There are two social media platforms where I seldom speak but frequently browse: one is the recently popular Xiaohongshu, and the other is Zhihu. On Zhihu, I most often indulge in literature and history content, followed by sports and language content (as I'm currently learning Japanese). Over the past two years, as large AI models gained traction, I read numerous literature reviews on Zhihu, achieving a good understanding of large model technology. Given that I spend dozens of hours on Zhihu every month, I am quite familiar with the platform and feel qualified to comment.

Certainly, Zhihu has its limitations, as any content platform does. However, Zhihu's irreplaceability lies in its high-quality professional content. I believe there are only two genuinely professional content platforms on the Chinese internet: WeChat Official Accounts and Zhihu. While not all content on Zhihu is "professional," in terms of the richness and depth of professional content, few platforms can compare.

Some argue that as the community expands, Zhihu has lost its initial "professional barriers" and become indistinguishable from other ordinary platforms. Is this really the case? Let's consider two recent examples: At the beginning of this year, Zhihu became the exclusive Chinese content platform for DeepSeek, releasing technical details and cost-benefit analyses of its open-source models first, thereby becoming a crucial source of "effective information" in the market. Recently, Ma Xiaochun, a master in the Chinese Go community, joined Zhihu, not only sharing extensively about his Go career but also commenting on current global Go trends. These instances clearly demonstrate that Zhihu still maintains deep barriers in the realm of top-tier professional content.

Besides top-tier content, Zhihu retains a strong UGC (User-Generated Content) flavor in broader professional areas: a plethora of professional users continuously create and share content on Zhihu, driven by passion rather than commercial motives. Taking the literature and history content I frequently browse as an example, most creators are "powered by love," with some having persisted for over a decade, contributing since Zhihu's inception! In my view, such a UGC atmosphere exists only in a handful of communities like Zhihu and Xiaohongshu. Some see this as indicative of Zhihu's insufficient commercial development, needing improvement; however, this is precisely why a large number of users, including myself, stay on Zhihu and are willing to absorb its content.

For a long time, the capital market has believed that content companies struggle to make money, with specialized content platforms facing an even greater challenge. Historically, Zhihu has often been in a state of loss, seemingly reinforcing the above view. However, this quarter witnessed significant changes, with Zhihu achieving double profits at both GAAP and Non-GAAP levels for the first time. While this is partly due to cost reduction and efficiency enhancement, a closer look at the financial data reveals a more nuanced picture:

Despite a substantial reduction in sales and marketing expenses, the number of paid users for the entire year still increased, indicating that a significant portion of Zhihu users have developed paid habits with high payment stickiness.

In 2024, the number of content creators on Zhihu increased by 8.9%, and the number of content creators earning income grew by 20%; the total number of content pieces rose by 12.9%. This shows that cost reduction measures did not dampen Zhihu users' enthusiasm for creation, and Zhihu's content ecosystem remains resilient and vibrant.

Compared to short-term financial figures, I am more intrigued by the long-term opportunities presented by the proliferation of AI technology. Since ChatGPT's emergence, there has been a misconception in the market: the popularization of generative AI heralds the end of "content," and both content creators and content platforms will see their value diminish as AI can produce and disseminate content more efficiently. As a heavy user of both ChatGPT and DeepSeek, the deeper I delve into AI, the more I realize the absurdity of this view! The revolutionary significance of generative AI does not lie in "replacing content creators"; rather, this technological revolution will elevate the significance and value of high-quality content platforms to unprecedented heights.

Over the past month, the topic that has garnered the most discussion in my social circle is "large model illusion": large models integrated with inferior information can produce garbage at an unprecedented rate, polluting the entire internet with "illusions." Only "large models + high-quality information" can enhance productivity and broaden human knowledge boundaries; "large models + inferior information" have the opposite effect. Meanwhile, due to vigilance against "large model illusions," general users are increasingly focused on verifying large model conclusions with professional and credible information. The result is that those who possess the most reliable information can maximize the utilization of generative AI technological advancements. In recent quarters, Zhihu has emphasized building a "trust-centered community ecosystem," which is undoubtedly Zhihu's self-positioning in the era of generative AI!

Since mid-February this year, Tencent Yuanbao has enjoyed immense popularity, primarily due to the integration of high-quality content from WeChat Official Accounts into DeepSeek, which is particularly crucial for vertical categories with strong professionalism such as finance and technology. In fact, Zhihu's AI search product "Zhihu Direct Answer" was integrated with DeepSeek V3/R1 as early as February 11, two days before Yuanbao. All internet platforms are vying to integrate with DeepSeek, but Zhihu's "professional + vertical/long tail" content ecosystem forms an irreplaceable competitive advantage, perhaps comparable only to the WeChat ecosystem across the entire Chinese internet. For instance, just yesterday, after a magnitude 4.2 earthquake struck Langfang, Hebei, I searched for topics related to earthquake early warning and risk avoidance on Zhihu Direct Answer for half a day, fully understanding the earthquake risks and response measures in North China - such content is hard to find on any other platform.

People often believed in the past that the "professionalism" and "popularity" of content were difficult to balance, with "too highbrow to appeal to many" being the norm. Taking Zhihu as an example, high-quality long answers often have certain reading thresholds, and there are often multiple professional answers with different perspectives but all having reference value under the same question; non-professional users may find this overwhelming. Generative AI excels at solving this kind of problem. Especially the recently increasingly popular deep reasoning models can efficiently summarize complex professional content and express it in a form understandable to general users. This could be another fundamental shift in the field of content dissemination, no less significant than the advent of the deep learning algorithm recommendation mechanism over a decade ago.

In the coming years, the path for the dissemination of high-quality content may evolve as follows: users with strong professionalism will continue to read the original text to obtain the most comprehensive first-hand information, much like how professional scholars today read their peers' full-text papers; general users will leverage the power of AI search to swiftly obtain the content they need. On Zhihu, this process is already progressing rapidly, with Zhihu Direct Answer's monthly active users growing from millions to tens of millions since its official launch in June last year.

The above discussions center around the content ecosystem level. Moreover, I believe the greatest significance of AI for Zhihu actually lies in the commercial realm. How can we truly unlock the commercial value of professional content? More colloquially, how can we make professional content more profitable? Strictly speaking, this problem has not been adequately solved in the internet era. Generative AI won't solve it overnight either, but we can already see a clear roadmap for the future.

Specifically for Zhihu, many overlook the fact that its advertising eCPM (effective cost per thousand impressions) has always been relatively high but has historically lacked sufficient advertising formats and inventory. AI search tools can provide efficient and precise user interaction scenarios, making them naturally suited for advertising. Currently, Zhihu Direct Answer has not yet begun loading advertisements, but I believe it's just a matter of time. Much like how Taobao opened up its commercialization ceiling years ago through its "direct train" search function, AI search is poised to become the breakthrough point for Zhihu's commercialization. Especially advertisers in industries with high unit prices and longer decision-making times, such as automobiles, home appliances, electronics, and tourism, will find such advertising scenarios very appealing.

Beyond advertising, AI-based user payments will become another breakthrough. Over the past few years, Zhihu has initially cultivated users' paid habits through paid reading, vocational training, and other methods. Among my friends, there are even over a dozen who have paid for Zhihu services. The expansion of AI functionality will further broaden Zhihu's user payment business landscape on this foundation. Specific business models can include paid consultations, customized content, and even novel forms of content; services can be provided to users by individual creators, AI, or a combination of both. As of now, the integration of generative AI and content communities is still in its nascent stages, and future commercialization methods will inevitably diversify. Over a decade ago, no one could have foreseen the immense commercialization scale of social media platforms based on Web2.0; similarly, at this juncture, we cannot predict the monetization models that content communities leveraging generative AI will devise in the future, but they will undoubtedly be significant.

Throughout 2024, Zhihu's operating revenue was approximately RMB 3.6 billion. Among the major content platforms on the Chinese internet, this revenue figure is relatively modest. Zhihu's single-user value (based on MAU) is about RMB 40+ annually, which is also not exceptionally high, inconsistent with Zhihu's content quality and actual influence. This certainly suggests that Zhihu's commercialization has hit bottlenecks so far, and its platform value has not been fully realized; however, it also implies that once generative AI is fully deployed, the resulting business elasticity for Zhihu may far exceed that of other content platforms - a true "turnaround" scenario.

Of course, this transformation won't happen spontaneously and requires concerted efforts from Zhihu's management team, as well as its R&D, product, and operations teams. I eagerly anticipate this transformation because I am also a professional creator. If Zhihu can comprehensively and substantially enhance the value of professional content through AI technology, millions, including myself, will be deeply encouraged. I hope this day arrives soon!