Plumyra

X algorithm FAQ

25 straight answers on how X ranks posts in 2026. Tap a question for the short answer, or open its guide.

How X ranks posts

How does X decide who sees my posts in 2026?

The For You feed is ranked by a Grok-based model called Phoenix. For each viewer it predicts how likely they are to reply, share, like, mute, report and so on, multiplies each chance by a weight, and adds them up. The highest totals are shown first.

So nothing is scored on your raw like count. A post ranks well when the model expects the people seeing it to act on it.

Source: xai-org/x-algorithm, home-mixer

Open the guide

What counts most: likes, replies or shares?

Shares by copy link count most (20.0), then replies, quotes and DM shares (5.0 each). A like is worth 0.66. Replies between people who follow each other get an extra +15 since July 2026.

The negative weights are much bigger: a predicted report is −234, a mute −58.8, "not interested" −52.3 and a block −31.2. These are the published defaults from October 2026:

For You ranking weights, October 2026 defaults
Share via copy link20.0
Reply5.0 (+15.0 between mutual follows)
Quote5.0
Share via DM5.0
Follow the author4.0
Share (general)2.0
Repost1.0
Like0.66
Post click0.3
Open link0.2
Dwell0.05 + continuous terms
Profile click0.0
Not dwelled−0.02
Block author−31.2
Not interested−52.3
Mute author−58.8
Report−234

Source: home-mixer/params/param.rs, Oct 2026 defaults

Open the guide

Is one report really worth 354 likes?

No. That viral claim misreads the code. The weights multiply each viewer's predicted chance of reporting or muting you, not your real report count. X said so in August 2026.

It's still true that anything likely to annoy people, such as bait, rage, off-topic rants or flooding the feed, drags a post down hard, because those weights are so large.

Source: param.rs; X statement, Aug 2026

Open the guide

How long does a post stay in For You?

48 hours. After that it leaves the pool the ranker picks from.

Fresh posts under two hours old with few impressions also get a cold-start lift, for accounts under 50,000 followers. Reply to commenters inside those first two hours.

Source: AgeFilter and ColdStart params

Open the guide

Does X care whether people read the whole post?

Yes, more than before. Since October 2026 the model predicts time spent much better, and retrieval is trained to surface posts people read all the way through.

Cut the throat-clearing, keep one idea per post, and make long posts worth finishing.

Source: x-algorithm release notes, 8 and 13 Oct 2026

Open the guide

Does X penalise AI-written or generic posts?

A Grok classifier gives every post a quality score and a "slop" score, with a pass threshold of 0.4. Template phrasing, generic motivation and AI-flavoured wording are exactly what it, and readers' mute reflexes, catch.

Write from something you did or saw, with your own numbers and words.

Source: grox/classifiers/content/banger_initial_screen.py

Open the guide

Writing posts that travel

What kind of post gets shared?

Copy-link and DM shares are the heaviest positive signals. Before posting, ask whether someone would send it to a colleague with "you need to see this".

Checklists, a real number, a framework, or an answer to a question your niche keeps asking all qualify.

Source: param.rs

Open the guide

How do I get more replies without engagement bait?

End with a question people can answer from their own experience, or take a position someone could disagree with.

Avoid "thoughts?", "agree?", "like if" and "RT if". X discounts impressions from posts that tell people how to engage.

Source: Nikita Bier on engagement bait, Aug 2026

Open the guide

How important is the first line?

It is the post for most readers. Long posts cut off after about 280 characters in the feed, and most people seeing it don't follow you.

Open with a specific claim, result or question that makes sense on its own. Don't start with "this" or an inside reference.

Source: Feed truncation; out-of-network discount

Open the guide

Do "here's how:" and "read more" hooks work?

No. A post click is worth 0.3 and a profile click 0.0. Curiosity gaps earn almost nothing. Put the point in the post and earn the read instead.

Source: param.rs

Open the guide

Do hashtags help reach?

No. The ranker has no hashtag signal, and X's spam rules target posts stuffed with them. Use none or one.

Habitual "BREAKING" is also deprioritised.

Source: X spam policy; Nikita Bier, Apr and Aug 2026

Open the guide

Do video and long posts do better?

They can, because they earn time spent. Use long-form when the idea needs room, native video over about 60 seconds when you can, and images only when they carry information.

Never repost watermarked clips from other platforms (about a 90% impression cut) or copy viral text.

Source: VQV gating; Nikita Bier, Apr to Aug 2026

Open the guide

Posting habits and growth

How often should I post on X?

One to five original posts a day, spaced 30 to 90 minutes apart. Your second and third post in a viewer's session are scored down (author-diversity decay), so bursts waste posts.

Nikita Bier's bar: one unexpected insight from your own experience every day, under five sentences, for six months, in one lane.

Source: author_diversity_scorer.rs; Nikita Bier, Sep 2025 and May 2026

Open the guide

When is the best time to post?

There's no time-of-day signal in the ranker. Timing matters only because the first two hours decide the cold-start lift and the reply-back loop.

Post when your readers are online and you can reply. Plumyra's Grow tab shows which of your own time slots have done best.

Source: ColdStart params

Open the guide

How do small accounts grow on X?

Through replies. Under about 1,000 followers, your replies are where strangers find you. They're seen inside threads, not pushed to For You.

Grok scores every reply from 0 to 3 and runs an extra spam check on small accounts. A concrete addition, a counter-example or a pointed question scores; "this" or "great post" scores about zero.

Source: grox reply_spam classifier

Open the guide

Does following people back help?

Yes. Since July 2026, replies to people you mutually follow get a +15 boost. Build two-way relationships in your niche: reply, and follow back the people who engage with you.

Source: docs/BIDIRECTIONAL_BOOST_CHANGE.md

Open the guide

Reach problems and X's report

Am I shadowbanned on X?

X doesn't use the word, but it does put visibility labels on accounts and posts that limit reach. Its own Under the Hood report lists them, and that's the only direct view of what X has flagged.

Never give your X login to a "shadowban checker". The real report is free and comes from X.

Source: X Under the Hood report

Open the guide

Who can see X's Under the Hood report?

Accounts at least a year old with 10 or more posts in the previous month. No Premium is needed. Each report covers one calendar month and appears about ten days after the month ends.

X lists the exact posts behind a label only when 5% or fewer of your posts carry it; otherwise it shows a count.

Source: xai-org/x-algorithm, under-the-hood

Open the guide

The X app has no download button for the report. What do I do?

That's expected: the app shows the report on screen only. In Plumyra, choose Get X's report and type in what it shows. Usually that's "None in this period" twice.

On the X website you can download the file and add it instead.

Open the guide

Should I delete old posts that might hurt my account?

Delete the ones that could be read as spam, bait or rule-breaking, plus replies left under deleted posts. They can weigh on how X and readers judge the account.

Plumyra flags candidates from their text and timing only, so read each one before deleting.

Open the guide

About Plumyra's report

Where does Plumyra's report come from?

From your posts, read in your browser: your X archive, the helper, pasted posts, or X's official API with your own key. It checks them against the X Rules and X's open-source ranking code, plus X's own Under the Hood report if you add it.

Source: The X Rules

Open the guide

How accurate is it?

X's own report is exact. Everything else is Plumyra reading captions and timing. It doesn't see images or videos, and X doesn't publish everything it does, so treat flags as candidates to review, not verdicts.

Open the guide

What does the health score measure?

Out of 100: labels from X (30), risky posts (25), reach trend (15), posting habits (20) and account trust signals (10). Parts Plumyra can't see are left out, and the rest is scaled up.

Open the guide

Is it safe? Does Plumyra see my X account?

Plumyra never logs in to X, never posts, and never asks for your password. Your posts stay in your browser. AI keys and X API keys pass through Plumyra's relay to Google or X without being saved.

Open the guide