WEPR INSIGHTS

Xiaohongshu Recommendations in 2026, Finally Explained

The supplied nine-section guide to matching, CES, engagement, traffic pools, search and content reviews, with separate FAQs distinguishing practitioner models, unverified weights and governance rules. Original figures are not confirmed official algorithm parameters.

Why do some Xiaohongshu posts receive thousands of views within an hour and keep growing beyond 100,000 impressions, while others remain at 200–500 views?

Many people attribute the difference to account weight, traffic pools, or suspected distribution restrictions.

To genuinely understand Xiaohongshu traffic, however, you need to understand at least four things:

① How does the platform understand content and match it to users?

② Which engagement signals influence subsequent recommendations?

③ What kinds of content can keep receiving exposure?

④ What changed in the platform’s rules in 2026?

This article systematically examines Xiaohongshu recommendations through existing CES scoring, traffic-pool and content-quality models, together with publicly available platform governance rules from 2026.

I. The core logic of Xiaohongshu recommendations

Xiaohongshu uses personalised recommendations.

Put simply, the platform combines information about content and user interests to recommend posts to potentially interested people, then adjusts subsequent distribution in response to feedback.

A common operating framework breaks this into six steps:

1️⃣ Publishing and content recognition

After a post is published, the system processes it through content understanding, compliance review and duplicate-content detection. The platform uses MD5 checks and semantic recognition to detect copied content, although the complete technical combination and decision thresholds have not been disclosed.

2️⃣ Content review

Content must follow community rules and avoid prohibited marketing, false information, infringement and other violations.

3️⃣ Initial audience matching

The system uses the post’s content, user interests and historical behaviour to find potentially interested readers.

4️⃣ User feedback

Observe clicks, reading, likes, saves, comments, shares and follows after exposure.

5️⃣ Subsequent recommendations

If a post continues receiving positive feedback, it may be recommended to more relevant users.

6️⃣ Long-tail traffic

The recommendation cycle is usually seven days.

However, seven days is not a universal traffic cutoff. Some posts still gain new traffic through search and recommendations weeks or months after publication.

The core idea: content matching influences who sees a post first, while user feedback affects its subsequent spread.

II. CES scoring: which interactions are more valuable?

CES, or Community Engagement Score, is a commonly used engagement-analysis concept in the operations industry.

One widely circulated CES formula is:

  • Likes × 1 point
  • Saves × 1 point
  • Comments × 4 points
  • Shares × 4 points
  • Follows × 8 points

According to this practitioner model:

Follows > Comments/Shares > Likes/Saves

This suggests a direction worth investigating:

A like and a willingness to comment, share or follow do not necessarily express the same level of involvement.

Important qualifications:

1. Engagement patterns differ between content types.

For example, educational posts may attract more saves, opinion posts may generate comments, and video analysis also needs viewing metrics.

2. Engagement quality cannot be reduced to engagement quantity.

Genuine discussion that contributes information is more useful to examine than large volumes of repetitive or meaningless comments.

Do not deliberately solicit follows, manufacture controversy or inflate comments merely to pursue a CES score.

III. Four key factors affecting recommendation weight

The original framework divides recommendation weight into four dimensions:

1️⃣ Content quality: 40%

Consider these indicators:

  • Visual clarity, such as 4K video
  • Original writing
  • Information density
  • Practical reference value

There are two common misconceptions:

First, 4K resolution does not automatically confer greater recommendation weight.

Clarity can improve viewing, but visual quality cannot replace content quality.

Second, changing a few words in AI-assisted content does not automatically make it high-quality original work.

Human improvement of AI-generated content is a useful direction, but the platform has not disclosed a universal required editing percentage.

2️⃣ User engagement: 30%

Focus on:

  • Video completion
  • Reading time for image-and-text posts
  • Likes, saves, comments and shares
  • Engagement quality and subsequent follows

Note:

The first three seconds of a video are important, and image-and-text reading time above 30 seconds is worth examining.

These can serve as creative optimisation directions.

For example, establish a video’s value within its first three seconds and show the user’s problem on the first screen of a text-and-image post, reducing unnecessary introductions.

3️⃣ Account performance: 20%

Observe:

  • Topic focus
  • Account activity
  • Content authenticity
  • Historical compliance

Consider focusing on one or two fields and publishing three to five posts per week.

This can be a planning reference during an account’s early stage.

However, posting across fields does not necessarily make an account a marketing account, and no publishing frequency is optimal for every account.

4️⃣ Real-time popularity: 10%

This mainly includes:

  • Topic popularity
  • Timeliness
  • Relevance to holidays and events

For example, topics linked to seasonal needs, industry developments and current user questions may better match present content-consumption needs.

IV. Traffic pools: how do posts gradually gain exposure?

Xiaohongshu follows a layered, personalised recommendation approach, often described in the operations industry as a stepped traffic pool.

Stage one: initial exposure

Some practitioner accounts mention roughly 300–400 impressions; others mention 500–1,000.

Stage two: performance testing

Circulated progression benchmarks include:

  • Click-through rate, CTR > 10%
  • Engagement rate, ER > 3%

Stage three: expanded recommendations

Content that continues receiving positive feedback may reach more people, and exceptional posts can reach millions of impressions.

An important distinction:

300–400 impressions, CTR above 10% and ER above 3% are not publicly confirmed universal progression thresholds.

Actual performance may vary significantly by account, category, audience and recommendation context.

In practice, I would focus on three areas:

① Click-through rate

Do the cover and headline attract the intended audience?

② Reading and engagement

After opening the post, do users continue reading, save, comment or share?

③ Growth curve

Does traffic continue growing one hour, 24 hours and 72 hours after publication?

Also observe changes in search versus recommendation traffic rather than cumulative views alone.

V. Content quality: which posts are more likely to be recommended?

Alongside engagement, Xiaohongshu’s content-quality evaluation is important.

The platform has mentioned content-quality and sincerity scores in public materials, emphasising quality and genuine sharing.

In practical content creation, focus on three directions.

1️⃣ Information gain

Offer new perspectives, experience or solutions.

For example, a post about SEO should not only define the term; it can also provide:

Keyword research, headline planning, search-result analysis, practical examples and performance reviews.

2️⃣ Subject expertise

Content in a specialised field should demonstrate genuine knowledge and judgement.

For Xiaohongshu search optimisation, for example, you could consistently address:

Autocomplete terms, blue search terms, navigation terms, long-tail keywords, search positions and GEO.

3️⃣ Emotion and resonance

Emotionally valuable content can encourage discussion, but this does not require manufacturing conflict.

It is more useful to examine whether users identify with the content and whether it supports genuine conversation.

Different content types can use different structures:

  • Tutorials: steps + tools + results
  • Reviews: comparison criteria + measured data + conclusions
  • Personal stories: a specific situation + personal experience + a viewpoint

VI. Four algorithm and platform-rule developments to watch in 2026

1️⃣ Greater emphasis on authenticity

On 19 January 2026, Xiaohongshu introduced Community Convention 2.0.

It emphasises genuine sharing, friendly interaction and orderly commercial activity, opposing invented personas, false marketing and malicious competition.

Accounts therefore need to care about truthful content and commercial identities, not only engagement figures.

2️⃣ Stronger AI-content governance

On 27 April 2026, Xiaohongshu published its AI governance position.

It encourages AI-assisted quality creation while opposing:

  • Fully AI-managed accounts
  • AI-fabricated experiences, screenshots and personas
  • Mass-produced homogeneous AI content
  • AI infringement and false information

Content involving AI should be proactively labelled in accordance with the rules.

What matters is not the percentage changed but whether content is truthful, original and valuable, and whether AI involvement is labelled as required.

3️⃣ Search relevance deserves separate attention

Xiaohongshu traffic is not limited to the recommendation feed.

Recommendations primarily concern user interests and feedback; search also involves keywords and search intent.

Public 2026 guidance on Xiaohongshu search advertising identifies matching between search terms and content as an important consideration.

This guidance concerns search advertising and cannot simply be treated as the organic-search ranking formula.

Content operations can nevertheless borrow the principle:

Use the main term in a relevant headline, answer a specific question in the body, and develop content around long-tail queries.

For example:

“Xiaohongshu SEO” is a core term.

“How do Xiaohongshu autocomplete terms work?” and “How can Xiaohongshu search positions improve?” address more specific questions.

Help the platform understand the topic, but also help people find the content through their actual questions.

4️⃣ Continued governance of false marketing and comments

On 6 November 2026, Xiaohongshu announced requirements targeting false and low-quality marketing.

The focus includes:

  • Pretending to offer genuine product experiences
  • Mass-producing homogeneous marketing posts
  • Coordinating multiple accounts to create false comments
  • Inflating positive comments and manipulating reputation

Mass-posting, fabricated ordinary-user experiences and artificial engagement are not worthwhile long-term growth methods.

In 2026, content performance and authenticity need to be considered together rather than treating recommendations as an engagement-points game.

VII. Practical strategies for improving post exposure

Consider the following areas.

1️⃣ Cover design

Use a three-colour principle and an information anchor: control the main colours and highlight the central value with clear text.

2️⃣ Headline optimisation

Build the headline around a user problem and a solution.

For example:

“No traffic on your Xiaohongshu posts? Check these five indicators first.”

3️⃣ Body layout

Use a modular structure, manageable paragraphs and standalone key points, reducing unnecessary introductions.

4️⃣ Publishing times

Suggested starting points:

Weekdays: 07:00–09:00, 12:00–14:00 and 19:00–21:00.

Weekends: 10:00–12:00 and 15:00–17:00.

These are testing starting points. Decide based on audience activity and actual publishing data.

5️⃣ Topic tags

One suggested combination:

One broad tag + two or three precise tags + one event tag + one individual tag.

In practice, prioritise relevance to the body rather than mechanically meeting a tag count.

6️⃣ Engagement maintenance

Respond to genuine questions and encourage substantive discussion instead of manufacturing meaningless interaction.

VIII. How can data diagnose a post’s problems?

Consider the following indicators.

① Click-through rate, CTR

Above 10% is suggested as a reference.

Interpret it in relation to the source of impressions, format and account history.

② Engagement rate, ER

A common simplified formula:

(Likes + saves + comments) ÷ views × 100%

3% is used as a reference line and above 8% as a reference for high-quality content.

However, different categories cannot use identical standards without qualification.

③ Early exits

The original framework suggests checking the first three seconds when the exit rate exceeds 70%.

The diagnostic idea can be retained, but 70% is not a publicly confirmed universal restriction threshold.

④ Save and comment patterns

More saves may indicate reference value.

More comments may indicate discussion value.

Comments substantially exceeding likes and saves do not automatically establish better quality; controversy and conflict also need to be considered.

⑤ Follow conversion

If a post receives many views but almost no new follows, consider whether it fits the account’s long-term positioning.

⑥ Sustained traffic

Look beyond the first 24–72 hours and examine search and recommendation performance over seven and 30 days.

This is particularly relevant to SEO-oriented posts.

IX. Conclusion: what does the Xiaohongshu algorithm consider?

The recommendation logic can be summarised through five ideas:

1. Content matching: help the platform find people suited to the content.

2. Qualified clicks: attract the intended audience with a clear cover and headline.

3. Reading value: provide genuine information once users open the post.

4. Meaningful engagement: collect feedback through authentic comments, saves, shares and follows.

5. Sustained traffic: combine recommendation-oriented content and search-keyword planning for long-term value.

A common mistake is treating CES formulas, traffic-pool thresholds and recommendation weights as absolute, unchanging algorithm rules.

The enduring priorities are content quality, matching user needs, genuine interaction and search coverage.

Algorithms will change, but content that consistently solves user problems has a better chance of earning stable traffic.

Frequently Asked Questions

Is the CES 1/1/4/4/8 formula Xiaohongshu’s official algorithm?

This review did not establish public evidence that these coefficients form a universal official ranking formula for 2026. They can be treated as a custom review model, not a basis for promising reach, comparing every format or claiming one follow always equals eight likes. Preserving the supplied body is not official validation of the formula.

Have the 40/30/20/10 percentage weights been confirmed?

No. This review did not verify an official confirmation of these four fixed percentages. The dimensions can organise a checklist, but the figures should not be presented as model parameters. Terms such as quality or sincerity scores also need a specific product, document version and scope; their names do not establish a platform-wide organic ranking formula.

Can MD5 and semantic checks be confirmed as the platform’s copying detector?

This review did not obtain official technical documentation sufficient to confirm the complete 2026 detection pipeline. MD5 is a file-checking method, but that fact alone does not establish how this platform detects copying. Do not alter hashes, wording or images to bypass review; check originality, permissions and attribution instead.

Does a post stop receiving recommendations after seven days?

No such conclusion is justified. This review did not verify an official seven-day cycle applying to all posts. One hour, 24 hours, 72 hours, seven days and 30 days can be observation windows for separate search and recommendation sources, not an automatic expiry or redistribution schedule.

Are 300–400 impressions, 10% CTR and 3% ER traffic-pool thresholds?

Do not treat them as universal thresholds. The same applies to 500–1,000 impressions, 8% ER, 70% exits and 30-second reading time. Compare the same account, format, source and observation window first. A post with 200–500 views alone does not prove a penalty or a fixed distribution cap.

How should CTR, ER, completion and dwell time be compared?

Record the available report definition and permissions: whether the denominator is impressions, views or unique users, and whether shares, follows or repeated actions enter the numerator. The body’s ER is a simplified custom definition, not necessarily the platform report. Separate formats and paid versus organic results, and mark unavailable data rather than deriving growth from incompatible measures.

Can 4K video or manually editing AI content guarantee more recommendations?

No. Resolution and editing may improve the experience but cannot replace genuine information, original judgement or asset permissions. This review established no fixed required editing percentage. Follow the current applicable rules and publishing controls for AI labelling, and do not present invented experiences or synthetic reviews as real product use.

Must an account cover only one or two fields and publish three to five times a week?

No. These are planning suggestions, not verified eligibility rules or a universally optimal schedule. Plan around audience, materials and capacity, then record feedback by topic. Changing fields or pausing publication alone does not establish a marketing-account classification; assess any actual notice and rule.

Must the publishing times, three-colour principle and tag combination be followed?

No. They are starting points for readability and publishing tests, not official algorithm requirements. Consider audience time zone, actual activity and format, change few variables at a time, and compare equivalent windows for useful consumption and business feedback. Avoid irrelevant tags or artificial interaction.

Can search-ad relevance scores prove organic recommendations or search positions?

Not directly. Spotlight’s relevance precheck supports keyword-to-post matching for search advertising, not an organic recommendation scorecard. The principle of answering the query can inform writing, but record ad impressions, organic search and feed traffic separately. Do not convert an ad score or paid growth into proof of organic ranking improvement.

Do governance announcements establish changes in recommendation weights in 2026?

No. Governance rules and ranking parameters are different. The official professional-account notice reviewed here is dated 30 October 2026 and specifies 4 November as its effective date. That does not independently verify the separate 6 November announcement referenced in the body; the exact version remains unresolved. It informs compliance checks, not an inferred increase in any engagement coefficient.

What is a sensible diagnostic sequence for persistently low traffic?

Check public visibility, review status and notices first, then topic–audience fit, cover and headline, opening delivery, reading and genuine engagement, followed by search sources and qualified enquiries. Preserve versions, times and paid/organic definitions. Without evidence, do not attribute fluctuation to suppression, account weight or a universal algorithm change, or respond with fake engagement or repeated deletion and reposting.

Source:Xiaohongshu Spotlight: Search tools (search advertising)

Source:Xiaohongshu: Professional-account compliance guide

Source:Xiaohongshu: Governance announcements and violation notices