The X algorithm, what actually changed in 2026.

Between November 2025 and April 2026, X shipped seven meaningful changes to its For You ranking model. None of them were publicly announced with specifics. All of them have been inferred by operators comparing distribution data before and after, corroborated by leaks from employees, and confirmed through indirect channels over the following weeks. This post is the consolidated operator-grade summary of what changed and what it means for content strategy in 2026.

Change one: bookmark reweighting (November 15, 2025)

Bookmarks went from roughly 1.0x weight (equal to likes) to approximately 2.5x weight in the composite engagement score. Retweets stayed at 2.0x. Replies dropped slightly from 1.8x to 1.6x. Dwell time above 4 seconds was folded into the composite at meaningful weight for the first time.

Operator impact: content optimized for likes now earns 25 to 45 percent less distribution than it did pre-reweight. Content optimized for bookmark-worthy value (frameworks, decision guides, reference material) earns 30 to 80 percent more. Full teardown of this change here.

Change two: reply-thread decay curve flattening (December 8, 2025)

The algorithm previously applied a steep time-decay to tweets: impressions dropped 70 percent or more by hour 6 and 90 percent plus by hour 24. After this change, high-quality threads (defined by the composite engagement profile in change one) now have a flatter decay curve, maintaining meaningful distribution for 36 to 72 hours instead of 12 to 24.

Operator impact: posting time matters less than it did. A great thread posted Friday at 4pm that would have died by Saturday morning under the old decay now earns meaningful distribution through Monday morning. Bad threads still die fast. The spread between great and average content widened.

Change three: cold-start threshold tightening (January 3, 2026)

The minimum quality threshold for a new account or a new tweet to be eligible for For You candidate generation tightened. Previously, accounts under 500 followers could occasionally break through with a single viral tweet. After the change, the threshold is closer to 800 followers plus an account-quality score minimum that excludes most accounts under 6 months old with low engagement history.

Operator impact: cold-start is harder than before for new accounts. Existing accounts above the threshold are unaffected. Accounts below it rely more heavily on second-degree amplification (replies under big accounts, quote tweets from established voices) to reach the For You candidate pool. Our growth playbook addresses the cold-start situation specifically.

Change four: quote-tweet downweighting in some contexts (January 22, 2026)

Quote tweets that add criticism or disagreement to the original post no longer carry the same signal weight they did previously. The algorithm seems to distinguish between constructive quote tweets (adding information, context) and dunk quote tweets (criticism without substance). The latter carry reduced amplification.

Operator impact: dunk content is less rewarded. This has compressed the "main character discourse" cycle on the platform by roughly 30 percent in observed frequency. Accounts that built audiences on dunk content have seen 20 to 50 percent reach declines. Accounts that engage in substantive quote-tweet discourse have seen reach gains.

Change five: Premium boost recalibration (February 14, 2026)

X Premium subscribers have always gotten a distribution boost. The boost size was approximately 1.4x on composite ranker scores. After this change, the Premium boost appears to be more context-dependent: larger (perhaps 1.6x) for Premium users posting substantive content, and smaller (perhaps 1.1x) for Premium users posting low-quality spammy content. The "gaming Premium for distribution" strategy is less effective than it was.

Operator impact: Premium is still worth it for creators publishing quality content. It is meaningfully less worth it for accounts using Premium to amplify low-effort posts. Premium subscribers posting at volume without quality curation have seen gains partially reversed.

Change six: niche clustering reintroduced (March 4, 2026)

The algorithm reintroduced niche-cluster amplification. Content that earns high engagement from a concentrated niche cluster (say, the crypto founder cluster or the indie hacker cluster) now gets amplified within that cluster at roughly 1.3x the composite score, even if overall cross-platform engagement is modest.

Operator impact: niche specialization pays. A tweet that does 500 impressions across everyone but 2,000 impressions within your specific niche now outperforms a tweet that does 1,500 impressions across everyone but only 600 within your niche. Building a recognizable niche identity is more valuable than it was in 2025.

Change seven: media content reweighting (April 1, 2026)

Video and image content are being weighted slightly higher than pure text for equivalent engagement profiles. The size of the bonus is modest (perhaps 1.1 to 1.2x) but the direction is clear and new. Text-only tweets still earn substantial distribution if the engagement profile is strong, but the playing field has tilted toward media.

Operator impact: adding a relevant image or short video to a text tweet now earns measurable incremental reach. This is not a mandate to post video content; text still works. It is a tailwind for operators who can easily produce media.

The compounded effect on operator strategy

The seven changes in aggregate produce a coherent shift in what X rewards.

1. Substantive over reactive. Bookmark reweighting, dunk downweighting, and Premium quality-dependent boost all point the same direction. Content that makes people think, bookmark, or genuinely reference later wins. Content that just provokes clicks, agreement, or outrage loses ground.
2. Niche over broad. Niche clustering reward means identifiable specialists outperform generalists for equivalent engagement volume. Picking a lane and sticking to it is now measurable content strategy.
3. Evergreen over ephemeral. Flattened decay curves mean a high-quality piece of content produced once now earns distribution for days. This rewards investing more in fewer pieces.
4. Media as tailwind, not requirement. The media-content bonus is small enough that text-only operators are not disadvantaged in absolute terms, but operators who can easily produce media have an incremental edge.
5. Cold-start is tougher. New accounts below the cold-start threshold face a higher barrier than they did in 2025. Organic rise from zero is still possible but slower.

What this means for content mix

The practical action list for the next 90 days

1. Shift content mix toward bookmark-worthy content (frameworks, decision guides, teardowns). Target 50 to 70 percent of posts in this category.
2. Sharpen niche identity. Your bio, pinned tweet, and last 20 posts should together paint a clear picture of what you cover. Generic accounts lose distribution under niche clustering.
3. Reduce reaction content. Fewer news takes, fewer dunks, fewer agreement-seeking posts.
4. Add one media element per week at minimum (image or short video tied to substantive text).
5. Invest more in fewer pieces. One long thread per week, edited carefully, beats seven short tweets of average quality.
6. If you are stuck in cold start (under 1,000 followers), focus on reply-under-big-accounts strategy from the cold-start playbook or consider a targeted growth purchase to cross the threshold.

What we are watching for next

Based on patterns from previous recalibration cycles, we expect X to ship at least two more meaningful changes before end of 2026. The most likely areas: further bookmark weight adjustments (possibly a pullback if operators over-index), reply quality scoring that distinguishes substantive from perfunctory replies, and Premium boost restructuring tied to creator monetization metrics. We will update the journal as each change ships and we have enough data to describe it reliably.

If you want the deeper pre-2026 context, our full algorithm guide covers the candidate generation and neural ranker architecture that these changes are built on top of.