China's First AI-Ghostwritten 'Seeding Post' Case — a Duty of Care for Generative-AI Providers

Editor’s Note — DCC.

This is DCC’s case summary of China’s first unfair-competition action over AI-ghostwritten “种草笔记” (seeding posts) — the second-instance judgment of the Hangzhou Intermediate People’s Court (杭州市中级人民法院), case no. (2025) Zhe 01 Min Zhong No. 3998, on appeal from the Hangzhou Internet Court ((2024) Zhe 0192 Min Chu No. 3396). The judgment was circulated as ~44 page-images by the 数据法盟 channel citing 知产库; DCC ran OCR over those images and summarises the court’s findings and holding below. In the circulated copy the defendant companies’ names are redacted (“XX”); the plaintiff is named. All English renderings of the court’s language are DCC’s. The framing for overseas counsel is ours.

Case at a glance

Plaintiff (respondent on appeal): Xingyin Information Technology (Shanghai) Co., Ltd. (行吟信息科技（上海）有限公司) — the operator of Xiaohongshu (小红书 / RED).
Defendants (appellants): the operators of the “AI写作鹅” (“AI Writing Goose”) tool — two Hefei companies (names redacted) plus a third company that distributed the app via the “单词乎” (dancihu.com) site.
Claims: (1) copyright infringement and (2) unfair competition.
Result: copyright claim rejected; unfair competition established under Article 2 of the Anti-Unfair Competition Law (反不正当竞争法). On appeal: liability affirmed, injunction upheld (reworded), damages reduced from RMB 200,000 to RMB 100,000, and the download-distributor’s joint liability reversed. Final judgment.

What the service did

“Seeding posts” (种草笔记, lit. “planting-grass notes”) are the staged, enthusiastic, first-person product-recommendation notes that drive discovery commerce on Xiaohongshu. Their persuasive power depends on the reader believing they reflect a real user’s genuine experience.

The “AI写作鹅” tool offered modules named “小红书种草文案” (Xiaohongshu seeding copy), “小红书旅游攻略” (Xiaohongshu travel guides), “小红书文案”, and “小红书笔记标题” (Xiaohongshu note titles). A user could enter a few keywords and “one-click create” a complete Xiaohongshu seeding post; the output fabricated the user’s first-hand experience and genuine feelings. The tool was a paid service after first use (the court noted membership prices of RMB 168 lifetime / 98 a year / 40 a month), and its pages advertised that it would “generate click-inducing titles for your Xiaohongshu notes” and “produce share articles that fit Xiaohongshu’s tone.”

Two claims, two outcomes

Copyright — rejected. Xingyin asserted copyright (via its user-agreement licence) in platform content and pointed to one example post copied into the tool and regenerated. The court found the object of the claim unclear: the example copy-and-upload was actually Xingyin’s own act during evidence collection, not the tool operator’s, and a blanket claim to “all content on the platform” did not identify a specific work or establish authorship and originality. So the copyright claim failed.

Unfair competition — established under the Article 2 general clause. Notably, Xingyin did not plead confusion (AUCL Art. 6) or false advertising (AUCL Art. 8); it invoked the general clause, Article 2, which courts apply to conduct that is not specifically enumerated in Chapter 2 but breaches good faith and commercial ethics. Because the case was about whether a generative-AI product at the application layer disrupts an existing content ecosystem — something Chapter 2 does not address — the court (both instances) analysed it under Article 2 across three questions: does the plaintiff hold a protected interest; is the conduct improper; does it cause harm.

The protected interest was the authentic “seeding” content ecosystem of Xiaohongshu — a UGC community built on real user experience. The court credited Xingyin’s heavy, sustained investment (community rules and conventions, a “Woodpecker” anti-fake-promotion campaign, governance reports, and express bans on “using AI to fabricate usage experience or effect”), and the resulting traffic and stickiness (by figures in evidence: 20M+ monthly-active creators and 30B+ daily note impressions by early 2023; ~300M monthly-active users by 2024, ~90% UGC). That ecosystem-based competitive advantage, it held, is protected by the AUCL.

The heart of the case: a duty of care for generative-AI providers

The most consequential part of the appeal judgment is an explicit framework for when a generative-AI service provider owes, and breaches, a duty of care. The court stressed that generative AI has a dual nature — technical service and content supply — unlike a traditional search-link or hosting provider, and that its output is non-deterministic (each generation differs), so infringement risk is “highly random” and the provider must build in reasonable measures to avoid or reduce it. It then weighed four factors:

Is it generative AI? Citing Article 22 of the Generative AI Services Interim Measures (definition of generative-AI technology), the court found “AI写作鹅” autonomously generated new content from user prompts via its own model, algorithms, compute, and training data — so it is a generative-AI service, and its provider bears the corresponding duty of care.
Does it use a specific scenario / another’s product as its “application layer”? The modules were named after “小红书” (Xiaohongshu) — a service “designed as a directional scenario oriented to another’s product.” A provider that uses Xiaohongshu as its application layer should be bound by Xiaohongshu’s platform rules, which do not ban AI but do ban using AI to fabricate experience or effect; the tool’s “seeding” output was, in substance, AI-powered “fake seeding” (虚假种草) that violated those rules.
Is it directional and inducing? Citing Article 5(1) of the Measures (encourage innovative AI applications that generate “positive, healthy, uplifting” content), the court read the tool’s marketing (“generate click-inducing Xiaohongshu titles,” “share copy matching Xiaohongshu’s tone”) plus the “one-click” fabricated output as showing a subjective purpose to supply fake seeding — so the provider “can hardly be said to be in good faith” (难谓善意).
Is it a paid, for-profit service? Because the tool charged a membership fee — unlike free, general-purpose generative-AI platforms — the court held it should bear a higher duty of care.

Putting it together: the providers should have foreseen that a Xiaohongshu-targeted, directional, inducing, paid tool would lead users to generate fabricated seeding posts and publish them to Xiaohongshu — yet they took no reasonable, necessary notice or reminder measures (告知、提醒). That failure meant they did not discharge the reasonable duty of care of a generative-AI provider and were at fault (过错); the conduct violated good faith and commercial ethics and was therefore unfair competition. The court was careful to add that the duty of care must match the provider’s technical control capability and not be so heavy as to deter innovation.

The court also drew a boundary: this case is not about regulating platforms that offer AI “seeding copy,” “travel guides,” or general text generation. “AI technology itself is neutral … but that does not mean its application is neutral”; what is regulated is a provider that, at the application layer, designs a scenario targeted at another operator’s product and improperly free-rides on the market results that operator has already achieved — guided by the principles of governing AI “according to law” and “AI for good” (智能向善), and an “inclusive and prudent” (包容审慎) posture that dynamically balances innovation, rights-holders, consumers, and the public interest.

What changed on appeal

The Hangzhou Intermediate Court affirmed liability and the injunction (reworded to “stop providing, via AI写作鹅, the ‘Xiaohongshu seeding copy / travel guide / copy / note title’ services”), but made two modifications:

Damages cut from RMB 200,000 to RMB 100,000. With no proof of actual loss or infringer gain, Xingyin sought statutory damages. The appeal court trimmed the award, expressly because AI legislation and industry norms remain immature and courts should stay “inclusive and prudent,” and because the harm was unproven — while still holding the providers at fault.
Joint liability of the download-distributor reversed. The third defendant ran the “单词乎” (dancihu.com) site and merely offered the app’s Android download and an iOS link — a network service. With no evidence of common intent or knowledge (the tool also had general “AI writing” and “AI drawing” features), the court found no fault and no joint liability, allowing that defendant’s appeal. Liability rests on the two Hefei operators of the tool.

(Legal basis: AUCL Articles 2 and 17; Article 23 of the Supreme People’s Court Interpretation on the AUCL; Civil Procedure Law Articles 67 and 177. The judgment is final.)

Why overseas counsel should care

Generative-AI providers in China carry an affirmative, scenario-sensitive duty of care. The four-factor test is the takeaway: a tool is most exposed when it (i) is generative AI, (ii) targets a specific platform or another’s product as its “application layer,” (iii) is marketed in a directional, inducing way, and (iv) is monetised. Tick those boxes and a Chinese court will ask what notice, reminder, and abuse-prevention measures you built in.
Naming and marketing are evidence. Naming features after a third-party platform was treated as “free-riding,” and promotional copy was read as proof of intent. Align product names and marketing with permitted-use claims; do not brand a feature on someone else’s ecosystem.
Build guardrails into the product, and weigh your business model. A paid service drew a higher duty of care. Pair the regulatory duties under the Generative AI Services Interim Measures and the AI-content-labelling Measures with product-level controls — usage warnings, content labelling, and friction or refusal for deception-prone prompts.
The general clause (AUCL Art. 2) is the live instrument for AI-content harms. With no enumerated tort to fit, Chinese courts are using the good-faith / business-ethics general clause to reach novel AI conduct — and reading the Generative AI Measures into the standard of commercial ethics, even where the case is decided on competition grounds.
“Inclusive and prudent” cuts both ways. It produced liability and a reduced award; expect Chinese courts to find fault but moderate damages while the AI rulebook matures — useful for calibrating litigation risk.

DCC sources

Original: 《全国首例AI代写”种草笔记”案判决书》, circulated by the 数据法盟 channel citing 知产库 — mp.weixin.qq.com. Underlying decision: Hangzhou Intermediate People’s Court, (2025) Zhe 01 Min Zhong No. 3998 (second instance; first instance Hangzhou Internet Court, (2024) Zhe 0192 Min Chu No. 3396); reported by CCTV, May 2026. DCC reconstructed the text by OCR over the judgment images.
Cross-references on DCC: the Generative AI Services Interim Measures (Arts. 4(3), 5(1), 22, cited in the judgment) · the Measures for Labelling AI-Generated and Synthetic Content (the regulatory analogue of the “notice/labelling” duty the court found missing).
Part of the AI Governance and Enforcement domains on DCC.

This is an editorial case summary written in DCC’s own words for overseas readers, reconstructed from OCR of the judgment images circulated with the source post — not an official transcript or a certified translation. Defendant names were redacted in the source; quoted phrases are short and attributed. Figures and holdings are as DCC read them from the images and may contain OCR error. Not legal advice.