Ten years ago, the number everyone quoted about China was that up to 40% of the country’s digital ad spend was fraudulent. Display banners nobody saw, clicks generated by scripts, video pre-rolls served to server farms. That number did its job: it scared brands into asking questions. But it belongs to a market that no longer exists. The fraud did not disappear. It moved.
Today the waste is not in banner impressions. It is in engagement. Views, likes, comments, followers, live streaming sales figures. The Chinese word for it is 刷量, literally “brushing up the numbers”, and it is a mature service industry with price lists. That is my position, and I will defend it: if you are a foreign brand spending money in China in 2026 and you are still auditing your media plan for click fraud, you are guarding the wrong door.
Written by Olivier Verot, founder of Gentlemen Marketing Agency, based in Shanghai since 2012. I have signed, renegotiated and cancelled several hundred KOL and live streaming contracts for European and American brands, and I have seen what the reporting screenshots look like on both sides of the deal.
The old problem is smaller. The new one is bigger.
Classic ad fraud, the kind that happens on programmatic inventory, has genuinely improved in China. Platforms consolidated. Baidu, Tencent and ByteDance sell most of the inventory directly, and a walled garden has less incentive to let a third party steal from it. AppsFlyer’s 2026 mobile advertising fraud report puts iOS install fraud at 11.7% in Q1 2026, down from 17.5% a year earlier, with Android sitting stubbornly in the 14 to 15% band. Those are not single digits, but they are a long way from 40%.
Meanwhile the global picture went the other way. eMarketer reported that invalid traffic worldwide climbed from 26% in January 2026 to 40% by June, driven by AI-generated bot farms that behave convincingly enough to pass basic filters. So China is no longer the outlier it was in 2017. Its digital advertising problem is now everybody’s problem.
What is specific to China is the second layer.
In a market where a brand’s decision to work with an influencer is based on a screenshot of a dashboard, and where the influencer controls that dashboard, the incentive to inflate is enormous. Sixth Tone documented the price list: roughly 3 yuan for a hundred likes, 2 yuan for ten thousand views. At those prices, buying a credible profile costs less than the taxi to the meeting where you present it. Some hosts go further and place fake orders during a live stream to hit the sales guarantee they signed with the brand, then refund them a week later once the invoice is paid.
How a foreign brand actually gets taken
It rarely looks like fraud while it is happening. It looks like a good campaign.
The agency or the MCN proposes a KOL with 1.2 million followers on Douyin and a nice engagement rate. You approve. The live stream runs for four hours, the counter on screen moves fast, and the post-campaign deck shows 2.8 million views, 40,000 comments and 6,000 units sold. Everyone is pleased. Then two things happen. The comments, read individually, are generic and repetitive, because a barrage service was paid to fill the chat. And thirty days later your warehouse tells you the return rate on that batch was 55%.
That return rate is the tell most foreign brands miss. Return rates in live commerce are structurally higher than on shelf e-commerce, because the buying decision is impulsive and the customer has not searched for the product. Apparel is the worst category. A live stream apparel return rate of 30 to 40% is normal in China and nobody is cheating you. A return rate above 50%, concentrated in the first 72 hours, with orders clustered in tight time blocks, is not consumer behaviour. It is a host manufacturing a sales figure.
Chinese regulators have started to move on this. In 2026 a Chinese media company was fined 400,000 yuan for placing and cancelling orders to fabricate transaction data and then advertising the fake sales figures, and a separate case brought a 100,000 yuan fine for running dozens of controlled accounts to post fake comments and product reviews. The Cyberspace Administration also now requires creators talking about medicine, health, law, finance or education to prove real professional credentials before they post or go live, and explicitly bans purchased followers used to inflate credibility. Enforcement is real but it is slow, and it protects Chinese consumers before it protects your media budget.
What to check before you sign anything
You do not need a forensic team. You need four habits.
Pull third-party data, never the influencer’s screenshot. On Douyin, Chanmama (蝉妈妈) tracks live room performance, product-level sales and historical GMV per host independently of what the host tells you. ByteDance’s own 巨量算数 (Ocean Engine Trendinsight) gives you category search volume and audience composition. If the numbers in the media kit do not match the third-party numbers within a reasonable margin, you have your answer without a conversation. This one step removes most of the risk and costs a subscription.
Check that engagement and sales tell the same story. Real audiences convert at boring, predictable rates. A profile with 800,000 followers and 300 comments per post that suddenly delivers a huge live session is suspicious. So is the reverse: enormous engagement and almost no product movement means the audience is either bought or completely mismatched with what you sell. Look at the shape of the curve, not the peak.
Pay on performance wherever the influencer will accept it. The affiliate and KOC model has grown for exactly this reason. Instead of a 200,000 yuan flat fee to one big name, you distribute commission-based links across a hundred small creators who earn only on validated, non-returned orders. Nobody buys fake views to earn a commission on a sale that gets refunded. The economics of the fraud stop working. Flat fees are sometimes unavoidable with top hosts, and in that case tie half the payment to net sales measured after the return window closes.
Test small before you commit big. Run three creators at 15,000 to 30,000 yuan each before you run one at 300,000. You are not buying reach at that stage, you are buying information: which creator’s audience actually clicks, what the real return rate is on your product, which message survives contact with Chinese consumers. Brands that skip this step are the ones who get taken, every time.
A client example
Bruno runs an Italian kitchenware brand. He came to us after spending 480,000 RMB on two Douyin live streams booked through an intermediary he met at a trade fair. The reports were spectacular: 5.1 million cumulative views, 9,800 units. His actual net revenue after returns and platform fees was under 90,000 RMB. When we pulled the host’s history on Chanmama, the two sessions he had paid for were the only outliers in eighteen months of otherwise flat performance.
We rebuilt the plan around 60 mid-tier and small creators on Xiaohongshu and Douyin, all paid on commission, with product seeding instead of fees for the smallest ones. It was slower and far less impressive to look at. Over seven months he did roughly 1.9 million RMB in net sales at a blended cost of 21% of revenue, with a return rate of 19%. The reason it worked is not that small creators are more honest. It is that the payment structure made lying pointless.
Where paid media still fits
None of this is an argument against advertising in China. Paid links are a real part of a campaign. My problem is with strategies that are only paid links, because that is where the exposure to fraud concentrates and where the spending stops the moment you stop paying. Baidu PPC buys you visibility on a keyword this quarter. Ranking in the natural results, having your brand cited when someone asks DeepSeek or Doubao for a recommendation, and holding a Xiaohongshu presence people find on their own, those keep working after the invoice is paid.
That mix has shifted a lot in the last two years, and search behaviour has moved with it. It is worth understanding why brands are moving budget from Tmall to Douyin, and why owned channels like a WeChat mini shop matter more than they did: traffic you own cannot be faked to you by a third party. You count it yourself.
The old structural causes still apply. Publishers in China hold the data, third-party auditing is conceded rather than granted, and a lot of business still runs on relationships rather than contracts. What has changed is that the tools to verify exist now, they are Chinese, and they are affordable. A brand that gets defrauded in 2026 is usually a brand that did not look.
FAQ
How do I know if my agency is protecting me or profiting from the same problem?
Ask for three things. The third-party data source behind every influencer recommendation, before you sign. The net sales figure after returns, not the gross GMV shown on the platform dashboard. And the actual invoice from the MCN, so you can see the margin. An agency that resists any of the three is not necessarily dishonest, but it is being paid in a way that does not align with your result. We publish our numbers post-return for exactly this reason, because the alternative is reporting sales that come back to the warehouse.
Can I get my money back if a KOL faked the results?
Rarely, and only if you planned for it. Chinese platforms will sometimes act on a documented complaint about a host, and regulators have issued real fines for fabricated transaction data, but neither process refunds you. Your protection is contractual and it has to be written before the campaign: payment split into two parts, the second one released after the return window, with net delivered sales as the trigger rather than views or gross orders. Getting that clause into a Chinese contract is a negotiation. It is much easier than a recovery attempt afterwards.
We audit KOL and live streaming spend in China
GMA runs influencer and live commerce campaigns for foreign brands from Shanghai, with third-party verification on every creator we propose and reporting based on net sales after returns.
We also review campaigns brands have already run, when the numbers looked good and the revenue did not follow.
Send us the deck and the invoice: contact our team, or read more about how we work.
