Xiaohongshu spent four years protecting its community from AI. The strategy worked. Now urgency is replacing restraint.

On 30 April 2026, Xiaohongshu, the Chinese social-search platform known internationally as RedNote, sent an internal letter to all employees announcing the establishment of Dots, a first-level AI department. The department is responsible for model R&D, infrastructure, engineering, and product. It reports to the newly appointed president, Ding Ling, who goes by the alias Conan. The decision marks the public end of four years of deliberate restraint that the company maintained while every other major Chinese internet company was racing to build AI products.

The restraint was not accidental. It was strategic. And for most of the last four years it was correct.

When ChatGPT emerged in late 2022, Xiaohongshu founder Mao Wenchao borrowed an employee's phone and asked it whether Xiaohongshu would be disrupted. He began monitoring AI developments and requiring fortnightly updates from his team. In an August 2023 internal letter, he argued that while overseas users were asking ChatGPT life experience questions that overlapped with Xiaohongshu's territory, that pattern existed precisely because there was no Xiaohongshu equivalent abroad. Xiaohongshu's content moat, rooted in real user-generated lifestyle posts, would not be easily shaken by AI in the short term.

The thesis turned out to be largely right. While Doubao, Qwen, Yuanbao, and DeepSeek raced to dominate the AI chatbot category through 2023 and 2024, Xiaohongshu's user base kept growing on the strength of its community search and recommendation experience. AI products were not eating the platform. The community was strong enough to defend itself.

What is more interesting than the restraint is what justified it operationally. The product team at Xiaohongshu actively resisted AI integration that risked diluting the community experience. A former employee described the platform's homepage architecture in unusual terms. The misaligned two-column layout is designed to create the feeling of a city street. Users are like people wandering through a neighbourhood, seeing signs of different sizes staggered along the way. When they click into a post, it feels like knocking on a door. This is not metaphor. This is product philosophy made explicit. Any AI integration that disrupted the city-street feeling was rejected.

Even when Xiaohongshu's AI search product Wenyiwen was being tested, the team was cautious about how broadly to deploy it. Initial coverage was capped at one to two percent of user queries. The team wanted to expand to ten percent. Senior leadership refused. The compromise was three to four percent. The principle being protected was specific. AI summaries reduce the time users spend browsing, and browsing is where the platform's commercial value gets produced. The product was wary of any feature that improved efficiency at the cost of engagement.

Then DeepSeek emerged in early 2025 and the calculus changed. DeepSeek made high-quality AI essentially frictionless for the average Chinese user. Xiaohongshu ran internal data comparisons. If users had both DeepSeek and Xiaohongshu installed, would their search volume on Xiaohongshu decline? The team did not believe AI would replace traditional search. After DeepSeek, they started to worry.

The Wenyiwen feature was eventually expanded. The community impact was the opposite of what management had feared. Wenyiwen increased community user retention by approximately two to three percent, a significant gain for a mature product. Daily active users on the feature reached the tens of millions. Browsing time per question went down. Frequency of use went up. The net effect was that someone who used to ask one question a week was now asking questions daily. Total platform engagement increased.

This is the part that contains the lesson Asian platform operators should read carefully. Xiaohongshu's caution was rational, the protection of community experience was strategically correct, and the eventual integration of AI was beneficial. But the gap between when AI integration became technically possible and when Xiaohongshu finally committed to it was approximately three years. During those three years, the platform lost the ability to define how AI search would feel inside community products. ByteDance, Tencent, and Alibaba established their AI products as the default consumer reference points. Xiaohongshu is now playing catch-up in a category it could have led.

The structural problem is that restraint and urgency are difficult to manage in sequence. Companies that excel at one rarely excel at the other. Xiaohongshu was institutionally good at protecting product integrity. The same instincts that produced that strength made it slow to mobilise when the strategic environment changed. By late 2025, when Mao returned from a Silicon Valley visit and announced that AI's roadmap had converged enough for Xiaohongshu to invest more, the company's technical AI team numbered fewer than 100 people. ByteDance's Doubao team alone had more than 1,000. The compounded talent gap is significant.

Internal observers have used a specific phrase to describe what is happening now. Everyone is transforming into an AI product manager. Non-AI roles have been pulled into AI literacy training. Some employees are required to learn vibe coding to build AI tools and embed them into their workflows. Employees are being asked to evaluate their direct managers' AI usage capabilities through internal surveys. The 2026 campus recruitment has opened almost exclusively AI-related positions. This is the institutional energy of a company that has decided, simultaneously and across every function, that AI is now mandatory.

The risk of that energy is that it overcorrects. Conan, the newly appointed president responsible for integrating community, e-commerce, monetization, and the technology system, comes from a consulting and finance background. Stanford GSB MBA, Boston Consulting Group, Citi. Her appointment is widely read as a sign that Xiaohongshu's AI strategy will be product-led rather than technology-led. This is consistent with the company's strengths. It is also a quiet admission that Xiaohongshu does not have the technical bench to win on model capability alone.

Xiaohongshu's eventual AI position is likely to be a specific bet. Use the community data and the search positioning to power vertical AI products in lifestyle categories where the community already has authority. Travel. Fashion. Beauty. Food. Wellness. The bet is that AI does not have to be a general-purpose chat interface to succeed. It can be a series of specialised tools that solve specific user problems better than general chatbots can. This is a defensible position if executed well. It is also a substantially more modest ambition than what ByteDance or Alibaba are building.

The broader lesson is the part that matters for any Asian operator running a community or content platform. Restraint in the face of new technology is more defensible than enthusiasm, because most new technologies disappoint, and the cost of compromised product integrity is permanent. But restraint becomes a strategic liability the moment the technology stops being speculative and becomes part of the user's default expectation. Identifying that moment is the hardest call leadership makes. Xiaohongshu identified it late. Late is recoverable. The cost is being a follower in a category the platform was philosophically positioned to lead.

The other thing worth noting is what Xiaohongshu did not do. It did not abandon the community-first product philosophy as it integrated AI. It did not flood the homepage with AI features that compromised the city-street feel. It moved Diandian, the AI assistant, into the community as an embedded auxiliary tool rather than as a replacement interface. Even in catch-up mode, the company is still protecting what made it valuable. That is the discipline that will determine whether Xiaohongshu's AI bet pays off or whether it becomes another platform that diluted its own community in the panic of trying to compete on the wrong terms.

The strategic question Asian platform operators should ask themselves after reading this. What is the city-street feel of your product, and what AI integrations would compromise it. If you cannot answer the first part of that question, you cannot answer the second. Xiaohongshu's strength was that it always knew the answer. Its risk now is whether it can keep answering it correctly under the institutional pressure of the catch-up.