Why your China engagement rate keeps dropping

Your likes are down and your posting volume did not change. The audience did not leave. Through 2025 every major Chinese platform rewrote what it counts as engagement, and most brand reporting is still counting the old way. Platform mechanics move fast here, so we review this piece every quarter. Last checked July 2026.

If interaction on your WeChat, RedNote, Douyin and Weibo accounts slid over the past year, the number is real enough. It just is not measuring what you think it measures.

All four moved the same way last year. A like costs a user half a second. A comment costs a sentence, and a follow costs an actual decision about whether they want to hear from you again. The platforms started paying for the ones that cost something and quietly discounting the rest.

Almost every China dashboard still grades on the cheap signal. That is where your drop is coming from.

At a glance

What most brands still grade	What the platforms actually count in 2026
Likes per post	Comments at 4x a like, follows at 8x
Reach on day one	A climb that runs 7 to 15 days
Follower count	Search, which is now 65% of RedNote views
WeChat open rate	Shares, worth 20% of the health score
Video views	Completion, and the first three seconds
Posting more on Weibo	Posting inside a live trending window

Start with WeChat, because the drop shows there first

WeChat (微信) open rates have been sliding for years. Last year did nothing to interrupt that.

In the first quarter of 2025 the average open rate was 0.89% for subscription accounts and 1.32% for service accounts, around 0.2 percentage points below the same period a year earlier. Source: 壹伴 / Yiban, 2025

A lot of brands read that as a content problem. It is closer to a plumbing problem. Push stopped being the main door into a WeChat article a while back.

One finance account recorded 45.9% of its traffic arriving through social recommendation, passing subscription push for the first time. Source: 极致了数据 / Jizhile Data, 2025

Once you know that, the scoring makes sense. Several figures below come from 极致了数据 / Jizhile Data, a Hangzhou monitoring firm. We cross-check their mechanics against what we see in client back ends, and the direction holds even where the exact percentages move a point or two by category.

WeChat rates article health on open rate at 40% of the weight, interaction such as likes, Recommend taps (在看) and saves at 30%, share rate at 20%, and read-through at 10%. Source: 极致了数据 / Jizhile Data, 2025

Open rate is still the biggest single input, and it is the only one your follower list caps. That single fact reorders most of the WeChat plans we inherit. Everything else in that scoring decides whether a person outside the list ever sees the article at all. So the brand that publishes twice as often to lift its open rate is working on the ceiling while the door stays shut.

Douyin stretched the clock

Douyin (抖音) changed something in 2025 that broke a lot of monthly reports. The platform stopped reaching a verdict on a video within the first day.

New videos enter a cold start pool of 500 to 1,000 impressions. Completion has to reach 40% to move up a level, and the first three seconds carry 40% of the retention weight. The 2025 rebuild also extended the push cycle from roughly 24 hours to a 7 to 15 day climb, with daily interaction data adjusting the recommendation as it runs. Source: 极致了数据 / Jizhile Data, 2025

Which means the weekly report closes the file before the race is over. A video that looked dead on Tuesday can still be in testing the following Monday. We have watched agencies pull budget on day three, kill something mid-climb, and then write it up as a content failure.

Two more numbers from the same rebuild belong side by side, because together they explain why the safe corporate video underperforms the messy one.

Completion rates on hard advertising content fell 37% year on year in 2025. Source: 极致了数据 / Jizhile Data, 2025

And the algorithm now reads commercial outcome as a quality signal, not just a business one.

Videos converting at 3% pull roughly four times the organic traffic of videos converting at 1%. Source: 极致了数据 / Jizhile Data, 2025

On Douyin, selling well is not something that happens after you earn reach. It is part of how you earn it in the first place, which is an uncomfortable idea if your China plan treats commerce and content as two separate budgets.

On RedNote, likes are the least valuable thing you collect

RedNote (小红书) is the clearest case here, mostly because the scoring is documented well enough to run the numbers yourself.

The content experience score weights a follow at 8 points, a comment or share at 4 points, and a like or save at 1 point. Source: 极致了数据 / Jizhile Data, 2025

Run that against two notes. Nine hundred likes, five short comments, nobody new following. That scores around 920.

Comments of eight characters or more that describe a specific situation carry around three times the weight of a short comment. Notes saved often but never commented on can be flagged as low quality and throttled. More than 65% of RedNote traffic now arrives through search rather than the feed. Source: 极致了数据 / Jizhile Data, 2025

Now the quieter one. Two hundred and fifty likes, forty-five comments of the detailed kind, thirty new followers. That lands near 1,030. So a note showing under a third of the likes outscores the loud one, and the platform pushes it harder. Same brand, same budget.

You might reasonably ask why platforms still display like counts if likes barely move the scoring. Because likes are for you and for other users, as social proof. The scoring is a separate system, and it was never built to flatter anybody’s screenshot.

There is a second reason likes stopped predicting much here. Two thirds of the platform is not a feed at all.

A note that ranks for a query keeps pulling readers for months, and it collects almost nothing on the day it goes up. Judged on day-one engagement rate it looks like a failure, and on most accounts we audit it turns out to be among the more valuable things there.

RedNote sees about 39 million users a day showing clear intent to buy something, and 140 million daily instances of people actively looking for a product. Source: 虎嗅 / Huxiu, 2026

More brands are chasing the same eight hours

Three sets of platform mechanics in a row, so here is the part that has nothing to do with mechanics at all.

Some of the drop is just arithmetic.

Across 2025 RedNote led every platform on daily usage time growth at 12.1%, with Douyin second and Video Accounts third. Weibo fell 2.1%. On stickiness, Douyin’s daily-to-monthly user ratio reached 76.3%, Video Accounts 62.5%, and RedNote rose to 39.1% from 32.4% in 2023. Source: 知乎 / Zhihu, 2025

So attention is shifting between apps rather than growing.

The average Chinese user keeps around 29 apps and spends close to eight hours a day on mobile. Growth has shifted from finding new users to working the ones already there. Source: QuestMobile / 贵士移动, 2025

At the same time, brands kept publishing more into it.

On RedNote in 2025 the number of brands running paid placement in apparel grew more than 56% year on year, and average monthly likes on commercial beauty notes rose 76%. Source: 千瓜数据 / Qiangua, 2026

More brands, more posts, same eight hours. Average interaction per post has to fall, even when total interaction in the category is climbing. Your per-post number can drop in a year when your category got healthier.

The fix is boring. Report total interactions and interactions per post side by side, every month. If the rate is falling while the total is rising, you are winning and your dashboard is telling you the opposite.

Weibo (微博) is the exception in that table, and it is worth pausing on. It lost daily usage time last year, and the common response was to post more. That is backwards.

Weibo pays for participation in a live moment. Outside a trending topic a Weibo post is close to shouting in an empty room. Inside one it still travels further and faster than anything else in China.

The comment section is production work, not moderation

So much for the diagnosis. The rest of this is what we change on an account, and it starts somewhere most brands are not looking.

International brands tend to file comments under customer service. Somebody checks them twice a week and deletes the rude ones. If a complaint looks serious it gets an answer. On Chinese platforms that leaves the strongest lever sitting there untouched.

Douyin builds content labels in three layers. Core labels drawn from the video itself carry 50% of the weight, auxiliary labels pulled from user searches and the comment section carry 30%, and predicted labels carry 20%. Source: 极致了数据 / Jizhile Data, 2025

Almost a third of how Douyin decides who to show your video to gets built from words your audience typed underneath it. Leave that conversation alone and the platform labels you from whatever happened to show up.

What that looks like in practice, on a normal account:

• Reply inside the first two hours. On RedNote that window carries the most weight for pushing a note into the next traffic pool.
• Ask questions that cannot be answered in two characters. Short replies score, long specific ones score much harder.
• Seed the vocabulary you want to rank for. If you sell a serum for oily skin, the words “oily skin” need to appear in the comments, not only in your caption.
• Pin the comment that carries the product detail people keep asking about. It saves you from writing it forty separate times.
• Read the comments as research. The questions people repeat are your next three posts.

Our teams in Shanghai and Hong Kong sit on comment sections in Mandarin inside that first window, every publishing day. On a normal account it runs twenty to forty minutes per post, most of that in the first two hours. It is dull work and nobody on the team volunteers for it, but per hour spent it is the cheapest performance gain available on any of these platforms.

What to grade on each platform, and what to stop grading

Before you rebuild a content plan around a falling number, check that the number matches what the platform rewards. It usually does not, which means the content plan people are about to tear up was working the whole time.

Platform	Stop grading	Start grading
WeChat 微信	Open rate alone	Shares, Recommend taps, WeCom adds
RedNote 小红书	Likes per note	Long comments, search-driven views
Douyin 抖音	Day-one views	Three-second retention, day 7 to 15
Weibo 微博	Follower count	Topic entries, reposts while trending

The last column matters most on WeChat. A share, a Recommend tap and a WeCom add all survive the algorithm changing its mind next quarter, because each one moves a person a step closer to an audience you own rather than rent.

Our piece on WeCom covers how that private domain layer works, and why those numbers hold up when public reach gets expensive.

The next question is what to publish against it, and our piece on the content mix in China takes that one.

None of this requires more content. It requires grading what you already publish against the right column.

That distinction is also why our pricing page carries a fixed scope and a fixed price, published before you talk to anyone.

An agency paid by output has every reason to keep selling you volume in a year when volume is the thing that stopped working. Fixed scope removes that incentive. It matters more than it sounds, because the right answer is often to publish less and work each post harder.

Want your China numbers measured the way the platforms measure them?

We run WeChat, RedNote, Douyin and Weibo accounts from Shanghai and Hong Kong. Reporting is built on what each platform scores. You get comment depth and search-driven views, plus the full seven-day Douyin curve rather than a likes chart that flatters everybody and predicts nothing. Fixed scope, fixed price.