While the entire city is captivated by the grandeur of the Trojan Horse, the real peril that threatens its downfall lurks silently in the shadows.

Authored by | BlueHole Business, Jiang Zian

Xiaohongshu, poised on the brink of a Hong Kong Stock Exchange debut, has been grappling with a sense of unease lately.

Swirling rumors of its impending listing, a secondary market valuation soaring to US$50 billion, and a whistleblower report from a former employee on 'listing compliance' have repeatedly thrust this star company into the limelight.

Yet, for a commercial entity, going public is merely a阶段性考验 (a phased challenge). Beyond the IPO frenzy, Xiaohongshu faces a more pressing issue: its once-impenetrable commercial barriers are being gradually eroded by AI.

Over the past eight years, the company has meticulously built a formidable moat through its 'slow' and steady approach. Despite facing multiple encirclements from giants like Alibaba, ByteDance, and Meituan, Xiaohongshu has not only remained unscathed but has also blossomed into a quasi-giant with 170 million daily active users and 400 million monthly active users, achieving a net profit of US$3 billion in 2025.

However, the gravest threats often emerge from unexpected quarters. With the advent of the AI era, platforms like DeepSeek and Doubao have begun to encroach upon Xiaohongshu's 'useful' capabilities, giving rise to a palpable sense of AI anxiety. As users grow accustomed to the answers provided by AI, Xiaohongshu's lifeblood as a 'consumer decision-making gateway' is at risk of being severed, shaking its very foundations.

Much like the 'Trojan Horse' metaphor: while the entire city is mesmerized by the Trojan Horse's splendor, the real danger that could lead to its downfall is quietly creeping in.

Xiaohongshu's management has sensed this looming threat, leading to a significant internal restructuring recently. In late April, founders Mao Wenchao and Qu Fang stepped back to focus on strategy and organization, while Ding Ling was promoted to president, integrating Xiaohongshu's community, e-commerce, commercialization, and technology systems, and establishing an AI first-level department, Dots. This marks the first time Xiaohongshu has elevated AI to the pinnacle of its corporate strategy.

While the outside world scrutinizes Xiaohongshu's IPO journey with a magnifying glass, the AI war brewing beneath the surface may well be the true battle for Xiaohongshu's survival.

Elaine, a dedicated Xiaohongshu user in Shanghai, has noticed a significant reduction in her daily app usage, which once spanned five to six hours.

Her most intuitive feeling is: 'After having a child, I often turn to Xiaohongshu for parenting advice. However, whenever I encounter complex issues, Xiaohongshu loses its utility and instead creates anxiety.' Several times, after consulting a pediatrician with answers obtained from Xiaohongshu, the doctor would remark, 'You got this from Xiaohongshu again, didn't you?'

After multiple comparisons, Elaine found that her usage habits had irreversibly changed: the expressions in some Xiaohongshu posts tend to be extreme and emotional, whereas platforms like Doubao, DeepSeek, and Kimi offer more professional and systematic answers when faced with complex issues.

Elaine's experience is far from isolated. Micro-level changes in user behavior are corroborated by macro-level data.

According to a Nomura Securities report, in May 2026, the total user time spent on Xiaohongshu declined by 1% year-on-year. Although seemingly small, this marks the first decline since the institution began tracking Xiaohongshu six years ago. A breakdown reveals that while daily active users on the platform increased slightly by 3%, the average daily time spent per user decreased by 4% year-on-year.

Another set of data unveils the true destination of the lost time. Quest Mobile's May monitoring shows that the daily active users of ByteDance's AI application, Doubao, surged 3.6 times year-on-year to 158 million; DeepSeek's daily active users reached 30.5 million, surpassing Alibaba's Tongyi Qianwen to rank second in the industry.

As Nomura Securities judges, 'Xiaohongshu is encountering the same AI impact as search engines and information portals: users' demand for acquiring information and knowledge is gradually shifting to AI chatbots.'

This aligns perfectly with American business analyst Ben Thompson's 'Aggregation Theory': in the Internet era, whoever monopolizes the entry point for end-user demand can attract backend supply (content and goods) to converge towards themselves, thereby reaping maximum profits.

Over the past eight years, Douyin has failed to defeat Xiaohongshu because their underlying operating systems differ. Douyin's essence is entertainment and stimulation, while Xiaohongshu, with its 'slowly cultivated' sense of authenticity, has monopolized the 'consumer decision-making gateway' for high-value users, thereby aggregating a vast amount of useful content supply and ultimately forming an extremely high moat.

However, Xiaohongshu is now facing not just community content competition on the same dimension but a reshaping of underlying demand, a challenge far greater than any previous threat from Douyin.

AI large models are directly intercepting users' search demands through higher-dimensional interaction methods. Once users become accustomed to making decisions on the AI side, Xiaohongshu's core search foundation will be cut off.

Once it loses the 'decision-making gateway' as a source of traffic, the hundreds of billions of notes produced by millions of bloggers will ultimately have only two destinations: becoming corpora for general large models to learn from or losing monetization value as attention drifts away.

The 'interception' by Doubao and others is the external threat Xiaohongshu faces in the AI era. The internal worry stems from itself: every time commercialization accelerates, the advantage of AI large models in being 'ad-free and providing direct conclusions' is amplified.

Xiaohongshu once cultivated a 'sense of authenticity' in its community through a restrained commercialization pace, continuously improving user time spent through high-quality content. The combination of the two made it a highly valuable consumer decision-making gateway. According to Caixin, the average daily search volume on Xiaohongshu has increased from 600 million in the fourth quarter of 2024 to over 800 million in early 2026.

This has brought significant financial returns over the past few years. Xiaohongshu turned a profit for the first time in 2023, with profits reaching US$1 billion in 2024. Revenue is expected to reach approximately US$6 billion in 2025, with profits tripling to US$3 billion.

As revenue scales up, the demand for further commercialization has also become increasingly prominent.

Currently, Xiaohongshu still relies heavily on advertising for monetization. According to estimates, advertising's share of revenue increased from about 72% in 2024 to about 76% in 2025. However, the winds in the entire advertising industry have changed. According to a WPP Media report, the market share of domestic brand promotion advertisements continued to decline in 2025, while the share of performance-based and sales-result-oriented advertisements maintained growth. Advertisers are no longer satisfied with mere brand exposure but demand direct reflection in product sales (i.e., performance advertising).

In the past, performance advertising was a weak point for Xiaohongshu. To meet client demands, Xiaohongshu began vigorously promoting performance conversion from 2025. Beyond its own e-commerce closed loop, Xiaohongshu has connected with external platforms such as Taobao, JD.com, and Meituan, allowing clients to clearly see in-store and conversion data after advertising placement.

This strategy has driven rapid growth in advertising placement but has also changed Xiaohongshu's placement ecosystem. Elaine has also noticed that when she searches for many maternal and child products, five or six posts in the results recommend the same brand.

Initially, she thought it was because the product had a good reputation and many users. After several poor purchase experiences, she began to suspect. Now her habit has changed: after searching on Xiaohongshu, she asks Doubao the same question and compares the two.

What Elaine doesn't know is that this is the result of Xiaohongshu's brand partners changing their placement strategies: shifting more budget from high-priced top bloggers to more cost-effective mid-to-low-tier KOCs (Key Opinion Consumers). As a result, phenomena such as fake grassroots recommendations and influencer soft advertising have increased, and the consumer decision-making content in the community has begun to 'water down.'

According to Caixin, an internal source at Xiaohongshu admitted that soft advertising harms the platform, affecting both user experience and the commercial ecosystem. Although the platform has assembled a professional team for systematic governance, this is often a long-term battle. A marketing service provider told Caixin that the simultaneous increase in marketing content and genuine user content on Xiaohongshu means users need to spend more time and effort discerning and synthesizing information, raising the decision-making threshold for users.

More troubling is that even Xiaohongshu's own AI large model is fighting with its 'moneybag.' According to 36Kr, Xiaohongshu's AI search 'Ask' generates product recommendations based on discussions among community users. If major advertising clients are ranked lower, brand partners will inevitably be dissatisfied. The commercialization team once hoped that AI search would not actively generate shopping recommendations.

This puts Xiaohongshu in a dilemma: the more truthful and objective AI answers are, the more likely they are to retain users but will affect advertising revenue; conversely, it will lead to user loss.

However, Xiaohongshu has not slowed down its commercialization pace. Even though users feel that marketing content has increased and decision-making has become more difficult, Xiaohongshu's current actual information feed ad load rate is still only 6%-10%, lower than Douyin's approximately 15%. To support future revenue growth, this ratio may continue to rise.

On one hand is the unstoppable revenue growth, and on the other is the irreplaceable sense of authenticity. Xiaohongshu has maintained a balance between the two for nearly a decade through its 'slow' approach. The emergence of AI is now breaking this balance.

Facing the shift in user search demand and the dilemma of internal ecological balance, Xiaohongshu needs to find a breakthrough. Before that, it has already paid four years of tuition in AI.

Looking back, Xiaohongshu's AI attempts over the past four years resemble a series of 'duds' that failed to make a sound.

As early as 2023, under the impact of ChatGPT, Mao Wenchao asked GPT whether 'Xiaohongshu would be disrupted' and subsequently required relevant teams to report on the industry's AI progress every two weeks. In March of that year, Xiaohongshu established an independent large model team to develop a general large model foundation called 'Zhuji.' In August 2024, the first independent AI application, 'DianDian,' was launched, positioned as a lifestyle search assistant.

However, this path quickly became untenable. In February 2025, DianDian publicly announced its integration with DeepSeek—the reason was straightforward: the self-developed model's experience was unsatisfactory. Another AI search feature in the community, 'Ask,' used the open-source Tongyi Qianwen. The self-developed foundation model line was once considered by outsiders to have stalled.

In the following two years, Xiaohongshu continuously adjusted its approach: in early 2025, it integrated the large model team into the 'Humanities Intelligence Lab' (Hi Lab), recruiting a group of 'trainers' with humanities backgrounds for post-training the model. Mao Wenchao personally designed the '45-degree angle looking up at life' persona for DianDian; in July, DianDian 2.0 was launched, adding voice and multimodal capabilities; in November, it abandoned the independent app route and integrated DianDian into the top-left corner entrance of Xiaohongshu.

The truly successful product turned out to be 'Ask.' This product, incubated by the search team and initially strictly limited to 3%-4% of search requests, improved community retention by 2-3 percentage points in 2025 and achieved daily active users in the tens of millions. Ironically, it highly overlapped in functionality with DianDian and was the result of separate bets by two departments. Currently, the 'Ask' feature in the Xiaohongshu app has been replaced by DianDian.

One self-developed path failed to materialize, and no blockbuster product emerged. The only data highlight came unexpectedly. This is the answer Xiaohongshu has given after four years of AI exploration.

Notably, in August this year, the Dots Model Lab open-sourced the dots3 note preview version for the first time: with a total of 280B parameters, 16B activated parameters, supporting 512K ultra-long context, and possessing multimodal understanding capabilities in text, vision, and voice. The official claims it is optimized for complex reasoning and long-term Agent tasks. According to the plan, the dots3 series will have three levels: note, jazz, and aria, corresponding to different task complexities and computational costs from lightweight to heavyweight. The official version will also be open-sourced soon.

This indicates that Xiaohongshu's self-developed foundation model route has not stopped but has shifted from closed-door self-development to open-source. However, 280B parameters and 512K context are just entry tickets. The reputation of open-source models depends on real-world testing. For content platforms building foundations, the real challenge remains unchanged: whether the model's capabilities can ultimately be integrated into products to defend the 'consumer decision-making gateway' that is being intercepted.

While still exploring its path, the choices of overseas peers may provide references for Xiaohongshu.

Overseas platforms with community models similar to Xiaohongshu have two starkly different choices when responding to AI impact: Reddit and Pinterest.

Reddit made the decision to license its content for AI training purposes. In 2024, it inked a lucrative US$60 million per year deal with Google, granting access to its public posts to train Google's AI model, Gemini. However, merely 'cashing in on its content' is fraught with risks. According to The Wall Street Journal, Reddit's executives noticed a decline in their traffic and started to reassess the agreement, even contemplating revoking Google's access. By surrendering its content, Reddit risks ceding the gateway to 'user demand,' effectively demoting itself to a mere provider of data for large AI models.

Pinterest, in contrast, took a different approach: retaining control over its user entry points. Rather than delving into the development of a general-purpose large AI model, Pinterest continued to channel its investments into AI-powered visual search tools, leveraging its proprietary data. This strategy keeps users within Pinterest's ecosystem, enabling them to complete decision-making processes in a self-contained manner (forming what are known as "closed loops").

Confronted with these two divergent paths, Xiaohongshu (Little Red Book) unveiled its strategy in late April of this year: establishing an AI-focused first-level department named Dots. This new entity is tasked with overseeing the entire pipeline, from model development to product deployment, with DianDian (a feature or tool within Xiaohongshu) being integrated into this framework. Judging from this strategic move, it's evident that Xiaohongshu leans towards Pinterest's model—utilizing AI to safeguard its decision-making gateway, rather than relinquishing its community content to external entities.

Numerous Xiaohongshu users have reported that DianDian shines particularly in travel guide scenarios, seamlessly integrating fragmented content into structured, actionable paths. This is precisely the domain where Xiaohongshu's community content boasts its deepest reservoir of knowledge and experience.

Mao Wenchao, a key figure at Xiaohongshu, once remarked that the platform's crowning glory lies in bringing the life experiences and expertise of millions onto the internet. In the age of AI, the crux lies in whether these vast reservoirs of content can be transformed into efficient, trustworthy decision-making aids. This transformation will directly dictate whether Xiaohongshu can solidify its commercial footing as the go-to 'consumer decision-making gateway.'

36Kr once highlighted in an article that during an internal community search team meeting in 2024, a member of the algorithm team posed a pertinent question: Will AI-powered search supplant traditional search methods? Xiaohui, the leader of the community search algorithm team, responded at the time, "It's worth exploring, but it won't be my primary objective (O1 or O2, where O stands for Objective in the OKR framework)."

Now, it's incumbent upon Xiaohongshu's senior leadership to ponder and address this very question.