# X Algorithm Update: What Changed on November 5, 2026?

By Pawnie (@pawnie_ on X). Growth on X. Published 2026-11-07. Canonical page: https://www.w3match.com/blog/x-algorithm-update-november-2026

X changed seven public For You ranking defaults. Learn what the numbers mean, what they do not prove, and how creators can respond.

X's public For You code changed on November 5\. Its Home Mixer defaults now put a higher coefficient on predicted likes, shares and time spent, while the coefficient for predicted clicks fell. The coefficient for a video open dropped to zero.

That sounds like a straightforward shift toward attention. But there is a catch: **these numbers are not points awarded for each like, share or second watched.** They weight predictions about how a particular viewer may respond to a particular post. They also represent configurable defaults, not a guarantee that every user sees the same setup.

Here's what changed, how to read it and what creators can actually do with the information.

## The November 5 changes

We compared the [November 5 code snapshot](https://github.com/xai-org/x-algorithm/blob/021afa44/home-mixer/params/param.rs) with [the previous version](https://github.com/xai-org/x-algorithm/blob/b5089266/home-mixer/params/param.rs). Seven Home Mixer scoring-weight defaults changed:

For the wider mechanics, see our earlier [X Algorithm 2026 guide](https://www.w3match.com/blog/x-algorithm-2026-for-you-guide). A [later November 7 snapshot](https://github.com/xai-org/x-algorithm/blob/3b6919f87532e57063bb3f1e85c2f509323b513e/home-mixer/params/param.rs) still shows the same like, share, click and video-open defaults. The comparison below isolates the November 5 change.

| Signal in the scoring model     | Before  | After  |
| ------------------------------- | ------- | ------ |
| Predicted like                  | 0.66    | 0.708  |
| Predicted continuous dwell time | 0.006   | 0.0075 |
| Predicted click                 | 0.3     | 0.274  |
| Predicted video open            | 0.02    | 0      |
| Predicted share                 | 2.0     | 2.118  |
| Predicted click dwell time      | 0.4     | 0.385  |
| Predicted “Not interested”      | −52.272 | −53.07 |

The continuous dwell coefficient increased by 25%. That does **not** mean posts get 25% more reach, or that every additional second has the same effect across viewers. Other predictions, filters and ranking stages also matter.

The video-open coefficient reaching zero means this particular term contributes zero under that default. It does not mean video no longer matters, or that other video-related predictions and viewing behavior have stopped influencing recommendations.

The [same commit](https://github.com/xai-org/x-algorithm/commit/021afa44) also switched on the default for `EnableHomeExcursionScores`. Its name alone is not enough to establish what a creator should do or how much reach changes.

## Why you cannot calculate a post's reach from the weights

X describes its scoring weights as multipliers on **predicted probabilities or continuous values**, not raw action counts. The model estimates how likely a specific viewer is to act, or how much time they may spend, and the system combines those estimates with weights and other adjustments. [X explains the distinction in its README.](https://github.com/xai-org/x-algorithm/blob/021afa44/README.md)

For a simplified illustration, if a model estimated a 10% chance of a like, multiplying `0.10 × 0.708` gives a contribution of `0.0708` for that term. This is an illustration, not an actual score for your post. The real system has other terms and stages.

That distinction rules out a common claim: you cannot divide the report coefficient by the like coefficient and conclude that one report cancels a fixed number of likes. The same goes for claims that one copied link is worth a fixed number of likes or impressions.

Also, X says configuration values can vary through experiments. Its repository syncs defaults to primary production values, but the public snapshot cannot tell you the exact configuration for every viewer or expose every part of the production system. [Source: X's configuration and limitations notes](https://github.com/xai-org/x-algorithm/blob/021afa44/README.md#experiments-and-configuration).

## Shares, replies and negative feedback: context matters

In the November 5 snapshot, the coefficients for predicted replies, quotes and DM shares were each 5; predicted share via copied link was 20; and predicted follow-author was 4\. A separate bidirectional-follow reply boost was 15\. The last one is tied to a mutual follow relationship, not every back-and-forth between two accounts. [Source: Home Mixer parameters](https://github.com/xai-org/x-algorithm/blob/021afa44/home-mixer/params/param.rs).

Those numbers do not establish a shortcut. Asking people to copy a link, coordinating empty engagement, or trying to manufacture a conversation is not a proven way to gain distribution. Make something a reader has a reason to pass to another person: a useful comparison, a clear finding, or a warning with evidence.

Negative feedback also appears in the public configuration: the report and mute-author coefficients were −234 and −58.8 respectively. They apply to predictions in a personalized system. They are not a universal penalty you can calculate from a post's visible counts.

## Why the October updates matter too

Two earlier changes make the November numbers more interesting:

- On **October 8**, X said its Phoenix ranking model had improved at predicting how much time a viewer would spend on a post.
- On **October 13**, X said Phoenix retrieval was also trained to surface posts that people read all the way through. [Source: X's update notes](https://github.com/xai-org/x-algorithm/blob/021afa44/README.md#notable-updates).

Retrieval and ranking do different jobs. Retrieval finds potential posts for the feed; ranking helps order candidates for a particular person. A post must be relevant to the reader before any discussion of its ranking score matters.

The direction supports a sensible editorial hypothesis: make the payoff clear, then give the reader a reason to keep reading. It does not prove that longer posts win. A concise chart explanation may be more useful than a padded thread.

## How For You gets from a post to a feed

X's public overview describes a multi-stage system:

1. **Retrieve candidates.** Thunder supplies posts from accounts a person follows. Phoenix retrieval and SimClusters help find posts outside their network.
2. **Filter candidates.** Eligibility, safety and visibility rules can remove or limit posts in different contexts.
3. **Predict and rank.** Models estimate possible viewer responses; scoring combines predictions with configurable weights.
4. **Shape the feed.** Further adjustments, including author diversity, affect the final selection.

This is a simplified view of the public For You pipeline. It should not be applied automatically to Search, notifications, replies or every other X surface. Some rules and prompts are not public. [Source: X repository](https://github.com/xai-org/x-algorithm/blob/021afa44/README.md).

## What Web3 creators can test

The code tells you what the system is designed to consider. It does not tell you which post format will work for your audience. Use it to generate tests:

**Make the subject clear immediately.** If you are explaining a protocol update, tell readers what changed and who it affects before adding background. Clarity helps people decide whether the post is for them; it is not a guaranteed ranking factor.

**Earn attention with the next line.** Give readers a reason to continue: the affected users, the trade-off, a chart or a specific example. Avoid adding length solely to increase time spent.

**Create something worth forwarding.** A short breakdown of a governance proposal or a security change may be useful in a team chat. Test whether readers share it because it helps, not because you told them a copied link is worth more.

**Add substance to conversations.** When replying to a founder or researcher, contribute a data point, a question grounded in their post, or a considered counterargument. The public defaults do not prove that every short reply or every AI-assisted reply receives a specific penalty.

**Compare similar posts over time.** Track reach, meaningful responses, shares and any available reading or video metrics. Compare similar topics, formats and audience windows across multiple posts. Timing, news and audience composition can still explain differences, so treat the results as observations rather than proof of causation.

If your visibility changes unexpectedly, X's [Under the Hood tool](https://github.com/xai-org/x-algorithm/blob/021afa44/README.md#under-the-hood-label-transparency-tool) is intended to show aggregate information about certain visibility-impacting labels. It cannot reveal every reason a post performed differently.

## Is there a two-hour or 200-view distribution gate?

Creators often repeat a rule that a post must pass a test in its first two hours or first 200 views before X expands distribution. We did not find a universal threshold in the primary sources reviewed for this article. Early engagement may be useful feedback, but this particular cutoff should not be treated as established fact.

There is a narrower source for similar numbers: X's public code describes a configurable cold-start mechanism for some original posts from smaller accounts, with conditions including age and Home impressions. It selects an eligible candidate within a feed request; it is not evidence of a universal pass-or-fail gate. Our [earlier guide](https://www.w3match.com/blog/x-algorithm-2026-for-you-guide) explains the distinction.

## The practical takeaway

The November update changes several published scoring defaults, while the October notes show improvements in attention-related retrieval and prediction. Neither reveals a formula for guaranteed reach.

For creators, the useful next step is to publish clear, relevant work, give readers a real reason to finish or share it, and compare results over time. Use the source code to ask better questions—not to turn every interaction into a points game.

**Sources:** [November 5 commit](https://github.com/xai-org/x-algorithm/commit/021afa44), [previous parameters](https://github.com/xai-org/x-algorithm/blob/b5089266/home-mixer/params/param.rs), [updated parameters](https://github.com/xai-org/x-algorithm/blob/021afa44/home-mixer/params/param.rs), [X repository and update notes](https://github.com/xai-org/x-algorithm/blob/021afa44/README.md).