How Does the Twitter/X Algorithm Work?

You write a tweet. Some get 200 views. The next one gets 20,000 with no obvious reason why. That's the Twitter algorithm at work — and in 2026 it works pretty differently than it did even a year ago.

In January 2026, xAI open-sourced the new Grok-powered version of the X algorithm, replacing the legacy ranking system with a transformer model that actually reads every post. That changed which tactics work and which don't. This post walks through what the X algorithm is doing under the hood, what signals it rewards, and what to actually do about it.

How Does the X/Twitter Algorithm Work in 2026?

The X algorithm is a recommendation system that decides, for each user, which roughly 30 posts to surface out of the ~500 million tweets posted every day. It runs as a three-stage pipeline. First, candidate sourcing narrows the firehose down to ~1,500 posts a given user might care about — half from people they follow ("in-network"), half from people they don't ("out-of-network"). Second, a neural network ranking stage, now powered by Grok's transformer architecture, scores each candidate by predicting how likely you are to reply, like, retweet, watch, or click. Third, heuristics and filters reorder the list — diversifying authors, suppressing posts you've already seen, and downranking content from accounts you've muted or blocked. The big shift in 2026 is stage two: instead of leaning mostly on early engagement velocity, the model now reads the actual content of your post and watches the actual video, deciding relevance from the text and media itself.

What's the Difference Between the For You and Following Feeds?

The two main feeds on X work in fundamentally different ways, and understanding the split matters because each one has its own rules. The For You feed is fully algorithmic. It mixes content from accounts you follow with out-of-network posts the model thinks you'll engage with, then ranks the whole pile by predicted action probability. This is where viral reach happens — a single post can travel far past your follower count if the model decides it's good. The Following feed used to be strict reverse-chronological. As of late 2025, X started using Grok to sort the Following timeline by predicted engagement too, though it stays inside accounts you follow. So even your "chronological" feed is now lightly ranked. For creators, this means optimizing for the For You feed (where most discovery happens) is still the highest-leverage move, but you can't fully escape ranking even from your own followers.

What Signals Does the X Algorithm Reward?

The open-sourced code shows the model combining many predicted actions into a single ranking score. The exact weights aren't published for the new Grok ranker, but per opentweet.io's 2026 analysis — using the canonical legacy weights from the open-sourced repo as a baseline — the relative picture is clear. Replies are worth dramatically more than likes — roughly 27x — and a reply that triggers a reply from the original author is worth around 150x a like. Retweets sit somewhere in between, bookmarks and profile clicks are weighted heavily because they signal serious interest, and plain likes are the cheapest signal of all. The algorithm also factors in dwell time (how long readers stop on your post) and video watch time, both of which the model can measure precisely. Negative signals — mutes, blocks, "show less of this" — hurt your distribution the same way positive ones help, so chasing engagement bait that annoys readers actively backfires.

Time decay and freshness

Why Threads and Replies Matter More Than Tweets

The algorithm cares about conversations, not applause — that's the whole game. Because replies are weighted around 27x a like — and a reply you respond to is worth roughly 150x a like — the highest-leverage thing you can do on X is post things people want to talk back to, and then actually reply when they do. That's why "reply guy" strategies (commenting thoughtfully under bigger accounts in your niche) have historically worked: every reply you leave is both visible to that account's audience and sends a strong signal that you're an active conversational user. Threads — multi-post chains — also get rewarded, because they keep readers on the platform longer (more dwell time) and tend to generate more replies per piece. The takeaway isn't to spam threads on every topic; it's to ask, before posting, "is this something a reader can actually respond to?" If the honest answer is no, expect mostly silence.

Does Twitter Blue/Premium Actually Boost Reach?

Premium subscribers — the paid blue checkmark tier — get a real, measurable distribution boost. Buffer analyzed 18 million posts across 71,000 accounts and found Premium accounts received roughly 10x more reach per post than non-Premium ones, though most of that gap reflects who buys Premium (already-active creators), not pure algorithmic favoritism. Industry estimates suggest a meaningful pure algorithmic boost on top of normal distribution, with replies from Premium users prioritized at the top of conversation threads — which is the most visible perk by far, because it puts your reply in front of every reader of a viral post. Premium also unlocks longer posts, edit windows, and creator monetization. Bottom line: Premium amplifies content that's already working. If your posts get crickets without a checkmark, paying $8/month won't fix that. But if you're already getting decent engagement, Premium meaningfully widens the funnel.

How to Optimize Your Content for the X Algorithm

Given how the model now works, here's what actually works:

1. Write posts people will reply to. Strong opinions, specific questions, and "wrong on purpose" framings (a hot take begging for correction) all generate the conversational signals the algorithm rewards most.
2. Use native media. Posts with images, native video, or GIFs consistently outperform plain text — both because they get more engagement and because the model can now read media directly.
3. Keep links out of the main post. Off-platform links are downranked, sometimes severely. Drop the link in a reply to your own post instead — a workaround the X community has used for years and that still helps.
4. Reply to your own replies. Every author-reply you send doubles the conversation signal and tends to pull the parent post back into more feeds.
5. Post consistently, not constantly. Three to five posts a day from a focused account beats twenty scattered posts. Burnout is the real algorithm killer.
6. Engage outside your own posts. Spending 15–30 minutes a day replying to bigger accounts in your niche is still one of the most reliable growth tactics — every reply is a discovery surface.

If you want a deeper playbook on growing from these signals, our guide on how to grow on Twitter walks through the daily routine in more detail, and what makes tweets go viral breaks down the formats that consistently break out.

Creator Jacob C. Edmunds breaks down the post-2026 shift after testing it on his own account:

How Does the X Algorithm Compare to the LinkedIn Algorithm?

The same content strategy doesn't translate cleanly across both platforms. The underlying mechanics are similar (both are engagement-prediction models), but the rewards are tuned very differently. X rewards speed, volume, and conversation. A great post can travel from 0 to 100k views in an afternoon, replies are worth more than any other signal, and posting multiple times a day is normal and expected. LinkedIn rewards depth, dwell time, and professional relevance. A LinkedIn post often takes 24–72 hours to peak, "meaningful" comments (longer than a few words) are weighted more than likes, and posting more than once a day usually hurts you. Off-platform links get suppressed on both platforms, but more aggressively on LinkedIn. The cross-platform takeaway: the same idea can work on both, but the format almost never should. A 280-character X post becomes a 6-paragraph LinkedIn post with hooks, line breaks, and a clear professional angle.

Start writing algorithm-friendly X content with Postory

Knowing the rules isn't the same as writing for them daily — actually producing content that triggers replies, dwell time, and reposts takes ongoing reps. Postory's AI post writer drafts X posts tuned for the conversational signals the algorithm rewards (clear hooks, opinion-driven framing, native-media-friendly length), and the scheduler queues them up so you can stay consistent without living inside the app.

Try Postory free — create algorithm-friendly X content and schedule it automatically.