VisionHow do you improve Twitter algorithm ranking?
To improve Twitter algorithm ranking on X, write posts that earn replies, quotes and shares, and avoid anything that makes people mute or report you. In the 2026 published settings, a reply carries a default weight of 5.0 and a like 0.66, while muting an author carries -58.8.
Most people start out chasing likes. A post with lots of likes feels like a win. But the code tells a different story: likes carry one of the smallest positive weights, and replies carry several times more.
Here's the good news: ranking on Twitter (now X) isn't a mystery you have to guess at anymore. You can line your habits up with what the code actually rewards, one step at a time.
What decides your ranking on X?
Your ranking on X is a score built from predictions. For each post and each viewer, a model called Phoenix predicts the chance of many actions, and X adds those chances together using weights. Higher total, higher position in the feed.
The xai-org/x-algorithm README describes it as a weighted sum: each predicted action's probability times its weight. Positive actions add to the score. Negative ones take away. The weights live in the repo's param.rs settings file, and X can change them at any time.
For the full pipeline, my guide on how the X algorithm works in 2026 covers every stage. This page sticks to what you can do about it.
How have the ranking weights changed since 2023?
The weights changed a lot between the 2023 release and the 2026 code. Replies still beat likes by a wide margin, but shares, quotes and following the author now carry real weight of their own.
| Action | 2023 weight (April 5, 2023) | 2026 default weight |
|---|---|---|
| Like | 0.5 | 0.66 |
| Repost | 1.0 | 1.0 |
| Reply | 13.5 | 5.0 |
| Profile click | 12.0 (good profile click) | 0.0 |
| Report | -369.0 | -234.0 |
The 2023 column comes from the 2023 heavy ranker README. The two models are different, so don't compare the numbers one to one. Compare the pattern instead: conversation counts, and annoyance costs.
7 steps to improve Twitter algorithm ranking
These seven steps map directly to weights in the 2026 code. None needs a big audience or a budget, and each one gets easier with practice.
1. Write posts that invite a reply
A reply's default weight is 5.0, while a like's is 0.66. Ask a real question, take a clear position, or share a specific number from your own work. Open-ended posts get answers. Finished thoughts get nods.
2. Reply back, especially to people you know
Answer the replies you get, and build mutual follows. The 2026 file has a "bidirectional follow reply weight boost" of 15.0. Per X's note on the July 2026 change, it raises the weight on the predicted chance someone replies to an original post from an account they mutually follow.
3. Make posts worth sharing
Sharing carries big weights: 2.0 for a share, 5.0 for a share via DM, and 20.0 for copying the link. Checklists, clear how-tos and useful data get passed along. This one makes a bigger difference than most people realize.
4. Give people a reason to stay
The code predicts dwell, meaning time spent on a post. DwellWeight is 0.05 and ContClickDwellTimeWeight is 0.4. A post people actually read beats one they skim past. A quick check with a post length checker helps you keep longer posts tight.
5. Earn the follow
FollowAuthorWeight is 4.0. When a post makes someone want more from you, that counts. A clear bio and a consistent topic make the follow decision easy.
6. Avoid the negative signals
This part is actually really easy to get right. Skip rage bait, misleading hooks and spammy tags. "Not interested" carries -52.272, mute -58.8, block -31.2 and report -234.0. One annoyed crowd can undo many good posts.
7. Post when your audience is active
The For You code removes posts older than 48 hours. Early engagement has the most time to work. Post when your followers are online so the first predictions have real signals to learn from.
What should you stop doing?
Stop chasing likes alone, stop flooding the feed, and stop tricks that annoy readers. Each of these either has a low weight or triggers a penalty in the code.
- Like-chasing. A like is the smallest positive weight in the table.
- Flooding. Each extra post from the same author in one viewer's feed gets a shrinking multiplier.
- Engagement bait. "Like if you agree" posts may earn taps, but they also invite "not interested."
If your numbers are already sliding, my guide to recovering declining Twitter engagement starts with a diagnosis before any fix.
What this means for you
Improving your ranking comes down to one shift: write for the reply, not the like. Every step above is a version of that idea.
I'd start with steps 1, 2 and 6 for two weeks and watch your replies. For writing help, my guide to writing engaging Twitter posts has formats that invite answers. To measure progress, the guide to increasing Twitter engagement rate shows what to track. If you use XVision, my free extension, its growth dashboard rates each of your posts against your usual views, so you can see which habits are working.
You've got this
You can improve Twitter algorithm ranking without hacks. The weights are public, and they reward the same things good writers already do: start conversations, be useful, respect people's time. Pick one step this week. Once it becomes second nature, add the next.
Frequently Asked Questions
Do likes matter for Twitter ranking at all?
Yes, likes matter, but only a little. The 2026 default like weight is 0.66, compared with 5.0 for a reply and 5.0 for a quote. Likes also teach the system about a viewer's interests, which helps match future posts to similar people.
Does X Premium improve algorithm ranking?
Partly, for replies. X's Help Center says replies from verified accounts get a slight preference in reply ranking, and that it is testing those levels. The 2026 For You README makes no statement about a Premium boost in For You, and X hasn't published any multiplier.
How fast can ranking improve after changing habits?
Ranking can respond within days, because Phoenix reads each viewer's recent engagement history and For You only considers posts from the last 48 hours. Building a steady audience that engages takes longer, usually several weeks of consistent posting on a clear topic.
Do replies to my own posts count?
X hasn't documented how replies to your own posts affect ranking. The ranking model predicts whether a viewer will reply, so the value comes from replies by other people. Answering their comments keeps the conversation going and gives more viewers a reason to join in.
Are the 2026 weights final?
No, the weights are not final. The 2023 heavy ranker README said weights could be adjusted at any time, and the 2026 repo has had several updates since January 2026. Check the param.rs file on GitHub for the current default values before relying on any number.