How the X algorithm works in 2026
X ranks the For You feed with a Grok-based model that it published as open source. For every candidate post, the model predicts what you're likely to do with it: like it, reply, repost, quote, keep reading, open the author's profile or follow them. It also predicts the bad outcomes: tapping “not interested”, muting, blocking or reporting. Those predictions are weighed into a single score, and the highest-scoring posts fill your feed.
Two things follow. First, posts that start conversations win, because replies and quotes are strong positive signals. Second, anything that makes people back away costs you reach, even if some people like it. Bait, spam-like repetition, walls of hashtags and posts that only point somewhere else all do that.
What this checker scores
Seven factors, each from 0 to 100, combined into one algorithm score:
| factor | what it measures |
|---|---|
| Reach trend | Median views of your newest posts compared with the ones before. A falling trend is the clearest sign the algorithm has cooled on you. |
| Engagement | The share of views that turn into a like, reply, repost, quote or bookmark. |
| Conversation potential | Jev scores each post on how likely it is to get replies and quotes, which X's ranker weighs heavily. |
| Opening lines | Whether your first line stops the scroll. The ranker predicts how long people stay on a post. |
| Originality | Your own thinking compared with bare links, reposted news or filler. |
| Topic focus | X matches posts to people by interest, so a clear niche is easier to place in the right feeds. |
| Negative-signal risk | Bait, spam-like phrasing, link-heavy or hashtag-heavy posts make people tap "not interested", mute or block, and those signals count against you. |
The numbers (views, engagement, links, hashtags, how often you post) are computed from X's public counts. The judgement calls (will people reply, does the first line hook, is it bait) are answered by Jev. A score of 70 or more means the algorithm is working for you. Below 45, it's working against you.
Why your reach on X dropped
A sudden drop usually has one of four causes. The first is a shadowban: a search ban or hidden replies, which the shadowban checkertests in seconds. The second is a run of posts with outside links. The third is engagement-bait phrasing that people answer with “not interested”. The fourth is simply posting less often, which leaves the ranker fewer posts to test with new audiences.
How to make the X algorithm work for you
- 01Write for replies
End with a real question or a take people can disagree with. Replies and quotes are among the strongest signals the ranker predicts.
- 02Hook in the first line
Lead with the point, a number or a surprise. The ranker predicts how long people stay on a post.
- 03Keep it native
Post the content itself on X and put any outside link in the first reply.
- 04Pick a lane
Stick to one to three topics, so the ranker knows whose For You feed to put you in.
- 05Skip the bait
Drop "like if…", "RT to win" and follow-for-follow. They trigger the negative signals that cost you reach.
- 06Add media
An image, chart or short video makes people stop and read.
- 07Post steadily
A few posts a week at a human pace beats bursts followed by silence.
About Jev
Jev is a System One model from TypeSafe AI, released in September 2026. Most AI models write text. Jev answers typed questions instead: a score on a scale, a yes/no probability or a pick from a list, in roughly a tenth of a second. For this checker, Jev answers the same questions for each of your last 40 posts — would people reply, would they back away — plus a few questions about your timeline as a whole. Our code then combines those answers with your real view and engagement numbers.
Building something on X data? The posts this audit reads come from the twitr.sh timeline tool. It's paid per call in USDC, with no API key, and works from Claude Code, Cursor or any MCP client. See all X tools.
Questions
- How do I know if the X algorithm is working for me?
- Enter your username above. The checker compares the median views of your newest posts with the ones before, measures your engagement rate, and has Jev score your posts on what X's For You ranker rewards. You get a score out of 100 and a verdict: working for you, mixed, or working against you.
- How does the X algorithm work in 2026?
- X ranks the For You feed with a Grok-based model that it published as open source. For each post it predicts how likely you are to like, reply, repost, quote, read for a while or follow the author, and how likely you are to react badly (not interested, mute, block, report). The positive and negative predictions are weighed into one ranking score.
- What is Jev?
- Jev is a System One model from TypeSafe AI, released in September 2026. Instead of writing text it returns typed answers — a score, a probability or a choice — in roughly a tenth of a second. That makes it a good fit for scoring dozens of posts at once on the same questions.
- Why did my reach on X drop suddenly?
- The usual causes are a shadowban (search ban or hidden replies), a run of posts with outside links or heavy hashtags, engagement-bait phrasing that triggers "not interested" taps, or posting less often. Run the shadowban checker first, then fix whatever this audit flags.
- Do links really hurt reach on X?
- Posts with outside links have historically gotten less reach on X, because they send people off the platform. A common workaround is to post the content natively and put the link in the first reply.
- Does X Premium help with the algorithm?
- Premium accounts get a boost in reply ranking, but the For You ranker still predicts engagement and negative reactions for every post. Premium won't rescue posts people don't engage with.
- Is the X algorithm checker free?
- Yes. You get 3 free audits a day with no login, and results are cached for 6 hours. Signed-in users can run extra audits for $0.02 each from their twitr.sh balance.
- Can I audit someone else's account?
- Yes. The audit only reads public posts and public numbers. It never logs into an account, and it can't audit private (protected) accounts.
- How accurate is the score?
- The reach and engagement numbers come straight from X's public counts. The content scores are Jev's judgement of each post, not X's own model, so treat them as a well-calibrated second opinion. The fixes link back to what the ranker is known to measure.
Last updated October 1, 2026