X algorithm changes, April 2026.

What happened in January that made April look the way it did

You cannot understand April without understanding January. In January 2026, X replaced its legacy recommendation algorithm with a new one built on the same architecture as Grok, the AI system developed by xAI. This was not a small tweak. It was a new engine in the same car.

The old algorithm was optimized around follower-normalized engagement volume. The new one is optimized around conversation quality and content dwell. Those are fundamentally different objectives and they reward fundamentally different content behaviors.

The old system asked: "did people engage, and how many of them?" The new system asks: "did people engage meaningfully, for how long, and with what quality of response?" The answer to both questions can be the same or completely different depending on your content type. For accounts whose content had been designed around the old system, April felt like walking in a room where all the furniture had been rearranged in the dark.

The signal weights we observed in April

We track campaign performance data across tens of thousands of posts. In April 2026, the clearest pattern that emerged from that data was a dramatic divergence between post types based on their reply-to-like ratio. Posts with high reply counts relative to likes outperformed posts with high like counts relative to replies by a factor that we had not seen before in our data set going back to 2022.

Replies: the new dominant signal

The current weight estimate for replies relative to likes is somewhere around 27 to 1. That number comes from independent regression analysis that multiple research teams, including our own, have applied to public post performance data. The exact figure varies by account tier and content category, but the directional conclusion is consistent: replies are worth dramatically more than likes in the current For You ranking calculation.

A post with 50 replies and 200 likes is being scored higher than a post with 20 replies and 2,000 likes. That is a significant departure from how most creators have been thinking about their content performance metrics.

Bookmarks: still the stealth signal

We documented this in our full algorithm guide, but it deserves mention here because it continued to be true through April: bookmarks carry roughly 2.5 times the weight of a like. The logic is the same as the reply weighting logic. Bookmarks signal intent to return, which is a stronger quality indicator than a momentary reaction.

The practical implication for April 2026 specifically: posts that generated bookmarks and replies together, even at relatively modest absolute volumes, were routinely outrunning posts with much higher like counts. Composite signals always beat single-axis volume in this model.

Retweets: still important, just not king anymore

Retweets carry about 2.0 times the weight of a like and remain the second-tier signal behind bookmarks. But they have been dethroned from their historical position as the primary distribution signal. The reason is probably practical: retweets are cheap to automate and were heavily gamed under the old model. The algorithm moved to signals that are harder to produce at scale without genuine human behavior.

The external link penalty: what we measured in April

The external link suppression is one of the most discussed observations in creator communities in 2026, and our April data confirms it is real and significant. Posts with external links in the main tweet body are receiving measurably lower initial distribution than identical posts without external links.

Our estimate based on A/B testing across matched post pairs: the suppression is in the range of 40 to 50 percent reduction in early impressions for posts with links versus the same post without links. This is not a marginal effect. It means that posting a link in your main tweet body cuts your potential reach roughly in half before anyone has seen your post.

Why does this happen? X has an obvious commercial reason to keep users on the platform. Posts that immediately take users elsewhere are less valuable to the platform's engagement metrics and ad model. The suppression is presumably intentional, even if undocumented. The workaround is posting links as replies on your own thread. This appears to avoid or substantially reduce the suppression while still giving your audience access to the link.

One nuance worth mentioning: not all links are equal. Links to X itself (quoting another tweet, linking to an X profile) do not appear to carry the same suppression. Only outbound links to external domains trigger the consistent suppression we have observed.

Hashtag spam classification: what the April data showed

Before the Grok-based model, hashtags were essentially ignored by the For You algorithm while still being functional for search. They were harmless noise. The new model appears to have added a spam classification layer that uses hashtag density as one signal among many when deciding whether a post looks like low-quality promotional content.

What we observed in April: posts with three or more hashtags performed consistently worse than posts from the same accounts with zero, one, or two hashtags, controlling for content quality and engagement pattern. The suppression was not as dramatic as the link penalty but it was statistically clear in our data, especially for accounts that were already in a lower credibility tier.

The practical recommendation is simple: if you are posting more than two hashtags per post, stop. The discovery benefit from hashtags in the For You feed is zero, and the risk of triggering the spam classifier is non-zero. The math does not work in favor of hashtags anymore.

The dwell time shift and what it means for content format

Dwell time as a ranking signal is not new. What changed in early 2026 is how it is calculated for text posts versus video. The short version: X updated how it measures and weights dwell time on long-form native text posts, and the change made single long posts more valuable than equivalent content split across threads.

The mechanism we believe is at work: when a user reads a long post from start to finish, the session duration on that post registers as meaningful dwell time with a positive quality signal. When the same user reads a thread, they generate brief dwell time on each tweet and a series of scroll actions between them, which individually look like quick engagement and in aggregate still register less overall dwell than the equivalent long-form post.

In April 2026, posts between 500 and 1,200 characters from our test accounts were consistently showing 25 to 40 percent higher impressions per post than posts under 200 characters from the same accounts. That is a meaningful format signal.

What drove reach losses for most accounts in April

If you had a bad April on X, here is the list of likely contributors ordered roughly by frequency in our support conversations:

1. External links in main tweet body. This accounted for the largest single chunk of unexplained reach drops we diagnosed. Accounts with heavy link-first content strategies saw the clearest declines.
2. Low reply velocity in the first 30 minutes. Accounts that had been fine under the old model because they were getting lots of likes but not many replies suddenly found the new model not caring about their likes as much.
3. Follower base quality shift. Accounts that had bought followers from low-quality pools in the past saw credibility score effects as the new algorithm recalculated follow graph quality. The accounts in the follower base that had low posting history and no engagement records subtracted from, rather than contributing to, the credibility vector.
4. Inconsistent posting. Accounts that had been irregular or that had posting gaps coming into April were starting from a lower temporal freshness baseline and needed more content before the algorithm treated them as reliably active.
5. Three or more hashtags per post. Less dramatic than the other factors but consistent enough to mention.

The paid promotion disclosure change: March 1, 2026

This one is separate from the algorithm but important enough to include because it affects a lot of the accounts we work with. On March 1, 2026, X updated its paid promotion disclosure policy with actual enforcement teeth.

The change: any tweet that contains paid promotional content and does not clearly label it as "Ad" or "Promoted Content" is subject to removal. First violation: post removed and read-only mode. Repeated violations: suspension. Immediate suspension for accounts created explicitly to violate the policy.

Crypto promotions carry additional geographic restrictions: they are prohibited as paid promotions in Australia, the EU, and the UK regardless of disclosure. Elsewhere they are allowed with proper labeling.

This matters for growth services work because accounts that are running undisclosed paid promotional content are at elevated suspension risk in 2026 regardless of how good their follower base quality is. If you are doing paid promotions, label them. The accounts that went dark in April and blamed the algorithm were often actually hit by the disclosure enforcement, not by the ranking changes.

How to recover reach lost in April 2026

If your April numbers were bad and you want to recover, here is the practical sequence:

Week one: stop the bleeding

Remove external links from your main tweet body. Cut hashtags to zero or one. Start ending every post with a direct question. Do not buy or source engagement from new providers unless you know the quality is clean. These three changes will stop the active suppression signals.

Week two and three: rebuild velocity

The algorithm responds to behavioral change over time, not immediately. Post consistently at the same time each day. Spend 15 minutes per day leaving substantive replies on posts from large accounts in your niche. This behavior signals topic authority and starts rebuilding your credibility vector. You will not see the full effect until week three or four but the process starts now.

Week four onwards: amplify what is working

By week four you should be able to see which posts are generating real reply velocity in the first 30 minutes. Those are the posts to amplify. A composite engagement push on a post that is already generating organic replies is additive and produces better algorithmic signals than a push on a cold post with no organic activity. Our Engagement Suite is built specifically for this kind of amplification and ships the composite signal, not single-axis volume.

The follower quality point and why April exposed it

April 2026 was the month a lot of accounts with large but low-quality follower bases suddenly noticed their reach was proportionally much lower than expected. An account with 100,000 followers that cannot produce 500 likes on a decent post has a quality signal problem, not a content problem.

The Grok-based model weights follow graph quality as a component of the account credibility score. A follower base full of accounts with no posting history, no engagement patterns, and no real profile activity contributes a negative signal to your credibility calculation. The platform has gotten better at identifying and down-weighting these accounts as credibility inputs.

If you bought followers from cheap bulk providers at any point in the past and have not gotten them cleaned up, April 2026 may have been the month the quality gap became visible in your metrics. The fix is adding a layer of high-quality followers that bring real engagement signals with them, which dilutes the low-quality base. Our Standard Followers Engine is built specifically for this purpose: real 30-day active accounts that contribute positively to the follow graph credibility calculation, with a 24-month warranty so the new layer does not degrade over time the way cheap followers do.

Looking ahead from April into May and June

The behavioral patterns that defined April are still in effect in May. Based on our ongoing campaign data, we do not expect a major structural change between now and Q3 2026. The Grok-based model is operating at what appears to be its intended behavior. Future monthly adjustments will probably be fine-tuning of signal weights rather than architectural changes.

The one wild card is X monetization policy. The paid promotion disclosure change of March 2026 signaled that X is willing to enforce policy more aggressively when they have a commercial reason to do so. We would not be surprised to see additional enforcement clarity around follower quality expectations and audience authenticity in H2 2026, particularly as X Premium monetization scales and the platform has financial incentive to ensure advertiser-friendly engagement quality.

For the current May 2026 data and tactics, see our May 2026 algorithm update. For the comprehensive signal reference, see the full algorithm guide. For the complete growth playbook, see How to grow on X in 2026.