Bookmarks, likes, retweets. Which one the algorithm actually weighs.
Everyone knows bookmarks matter more in 2026. Almost nobody explains the exact weight difference, why X changed it, or what it means for what operators should optimize for. Here is the full breakdown with the reweighting history, the measurable impact on For You distribution, and the two engagement mix profiles that the algorithm rewards hardest right now.
In November 2025 X quietly reweighted the engagement signals in its For You ranking model. Bookmarks went from being roughly equal to likes in signal weight to being roughly 2.5 times a like. Retweets stayed at about twice a like. Replies dropped slightly. Dwell time, which is how long users actually pause on your tweet in the feed, crept up in importance. The reweighting was not announced. It was discovered by operators watching their own distribution data and comparing notes across niches. By January 2026 the consensus read of the new weights was widely circulated among serious operators and almost nowhere in the mainstream Twitter advice ecosystem.
This post walks through exactly what changed, why X changed it, what the new weights do to your content strategy, and which engagement mix profiles now produce the most For You distribution. If you are still optimizing for likes as your primary signal, you are playing the 2023 game.
The rough weighting, before and after
| Signal | Pre-November 2025 weight | Post-November 2025 weight | Shift |
|---|---|---|---|
| Bookmark | 1.0x | 2.5x | Up 150 percent |
| Retweet | 2.0x | 2.0x | Unchanged |
| Reply | 1.8x | 1.6x | Down 11 percent |
| Like | 1.0x | 1.0x | Baseline (unchanged by definition) |
| Quote tweet | 2.2x | 2.2x | Unchanged |
| Dwell time above 4 seconds | Minor | Meaningful (partial fold-in) | Up notably |
These are our estimates based on observed distribution patterns, not published figures. X has not confirmed the weights. The estimates correlate strongly with what operators across roughly 40 different accounts have observed and reverse-engineered independently.
Why X made this change
The stated reason, from what has leaked in employee chatter and third party coverage, is that bookmarks are a stronger signal of "content the user wants to revisit," which correlates with content quality better than likes do. Likes can be reflexive, performative, or thoughtless. Bookmarks require the user to think "I want to come back to this," which is a higher-commitment action. Higher-commitment engagement signals tend to correlate with content the algorithm wants to reward because they predict future engagement from other users at a higher rate.
The unstated reason, which is at least as important, is that likes had become increasingly gameable. Paid like services existed at every price point. Bot accounts were trained specifically to like popular content to maintain credibility, then quietly amplify target content alongside it. By reweighting toward bookmarks, X reduced the efficacy of the cheap like-buying attack without dramatically restructuring its UI or announcing a crackdown.
What this means for content
The implication is direct and actionable. Content that earns bookmarks now outperforms content that earns likes at roughly 2.5 to 1. That changes what to write. Bookmark-worthy content tends to be:
- Frameworks or decision guides the reader expects to reference later
- Resource lists, tool compilations, or curated collections
- Step-by-step explanations of something the reader might do themselves later
- Data or charts the reader might want to cite
- Deep thread-format teardowns of a topic
- Tweet-length quotes or insights worth quoting in future conversations
Like-worthy content tends to be:
- Jokes and one-liners
- Opinions the reader agrees with emotionally
- News commentary
- Personal updates
- Reactions to events everyone is already talking about
The two categories overlap occasionally (a great joke is also quotable and therefore bookmark-worthy) but mostly they do not. If you have been writing like-worthy content for the last three years, the 2026 algorithm will distribute it less than it used to. Your content has not gotten worse. The weights changed under you.
The two mix profiles that the 2026 algorithm rewards hardest
Using the new weights, you can reverse-engineer which composite engagement shapes produce the highest ranker scores. Two profiles dominate.
Profile one: the viral framework
A tweet or short thread that contains a framework, step list, or decision tree that readers recognize as useful for future reference. Typical mix: 100 likes, 25 bookmarks, 8 retweets, 6 replies. Composite score at new weights: 100 × 1.0 + 25 × 2.5 + 8 × 2.0 + 6 × 1.6 = 188.1. High ratio of bookmarks to likes (25 percent) pushes this into the algorithm's "ratio health" bonus modifier, adding another 10 to 20 percent to the effective score. For You distribution for this shape: strong.
Profile two: the quotable insight
A single-tweet statement of something punchy and true, or a short thread ending on a statement worth quoting in future conversations. Typical mix: 300 likes, 40 bookmarks, 15 retweets, 12 replies. Composite score: 300 + 100 + 30 + 19.2 = 449.2. Higher absolute numbers, lower bookmark-to-like ratio (13 percent), still a healthy profile. For You distribution for this shape: very strong in absolute terms because the like volume is high and the supporting signals are balanced.
The losing profile
For contrast, the profile that loses under the new weights is the "reaction post." Typical mix: 800 likes, 4 bookmarks, 20 retweets, 30 replies. Composite score: 800 + 10 + 40 + 48 = 898. The absolute numbers are the highest of the three but the bookmark-to-like ratio is 0.5 percent, which now triggers a ratio-health penalty of roughly 0.7 to 0.85. Effective score after penalty: 628 to 763. Under the old weights, this profile would have had the most For You distribution of the three. Under the new weights, it distributes less than the framework post despite having 5 to 8x the raw engagement.
How to actually earn bookmarks
You cannot optimize for bookmarks by asking for them. "Bookmark this" CTAs rarely convert above baseline. Bookmarks are earned by writing content the reader thinks, in the two seconds after reading it, "this is worth coming back to." The mental model is future reference value. Ask yourself: would a reader, six weeks from now when they need this information, remember to look for it. If yes, it is bookmark-worthy. If no, it is probably like-worthy at best.
Formats that earn bookmarks disproportionately:
- Numbered lists with specific tactical content ("Seven things to check before shipping a new SaaS feature")
- Decision frameworks ("How to decide whether to raise a seed round: four criteria")
- Cheatsheets or reference tables
- Long thread teardowns of specific case studies
- Mini-guides under 15 tweets that walk through a process end-to-end
Formats that rarely earn bookmarks, regardless of quality:
- Reactions to news
- Personal updates
- Jokes and puns
- One-off opinions
- Replies to other tweets (bookmarks on replies are technically possible but statistically rare)
The practical tactical shift
If you have been optimizing for likes, shift your content mix toward roughly 60 to 70 percent bookmark-worthy posts and 30 to 40 percent like-worthy posts. The like-worthy posts are not useless, they maintain audience connection and earn baseline engagement, but the bookmark-worthy posts are what actually buy you incremental For You distribution in 2026.
If you are running engagement campaigns (buying likes, retweets, bookmarks to prime algorithm signal), shift your purchase mix similarly. Our engagement product defaults to the updated ratio because we updated the defaults when we caught the reweighting in early December. If you are buying engagement elsewhere and your provider is still delivering predominantly likes, you are paying for a signal that produces less distribution than it used to.
How to measure the effect on your own account
Pull your analytics. Compare the 30 days before November 15, 2025 against the 30 days after. Look at two things: your overall impression count, and the bookmark-to-like ratio on your top ten posts from each period. Most accounts will find that impression count stayed flat or dropped slightly while top-post bookmark ratio rose. Accounts whose bookmark ratios stayed low lost more reach than accounts whose bookmark ratios rose.
If your bookmark ratio has stayed low and your reach has dropped, the fix is in your content, not in your strategy. Rewrite your next 20 posts around future-reference value and watch the ratio (and the distribution) shift within three to six weeks.
Further reading and tools
If you want the mechanics behind the For You ranker more broadly, our algorithm guide covers candidate generation and neural ranking in depth. If you want a specific product we built around the viral framework profile described above, the bookmarks product is one of the few places in the category that sells bookmark engagement specifically. If you are starting from scratch and want to understand follower dynamics that precede engagement optimization, our small account post is the adjacent reading.