WEPR INSIGHTS

Major X Algorithm Update: Once the Highest Weight, Now Zero

The supplied seven sections, interaction-weight table, ten recommendations and twelve cautions, with twelve FAQs. A pinned official commit verifies the copy-link change from 20 to zero; FAQs distinguish time zones, public defaults, live enforcement, model scoring and cold-start scope.

X, formerly Twitter, has made another important adjustment to its recommendation algorithm.

On 13 November 2026, its public recommendation code showed a notable change:

The weight for copying a post link fell directly from 20 to zero.

Previously, this was the highest value among regular engagement signals—four times the ordinary reply weight.

Strategies that sought a recommendation advantage from users copying links therefore need reassessment.

Teams working on X, overseas brand growth, personal brands and international customer acquisition should examine the update carefully.

I. The latest X algorithm change: 20→0

Start with the central change.

📌 Copying a post link: 20 → 0

Previously, X predicted the likelihood that a user would copy a post link and assigned it a high weight.

That weight is now zero.

An important distinction:

The copy-link function has not been removed. Users can still copy a link to share content elsewhere.

The change means the predicted copy-link action no longer directly contributes through this particular scoring weight.

The code also contains a notable anti-spam detection item:

Copy Link Bait

It is intended to identify content that may encourage link copying artificially.

However, its presence can currently be confirmed in test code; whether it is fully enabled and how strictly it operates remain uncertain.

II. Current weights: which interactions still matter?

The public 13 November version contains the following principal interaction weights.

① Positive engagement signals

  • A qualifying reply involving mutual-follow users: 20
  • Ordinary reply: 5
  • Quote post: 5
  • Sharing through a direct message: 5
  • Following the author: 4
  • General sharing: 2.118
  • Repost: 1
  • Like: 0.708
  • Post click: 0.274
  • Opening an external link: 0.2
  • Copying the post link: 0
  • Merely opening a video: 0

The qualifying mutual-follow reply value of 20 combines a base reply weight of five with an additional 15, subject to specific conditions.

② Dwell-related signals

  • Content dwell weight: 0.05
  • Continuous dwell-time weight: 0.0075
  • Continuous dwell time after a click: 0.385

These indicate that users’ willingness to stop and read remains relevant.

However, 0.385 does not mean that reading for more than ten seconds earns a fixed 0.385 points.

③ Negative feedback signals

  • Selecting “Not interested”: -53.07
  • Blocking the author: -31.2
  • Muting the author: -58.8
  • Reporting content: -234

These weights underline the importance of relevance, user experience and compliance.

Importantly, the values apply to model-predicted action probabilities; they are not simple additions or deductions for actual counts of likes, replies or reports.

III. What does the update mean for smaller accounts?

1. Likes still matter, but raw like counts are not everything

In early November, X raised the like weight from 0.66 to 0.708.

This is a relatively high recent like weight in the public code.

For smaller accounts, positive feedback from genuine users after publishing remains useful to observe.

Practical suggestions:

✅ State the central point in the opening sentence.

✅ Give the conclusion first, then explain it.

✅ Use real cases, data and personal observations.

✅ Remove unnecessary introductions and increase information density.

2. Replies, quotes and private sharing deserve attention

Ordinary replies, quotes and DM sharing currently each have a weight of five.

Of particular note:

When the relevant conditions apply, a mutual-follow reply weight can reach 20.

Building genuine ongoing relationships may therefore be more valuable than pursuing follower numbers alone.

Practical suggestions:

✅ Answer genuine questions.

✅ Follow peers and potential customers with whom you have continuing exchanges.

✅ Publish professional viewpoints worth discussing.

✅ Contribute information people want to quote.

✅ Create content people want to send privately to friends.

The objective is not engagement bait, but a genuine reason to communicate.

3. Sustained attention deserves more study than clicks alone

X previously increased weights associated with continuous dwell time.

Meanwhile:

The weight for merely opening a video moved from 0.02 to zero.

The post-click weight also fell from 0.3 to 0.274.

The operating implication is straightforward:

Opening content does not necessarily mean it holds attention.

Content that sustains reading, viewing and substantive discussion deserves greater investment.

Suggestions:

✅ Present the most useful information early.

✅ Structure long posts with clear subheadings.

✅ Explain complex ideas through examples and data.

✅ Give viewers a clear reason to watch in a video’s opening seconds.

✅ Avoid an attractive headline followed by an empty body.

IV. How can smaller X accounts reach non-followers?

Several other mechanisms in the public code deserve attention.

① Certain new posts can qualify for recommendation support

For qualifying original posts by authors with no more than 50,000 followers, within two hours of publication and below 200 Home-feed impressions, a mechanism can lift a post toward approximately the sixteenth display position.

This is not a fixed traffic reward for every new post.

② Original content is better suited to seeking reach beyond followers

Out-of-network recommendations apply additional filtering and ranking conditions to replies and reposts.

Smaller accounts should therefore prioritise original contributions rather than relying indefinitely on reposting others.

③ Repeated appearances by one author receive diversity adjustments

The feed applies a diminishing adjustment to subsequent posts from the same author.

Publishing many posts close together therefore does not necessarily produce proportional exposure growth.

④ Freshness remains relevant

Some recommendation candidates face a 48-hour age filter.

Do not build a strategy around the assumption that posts from several days earlier will keep receiving Home recommendations.

V. Ten priorities for operating an X account

The latest public weights suggest the following operating priorities:

  • 1. Prioritise original content: publish your own viewpoints, experience, cases and industry observations.
  • 2. Earn genuine likes: gain approval through useful, information-rich content.
  • 3. Leave room for discussion: offer ideas people can meaningfully add to.
  • 4. Support sustained attention: remove filler and improve long-post structure and video pacing.
  • 5. Build mutual relationships: maintain contact with peers, potential customers and regular participants.
  • 6. Test publishing times: publish when the intended audience is active and observe early feedback.
  • 7. Keep a clear topic focus: develop consistent subjects and reduce audience–content mismatch.
  • 8. Create private-sharing value: publish checklists, methods, industry analysis and practical tutorials.
  • 9. Improve the profile: align the bio, pinned post and history so visitors understand the account.
  • 10. Monitor continuously: assess non-follower exposure, interaction quality, profile visits, follow conversion and available attention metrics.

VI. X practices to avoid where possible

Negative signals and governance mechanisms make the following behaviours worth particular attention.

❌ 1. Repeatedly asking users to copy a post link to influence recommendations.

❌ 2. Posting AI-generated template replies in bulk under large accounts.

❌ 3. Copying identical comments at scale to create mechanical interaction.

❌ 4. Repeatedly republishing old material with only minor wording changes.

❌ 5. Using exaggerated headlines without delivering the promised information.

❌ 6. Switching subjects frequently for traffic while ignoring the intended audience.

❌ 7. Mass-mentioning unrelated users or tagging strangers arbitrarily in images and videos.

❌ 8. Manufacturing popularity through fake likes, replies or other abnormal engagement.

❌ 9. Publishing large volumes of similar posts in a short period.

❌ 10. Overloading hashtags while neglecting the body’s relevance.

❌ 11. Using low-quality links and overt promotional scripts that increase spam risk.

❌ 12. Ignoring account status, content labels and visibility restrictions.

The public code contains mechanisms related to spam, low-quality material, engagement bait and some abnormal behaviour.

However, each detector’s activation, scope and actual effect need separate assessment; none of these observations means every listed action automatically triggers reduced reach.

VII. What is most important about the change?

From higher like and dwell-related weights in early November to the copy-link weight reaching zero on 13 November, the recent public changes suggest a direction:

Reduce the weight of some easily pursued mechanical signals while retaining signals associated with genuine discussion, content consumption and user relationships.

For brands and international teams, I believe the priority should not be inducing more interactions, but:

Making users want to read, reply, share and continue following you.

Traffic tactics may change quickly; content value and genuine relationships are stronger long-term growth assets.

#XAlgorithm #TwitterOperations #XOperations #OverseasSocialMedia #InternationalMarketing #InternationalGrowth #SocialMediaMarketing #PersonalBrand #ContentOperations #AIMarketing

Frequently Asked Questions

Which public version corresponds to the 13 November update?

This review pins official commit 590f3926e5bc33c50d9ece13953623e1e75ee595 and verifies the copy-link parameter changed from 20.0 to 0.0. Its timestamp is 13 November 2026 at 21:42:04 UTC, or 14 November at 05:42:04 in China. That is the public commit time, not an established deployment time. Do not mix the 2023 repository, third-party interpretations and different parameter dates.

Does zero copy-link weight make link sharing worthless?

No. This coefficient is zero in the reviewed public defaults; that does not remove sharing or erase other distribution, visit or business value. The repository says defaults are synced from production configuration, while experiments and settings may differ. No live runtime logs were inspected, so the default is not a universal claim about every user.

Can actual likes, replies and reports be added or subtracted using these numbers?

No. Predictions, continuous-time outputs, conditions and subsequent ranking are not a counter of realised actions. One report cannot be translated into a fixed number of cancelled likes, nor does 5 divided by 0.708 establish an exposure multiplier. The table is preserved from the supplied body; actual effects need account data and consistent measurement.

Does the mutual-follow reply value mean following back automatically adds 20 points?

No. The parameter file confirms a base reply value of five and a relationship-related increment of 15, but applicability depends on the candidate, relationship and scoring path. This review did not reproduce the complete live pipeline. Do not interpret every mutual reply as fixed added points or replace genuine relationships with reciprocal-engagement schemes.

Does the 0.385 dwell value award fixed points after ten seconds?

No. Continuous-dwell parameters need their model outputs, units and processing context; they are not a fixed bonus after a chosen number of seconds. Improve substance, structure and reasons to keep reading, without withholding answers, manufacturing empty dwell or guaranteeing expansion after a duration.

Does a Copy Link Bait test establish full production enforcement?

No. The body’s test-item description is preserved, but this review did not independently locate and reproduce its complete trigger, configuration and live enforcement path. A test name, label, model output and penalty are different layers. Without runtime evidence, do not assert full rollout, invent a threshold or equate ordinary copying with a violation.

Do 50,000 followers, two hours, 200 impressions and position 16 guarantee a boost?

No. The reviewed defaults contain these caps and slot values, and the code also selects among eligible candidates. Meeting some conditions does not guarantee selection, impressions or a final position. This is a particular cold-start window, not proof that a post becomes undiscoverable through search, reposts or other paths after two hours.

Does a 48-hour filter mean older posts can never reappear?

That inference is too broad. An age filter checks candidates in the path where it is applied; it does not establish one lifespan for every surface, user and post. Check the caller, source and dated configuration. Older content can remain a reference or search asset without a promise of continuing Home recommendations.

Can author-diversity adjustments establish a maximum daily posting count?

No. Repeated-author adjustments within a feed are not a daily account quota or a universal publishing schedule. Plan around quality, audience time zones and capacity, then observe actual intervals. Do not turn posting density, one time slot or a topic count into a confirmed follower-growth formula.

Does a large negative coefficient mean one report automatically restricts reach?

No. Prediction weights, submitted reports, moderation and account enforcement are different mechanisms. No account labels or penalties were inspected here. Check notices, visibility and available official transparency or appeal tools when needed. A reach decline alone does not prove enforcement, and coordinated false reporting is not acceptable.

Should adaptation involve follow exchanges, automated comments or repeated reposting?

No. X’s authenticity policy restricts coordinated engagement exchange, bulk irrelevant replies, duplicative content and repeated posting/deletion used for manipulation. Research or editing assistance does not make every automated action compliant. Follow current rules and earn feedback through real viewpoints, answers and independent value rather than manufacturing metrics.

How can an account assess whether the change affected it?

Keep comparable records of topic, format, timing, audience, paid promotion and observation windows, separating impressions, useful discussion, profile visits, follows, website visits and qualified enquiries. Compare multiple similar samples and record concurrent changes. A public parameter update generates a hypothesis, not causal proof from one or two posts or an acquisition guarantee.

Source:X/xAI: Pinned commit changing copy-link weight from 20 to zero

Source:X/xAI: Home Mixer parameters at the reviewed commit

Source:X/xAI: Experiments and configuration scope

Source:X/xAI: Cold-start selection and lift at the reviewed commit

Source:X: Authenticity and platform-manipulation policy