matters. Follower accounts with their own engagement histories and posting histories contribute positively to your credibility score. Dormant accounts or accounts with no posting history contribute neutral-to-negative.

What is winning in May 2026

Across the campaigns we are running and the test accounts we maintain, here is what is consistently outperforming benchmarks in May:

• Question-forward posts. Posts that open with or embed a genuine question to the audience drive reply velocity faster than declarative posts. Not cheap engagement bait questions like "agree or disagree?" but actual questions that the audience has opinions about and knows things about.
• Counterintuitive takes on topics the audience already cares about. Confirmation bias is strong but contrarianism gets replies. A post that says something accurate but unexpected about a topic your audience follows generates comment because people need to correct it, agree loudly, or add nuance. All three count equally in the algorithm.
• Native video with genuine dwell time. Video under 60 seconds that people watch all the way through, then reply to or bookmark, is producing exceptional composite scores right now. The dwell time signal on video is longer per unit than text, and bookmarks on video posts are spiking in our data for reasons we are still analyzing but the effect is real.
• Consistent medium-frequency posting. Accounts posting three to five times per day on a regular schedule are outperforming both infrequent posters and volume-heavy accounts posting twenty-plus times per day. The consistency signal is real and the volume penalty for over-posting is also real.
• Threads designed as individual value units. When creators write each tweet in a thread so it works as a standalone post (not just "continued from last tweet"), the per-tweet engagement scores stay high throughout the thread instead of dropping off after tweet two or three. Harder to write but measurably better in distribution.

What is losing in May 2026

• Posts with links in the first tweet body. We covered this above but it keeps coming up in our data. Still a meaningful suppression, still easy to fix by moving the link to a reply.
• Hashtag stuffing. Three or more hashtags in a post is a consistent predictor of lower distribution in our May data. It is not catastrophic on every post but over a content calendar the compounding effect is significant.
• Engagement bait with no real discussion follow-through. The model is getting better at identifying posts designed to generate engagement without genuine conversation. "Like if you agree, retweet to disagree" style posts perform worse than they did two years ago. The algorithm apparently checks whether replies are substantive and if reply quality is low, the weighting on those replies decreases.
• Repost-only strategies. Accounts that mostly repost others are seeing reduced distribution on their original content. The model tracks the ratio of original to reposted content and deemphasizes accounts that do not contribute original thought to the platform. Reposting is fine as a supplement, not as a primary strategy.
• Cold audience posting with no real follower base. Accounts under 500 followers are getting reduced For You testing volume compared to Q1. The minimum social proof threshold appears to be rising slightly. If you are under 500 followers, getting to 500 as quickly as possible is still the single most impactful thing you can do for your distribution ceiling.

What the X Premium monetization picture looks like right now

A lot of people are asking about money in May 2026. Specifically: can they actually earn from X and what does it take. The honest answer is yes, creators are earning, and the requirements have not changed dramatically from what was published in late 2025.

The current eligibility gates are: 500 followers and 5 million impressions in the three rolling months before the payout period, both met simultaneously, with an active X Premium subscription. Miss either gate and the payout for that period is zero.

What does 5 million impressions in 90 days actually require? At a consistent four posts per week, you need roughly 416,000 impressions per week, which breaks down to around 104,000 impressions per post. For most accounts that is a lot. For accounts with 20,000 to 50,000 followers and decent engagement rates, it is achievable through organic content. For accounts under 10,000 followers, it is very difficult organically and most creators we see hitting the threshold are using paid impression runway to close the gap.

The payout rate is not published by X and varies significantly by account. The real-world range we see reported by creators in May 2026 is roughly $0.25 to $1.00 per thousand monetizable impressions, where "monetizable" means impressions from other Premium subscribers. Since Premium subs are estimated to be about 65 percent of engagement on most general content accounts, your effective RPM on total impressions is lower than the headline rate. A conservative planning assumption is $0.40 per thousand total impressions for most accounts, rising to $1.00 or more for accounts in high-CPM niches like finance, B2B tech, and legal content.

The algorithm and purchased signals: what actually happens in May 2026

Since this is a Twitterz guide, we should address the direct question: does purchased engagement help or hurt in the current algorithm?

The answer, as always, depends entirely on how the signal is delivered. The algorithm cares about pattern, not source. If 1,000 likes arrive on a post in 90 seconds from accounts with no posting history, the quality modifier on those signals is negative and you would have been better off with no engagement. That is the cheap panel version.

If the same signal arrives over 30 to 90 minutes from accounts with genuine posting histories, the quality modifier is neutral to positive, the temporal velocity looks organic, and the signal contributes to For You threshold calculations the same way organic engagement would. That is what quality providers deliver and what we ship.

The composite signal point matters even more in May 2026 than it did a year ago. A mix of bookmarks, retweets, and likes in a realistic ratio scores exponentially better than the same total volume in any single signal. 100 bookmarks, 50 retweets, and 250 likes arriving over 45 minutes produces a wildly different algorithmic read than 400 likes arriving in the same window. The ratio beats the volume every time now. This is why the Engagement Suite ships as a composite package and not as individual signal products.

Specific tactics for May 2026 that actually work

Here is the practical list. Things you can do this week that will move numbers based on what we are seeing in current campaigns:

1. Move all links to the first reply, not the main post. If you have been putting links in main posts, try a two-week experiment where every link goes in a reply. Compare impressions. The difference should be visible within the first few posts.
2. End every post with a direct question or a blank-fill prompt. Not engagement bait, but a genuine question that requires your audience to think. "What is the one thing you wish X would actually fix?" outperforms "What do you think?" in reply velocity by a factor of three in our tests.
3. Post longer. If your typical post is under 200 characters, try pushing to 400 to 700 characters for a week. You do not need to pad with fluff. Just write more of what you actually have to say. The dwell signal benefit is real and consistently measurable.
4. Cut hashtags to zero or one. Pick your audience profile and trust the semantic classification. You do not need hashtag labels to help the algorithm understand what your post is about. It already knows.
5. Post at the same time each day. Consistency in posting time appears to feed the account-level credibility score in a way that random timing does not. Your audience learns when to expect you and that learning shows up as faster early engagement, which compounds into better distribution.
6. Spend 15 minutes per day leaving genuine replies on posts by larger accounts in your niche. The algorithm tracks your reply behavior as a signal of your topic authority. Being an active participant in conversations about your topic area, not just a broadcaster, improves your credibility score over time.
7. Clear the 500 follower gate if you have not yet. For anyone under 500 followers, the single most impactful action is hitting 500. The distribution ceiling below 500 is low and rising to 500 gives you access to a meaningfully larger testing pool. Our Followers Engine clears that gate inside 24 hours for $29, which is the math most serious operators accept quickly when they understand the alternative is weeks of grinding toward a gate that unlocks a ceiling.

How this guide is updated and how to read the dates

We update this page every time we have enough new campaign data to confirm a behavioral shift worth documenting. The "last updated" date at the top reflects the most recent edit. When X makes a public announcement about the algorithm we check it against our data and update within 14 days. When our data shows a clear shift without a public announcement (which is more common), we document the observed behavior and flag it as observed rather than officially confirmed.

If you want the static version of our full algorithm analysis, the comprehensive reference document is at The X Algorithm in 2026. That guide covers the foundational signal weights and is updated less frequently than this one. The monthly update posts like this one are the fast layer and the main guide is the deep layer. Both are free and always will be.

For the growth playbook that uses everything on this page as its foundation, see How to grow on X in 2026. For the products we build against this model, see all products.

Common questions we keep getting in May

My impressions dropped after the April 2026 changes. Are they coming back?

Probably yes, if you adjust your content behavior. Impression drops after algorithm shifts usually reflect the platform reclassifying your content against new weights. If you were heavy on hashtags, heavy on external links, or relying on retweets as your primary signal, switching to reply-forward content and removing link suppression signals should restore most of the lost reach within two to four weeks of consistent new behavior.

Should I buy X Premium right now?

If you are trying to hit monetization eligibility, yes. Premium is required for ad revenue sharing. If monetization is not your goal and you just want reach, Premium gives you a modest credibility signal that helps but is not the primary driver of distribution. The bigger drivers are content behavior and follower count. Premium is a multiplier, not a foundation.

How do I know if I am shadowbanned?

Look at your impressions per post relative to your follower count. A rough health check: if you are regularly getting impressions below two percent of your follower count on posts with zero external links and decent content, something is suppressing your distribution. It might be a content penalty, an engagement quality issue, or an account credibility problem. Our Algorithm Checker gives you a composite signal score you can use as a baseline to track week to week.

Does posting in X Spaces help your account?

There is weak evidence in our data that regular participation in Spaces, including hosting, correlates with slightly elevated profile discovery over time. This appears to be an indirect effect through follow graph expansion rather than a direct algorithm signal on your posts. It is not a primary growth lever but it is not hurting you either.