How to read China social media reporting metrics and spot the padding

Your monthly deck shows big reach, steady follower growth and a healthy engagement rate. Sales haven’t moved. Before you blame the content, check the China social media reporting metrics themselves. Each platform defines a “read”, an “interaction” and a “conversion” its own way, and a report can pick whichever version looks best.

That isn’t always your agency’s fault. Some platforms count loosely, and Xiaohongshu’s creator tool couldn’t show paid boosts apart until January. But a report should tell you which version it used.

Metric in the deck	How it gets padded	What to ask instead
Reads or views	Times, not people; repeat views counted	Unique readers, split by source
Followers	Gross new follows; bought or inactive accounts	Net change: new follows minus unfollows
Engagement rate	The friendliest formula and denominator	The formula, written out and kept fixed
Campaign sales	The longest window; paid and natural mixed	The window, the attribution rule, the split
Best posts	A few screenshots instead of the full list	Every post in the period, exported

The rules below come from the platforms’ own help pages. The cases come from courts, regulators and dated Chinese press. All were checked on October 28, 2026.

The five metrics that mean almost nothing on their own

None of these numbers is fake. They answer questions you didn’t ask.

Total impressions tell you a post was shown, not read. A follower total can include accounts that went quiet long ago (the follower section below shows how many). Likes and plays can be bought by the thousand. Topic or hashtag views count everyone’s posts on that topic, not only yours, which is how a small brand can “reach” millions in a deck while its own videos reach a fraction of that. And an engagement rate with no formula next to it can be almost any number you like.

Treat them as background.

Impressions, reach and views: why the labels shift

Start with WeChat (微信). Its Official Account back end reports article reads two ways, and they are not the same number.

WeChat counts “reading people” without repeat clicks and “reads” with them. Both include readers who don’t follow the account. Its conversion rate is reading people divided by people the post was delivered to. Source: Tencent customer service (腾讯客服), confirmed October 2026. https://kf.qq.com/faq/161222VVVjAr161222EJbiMj.html

So a deck that quotes “reads” may be counting the same fan three times. It may also count strangers who opened a forwarded link. Neither is wrong. But only the people count, set against deliveries, shows how your followers responded. One more detail cuts in your agency’s favor: followers who mute the account aren’t counted as delivered, so a shrinking delivery number isn’t always a content failure.

Weibo (微博) goes further. Its read count behaves more like an impression.

A Weibo post’s read count is the number of times users saw it. For a single post, abnormal and repeat reads have not been removed. Source: Weibo customer service (微博客服), January 2026. https://kefu.weibo.com/faqdetail?id=14691

That is the platform’s rule, not your agency’s doing. But an English deck that calls the number “reads” should say what it measures.

On Xiaohongshu (小红书), the ad tool and the app don’t agree with each other.

Juguang, Xiaohongshu’s ad platform, counts a withdrawn like or repeat clicks from one account again, where the app filters them out. If an ad viewer then interacts with another note on the brand’s profile, Juguang credits that to the ad. Source: Xiaohongshu Juguang help center (小红书聚光), January 2026. https://ad.xiaohongshu.com/api/light/help/doc/content?shortcutId=c0b80b06af593eff0d0f2680b9df6701

Pugongying, the platform’s creator marketplace, says its figures and the app’s figures use different definitions and can’t be compared. For sponsored notes, Pugongying’s data is the one to use. Source: Xiaohongshu Pugongying help center (小红书蒲公英帮助中心), April 2026. https://pgy.xiaohongshu.com/api/pgy/help/doc?shortcutId=fcd9c98c29152a7826a676350f63bef9&role=4

If one chart mixes app screenshots with Pugongying exports, it is measuring with two rulers.

Douyin (抖音) is vaguer still. Its creator marketplace, Xingtu, builds its rates on plays, and we found no Douyin page that defines what a single play is. So ask your agency what it counted.

Xingtu defines completion rate as full plays divided by plays. Source: Ocean Engine (巨量引擎), Xingtu help center, June 2026. https://www.xingtu.cn/help-center/demander/115182

The follower growth trick: bought, seeded or earned

Follower growth in a deck is often gross growth. WeChat (微信) defines the number that counts in its own data interface.

Net new followers equal new follows minus unfollows. Source: WeChat developer documentation (微信开发者文档), confirmed October 2026. https://developers.weixin.qq.com/doc/subscription/api/wedata/user/api_getusersummary.html

Then ask who the followers are. Earned followers came for the content. A giveaway or a paid push brings in seeded ones, who are real but often not interested. And then there are bought accounts, sold by the thousand. The platforms keep finding them.

Xiaohongshu (小红书) cleared 170 million fake follows between August 22 and September 20, and handled 76,000 accounts for interaction cheating. Source: Xiaohongshu Pugongying help center (小红书蒲公英帮助中心), September 2026. https://pgy.xiaohongshu.com/api/pgy/help/doc?shortcutId=cdf229e526f64dcf3375462b89eb63e8&role=4

Weibo (微博) found 7,697 accounts that had bought followers, across 36 content categories, and named the 10 worst. Source: Beijing Daily (北京日报), quoting Weibo’s administrator account, January 2026. https://finance.sina.cn/2026-01-30/detail-inhkacxr0606435.d.html

The sellers are cheap, which explains a lot.

One ring broken up by police in Liangshan, Sichuan, advertised 100,000 video plays for 98 yuan and 10,000 likes for 248 yuan. Source: CCTV.cn (央视网), February 2025. https://news.cctv.cn/2025/02/11/ARTItQYvuvUQRKmqz2bMjWFY250211.shtml

Inactive accounts are the quieter problem. The last time the ad-measurement firm Miaozhen published the share, it was more than half.

The average KOL account had 53.2% invalid followers in 2022, a figure that counts inactive accounts as well as bots. Source: Miaozhen (秒针系统), March 2023. https://www.miaozhen.com/all-reports/marketingeffiency/806.html

So look at the curve next to the posting calendar. A jump in one week with no post, campaign or giveaway behind it needs an explanation.

Engagement rate: which denominator did they use

Take a made-up example. A Xiaohongshu (小红书) note has 2,000 readers and 100 likes, saves and comments, on an account with 50,000 followers. Over readers, that’s 5%. Over followers, it’s 0.2%. Same note.

The platforms don’t agree on the top of the fraction either.

Tool	Interactions it counts	Shares included
Pugongying (Xiaohongshu)	Likes, saves, comments	No
Juguang (Xiaohongshu ads)	Likes, saves, follows, comments, shares	Yes
Xingtu (Douyin)	Likes, comments, shares	Yes

Pugongying adds likes, saves and comments, and divides by reads. Shares are left out. Source: Xiaohongshu Pugongying help center (小红书蒲公英帮助中心), April 2026. https://pgy.xiaohongshu.com/api/pgy/help/doc?shortcutId=fcd9c98c29152a7826a676350f63bef9&role=4

Juguang’s interaction count is the sum of five actions: likes, saves, follows, comments and shares. Source: Xiaohongshu Juguang help center (小红书聚光), February 2026. https://ad.xiaohongshu.com/api/light/help/doc/content?shortcutId=d88f6b141046664219978e32619b31be

On Xingtu, the creator marketplace for Douyin (抖音), the interaction rate is likes, comments and shares divided by plays. Source: Ocean Engine (巨量引擎), Xingtu help center, June 2026. https://www.xingtu.cn/help-center/demander/115182

So a count built on Juguang’s five actions runs higher than one built on Pugongying’s three, for the same note. Neither is padding by itself. Switching between them from month to month is, and it’s hard to spot unless the formula is printed on the slide.

Ask for it in words. Then ask that it stay fixed for the whole contract.

Attribution windows that flatter a campaign

Sales numbers rest on two choices. How long after an ad click does a sale still count? And does paid traffic sit in the same pile as natural traffic?

Juguang reports order value within 15 days and within 30 days of an ad click, and credits the last ad click. Source: Xiaohongshu Juguang help center (小红书聚光), February 2026. https://ad.xiaohongshu.com/api/light/help/doc/content?shortcutId=d88f6b141046664219978e32619b31be

Xingtu’s product seeding report offers windows of 1, 3, 7, 14 or 30 days. Source: Ocean Engine (巨量引擎), Xingtu help center, June 2026. https://www.xingtu.cn/help-center/demander/140449

A 30-day number will beat a 7-day number for the same ads. And your store’s own back end may still disagree with both.

The store back end attributes sales by order time. Juguang attributes them by click. Source: Xiaohongshu Juguang help center (小红书聚光), January 2026. https://ad.xiaohongshu.com/api/light/help/doc/content?shortcutId=c0b80b06af593eff0d0f2680b9df6701

Long windows aren’t always a trick. Some products take months to sell.

Xiaohongshu’s (小红书) marketing chief said a robot vacuum’s return turned at about 45 to 60 days, and many buyers of one body oil took 60 to 90 days to decide. Source: Huxiu (虎嗅), via Sina (新浪), December 2024. https://finance.sina.com.cn/wm/2024-12-20/doc-ineacsym5596846.shtml

The paid and natural split is where padding hides best. Douyin (抖音) sells paid boosts for a single post, called DOU+, and Xiaohongshu sells heating (加热). A boosted post can look like an organic hit.

Xingtu reports show all traffic by default, paid included. Only when the traffic split is switched on do they show the non-marketing part. Source: Ocean Engine (巨量引擎), Xingtu help center, June 2026. https://www.xingtu.cn/help-center/demander/142935

Until January 23, 2026, Pugongying’s post-campaign report did not show paid heating data separately. Source: Xiaohongshu Pugongying help center (小红书蒲公英帮助中心), April 2026. https://pgy.xiaohongshu.com/api/pgy/help/doc?shortcutId=da9e48ae271a1da33022122af1f0d56a&role=4

So an older Xiaohongshu report that blended heated and natural reach may reflect the tool, not a trick. A report written since then has no such excuse. If a creator post “went viral” and the agency also paid to heat it, you’re entitled to see both halves.

The screenshot problem: six great posts, forty missing

A deck with six screenshots of six great posts tells you nothing about the other forty. The fix is the full list: every post in the period, with reads, interactions and paid spend, exported from the back end.

Courts have started drawing the same line between what was claimed and what was delivered.

An agency billed a Jiangxi brand 727,000 yuan and cited 15.752 million plays of a related topic. The court counted the brand films’ own exposure, 38,900, and rejected likes the agency bought during the dispute. Source: New Rule of Law Daily (新法治报), via Phoenix Jiangxi (凤凰网江西), October 2026. https://jx.ifeng.com/c/8x1euV2KjUG

In a 2026 typical case, an outsourced operator that ran platform stores for merchants faked deals to lift its clients’ rankings. Regulators fined it 100,000 yuan, and the court called the practice “data padding” that the law does not protect. Source: Supreme People’s Court (最高人民法院), August 2026. https://www.court.gov.cn/zixun/xiangqing/507691.html

Regulators haven’t let up either.

The Cyberspace Administration of China’s 2026 campaign targets follower and like counts faked with boosting software, phone farms and gray-market sellers. Source: Cyberspace Administration of China (中央网信办), June 2026. https://www.cac.gov.cn/2026-06/18/c_1783524858389800.htm

China social media reporting metrics that actually matter

That covers what to distrust. Now for what to ask for. These six are harder to inflate, and each ties to something you can check.

1. Net follower change, new follows minus unfollows, per platform.
2. Unique readers or viewers per post, split by source. WeChat’s data interface already separates follower pushes, chats, Moments, recommendation and search.
3. Interactions per post, with the formula written down and held constant.
4. Search after viewing, where the platform measures it.
5. Paid and natural results side by side, never blended.
6. Cost per lead or sale, with the attribution window and rule stated.

On Xingtu, the search-after-viewing rate is searches after viewing divided by video impressions. Source: Ocean Engine (巨量引擎), Xingtu help center, June 2026. https://www.xingtu.cn/help-center/demander/115182

China’s ad industry standard points the same way.

A China Advertising Association standard defines invalid traffic, including manipulation of measurement data for money. Traffic that looks invalid but may come from real people should be shown separately in the report. Source: China Advertising Association (中国广告协会), T/CAAAD 006-2024, in force October 2024. https://digitalad.china-caa.org/TechLab/pdf/TCCSA%20553-2024%20TCAAAD%20006-2024.pdf

Ten questions to send your agency this month

Copy these into an email. Most of the answers already sit in the platform back ends, so keep the replies and set them beside next month’s deck.

1. For each reads or views figure, is it people or times, and does it include non-followers?
2. What was our net follower change per platform, new follows minus unfollows?
3. Did any week’s follower jump come from a giveaway, a paid push or a creator post? Which?
4. What is the exact engagement rate formula, top and bottom?
5. Has that formula changed at any point since we started?
6. Which figures come from Pugongying, Juguang, Xingtu or the app itself, and does any chart mix them?
7. What attribution window and rule sit behind each sales figure?
8. How much of this month’s reach was paid, including heating and DOU+, and what did the natural part do alone?
9. Can you export every post from the period, not only the highlights?
10. Can we have read-only access to each platform back end?

If an answer is “we’ll check”, that’s fine. If it’s “that’s how everyone reports”, it isn’t. Ask again in writing, and ask for back-end access before the next invoice.

What our reporting covers, so you can compare

Hold us to the same ten questions. A monthly performance report is part of every package on our pricing page. Here is what our service pages say those reports cover, quoted as published.

Page	What it says our reports cover, as published
WeChat	Follower growth, article reads, engagement, ad spend at cost, and leads or sales where tracking allows
Strategy, Campaigns and Analytics	Follower growth, engagement rates, content performance, ad spend efficiency, and competitive benchmarking
Other platforms	We pull data from platform backends directly, not third-party tools; Chinese-language data attached
Advertising	Weekly reports during active campaigns

Those lines are commitments, not results. Our Strategy, Campaigns and Analytics service page also sets out how KPIs are chosen and reviewed. The test for any report, ours included, is whether it answers the ten questions above.

FAQ

What metrics should a China social media report include?

Net follower change per platform, unique readers or viewers per post split by source, and interactions per post with a fixed, stated formula. Add search after viewing where the platform measures it. The report should separate paid and natural results, and give cost per lead or sale with the attribution window stated. Raw impressions and gross followers are context, not results.

How is engagement rate calculated in China?

There is no single formula, and the denominator matters most. The same note can show 5% over readers and 0.2% over followers. On Xiaohongshu (小红书), the creator tool Pugongying divides likes, saves and comments by reads. On Douyin (抖音), Xingtu divides likes, comments and shares by plays. Ask your agency which formula it uses, and hold it steady from month to month.

How do I know if my agency’s numbers are real?

Ask for read-only access to each back end and an export of every post, not screenshots. Compare net follower change with the posting calendar. Ask for paid and natural traffic shown apart, and the attribution window behind every sales figure. A sudden follower jump with no campaign behind it needs an explanation.