Open-source algorithm review

X Algorithm Analysis

Three documents examine the open-source code behind X’s For You feed: what it says about reach and visibility, why accounts with similar audiences get very different results, and what the findings mean in practice.

All three are based on the November 19, 2026 state of the algorithm (commit b551517) and are reproduced from the mindprison-data repository.

  • 3 documents
  • 63,725 words
  • Reviewed Nov 19, 2026

What's new

November 19, 2026

The latest review covers commit b551517. No ranking weight moved and no filter was added or removed. Three changes are worth an ordinary account's attention.

Ads may move

A new ad-slotting default can shift where ads appear. Organic posts move around them, so small impression changes in this window are not a scoring change.

AI video is now tagged

X can label declared or detected AI-generated video and is building a separate retrieval pool for it. It does not affect ranking yet.

Post labels are finer

NSFW and soft-NSFW labels can now be applied directly to a post, with more review for image-heavy posts. Share-bait is still treated as spam; click- and reply-bait get their own labels.

Read the full November 19 summary

FAQ

Five questions the analysis answers

Short answers from the Account Analysis, with links to the full explanations.

Why can two accounts with the same followers get completely different reach?

Reach is decided one viewer at a time, not by follower count. The model learns what usually happens when an account's posts are shown, so an account that earns replies, shares and reading time keeps getting shown while a similar-sized account that doesn't can go quiet.

Read the full answer

Do blue checks help my posts get seen?

No. Verification is not an input to feed ranking. It can give an account more credibility in enforcement decisions, but it does not add reach.

Read the full answer

Does a repost from a big account guarantee reach?

No. A repost only adds your post as a candidate in more feeds. Each viewer is still scored separately, reposts are discounted, duplicates are removed, and your original post's reputation travels with it.

Read the full answer

Does replying to small accounts hurt me?

There is no rule that punishes it by name, but small threads get automated spam screening and produce weaker training signals. At volume, that can drag on how the model sees your replies.

Read the full answer

Can liking posts hurt my own reach?

No. Likes are a positive viewer signal and do not touch your reputation as an author. They do train what you are shown, so a steady diet of one kind of post can narrow your feed.

Read the full answer

Short version

The one-page cheat sheet

The rules that matter, distilled from the User Guide.

Do

  1. Design every post around one question: will someone reply to this, or copy-paste the link to send it to someone? (reply 5.0, copy-link share 20.0 — vs like 0.708; note the weights scale predicted probabilities, not counts — a DM share isn’t “7 likes,” Part 1). Hold the click: sustained post-click reading now pays 0.385 click-dwell.
  2. Post originals, not replies/RTs — especially if you’re small (cold-start slot-15 boost requires it; mutual-follow +15 requires it; replies/RTs are discounted). Post them inside the 2-hour launch window — the boost expires fast now.
  3. Make your best content a ≥10-second video (evergreen pools: 5 years of eligibility vs 48 hours for text — but since Nov, no video-open scoring bonus; and Articles get no loophole: they’re ordinary 48-hour posts, now with an unpublished-article VF drop)
  4. Win the first day: 1 fav makes the post eligible for discoverability (indexed, not guaranteed served); 32 favs in a day adds the hot pool — and the first fav must come from your seed audience, because base-pool membership doesn’t open until it does (Part 2.11). Replies and reposts help through predictions either way.. New: fav’d replies get their own index — good replies are now retrievable, not just ranked.
  5. Quote with your own words instead of reposting (repost credit goes to the original author)
  6. Reply into big threads (target/root > 250K followers): ranked, not spam-screened, bigger audiences, positive training signal. Don’t stack same-author replies in one thread (Oct demotes that to 0.0).
  7. Like deliberately; never like-then-unfollow; don’t like bait (your likes train your own feed)
  8. Pursue follow-backs from accounts you follow (+15 reply weight on your originals, plus a direct model feature)
  9. One strong post per story — the diversity step zeroes near-duplicates
  10. If you’re stuck: sustained positive outcomes at volume, via the side door (replies in big conversations); expect a long invisible phase before anything visibly changes
  11. Know your class and its levers: ≤50K → cold-start launch boost (2h/200 views); mid-size → follow-backs + big-thread replies; everyone → win the first day (Parts 2.1, 2.3, 4)

Don't

  1. Pay for verification for reach (zero delivery effect; it’s only an enforcement shield)
  2. Obsess over follower count, posting frequency, engagement rate, or profile completeness (none are scored)
  3. Habitually reply into small dead-end threads (spam-screened + full-weight “no action” examples attributed to your author row)
  4. Generate negative feedback from your own audience (report −234 / mute −58.8 / not-interested −53.07 / block −31.2 — and net-negative posts are compressed into a floor band)
  5. Expect a big-account repost to save you (scored per viewer, carrying your original’s reputation; engagement on it credits the original author)
  6. Worry about brigades or drive direct-link traffic expecting ranking impact (per-viewer predictions; only Home-Timeline-served impressions count)
  7. Post sub-10-second videos expecting video treatment (no video signal, no VQV score — make it ≥10 s or a photo — Part 5)
  8. Assume the enforcement shield is earned — it tracks size and verification, so stay label-clean (Part 5)
  9. Repost your own failed posts for a “second chance” (same post re-scored with the original’s age, deboosted, seen-filtered — you re-run the failure; post a new original or quote yourself instead — Part 2.4)

Read the full User Guide