BEIJING — For fifteen years, Zhihu taught the world that its value was answering questions. The company’s second-quarter 2026 earnings, filed in late August, told a different story. Paid content and IP operations accounted for 61.7% of revenue, while marketing services — long the company’s largest and most dignified business — slipped to 28.8%. The most revealing detail was not a number but a name. A serialized-fiction writer called Jiusen sold seven story copyrights in May alone and ten in the first half of the year; his novel, adapted into the short drama The Wrong Marriage Brings Joy, passed three billion views. A knowledge community built on asking and answering had quietly become a company that sells stories. The company has been laying the groundwork for years: it launched Zhihu Zhida, an AI search product over its community content, back in 2024, and later wired it into DeepSeek’s R1 model.
The advertising collapse that forced the pivot was real and steep. Marketing services brought in 1.653 billion yuan in 2023, 1.247 billion in 2024, and 844 million in 2025 — nearly halved in two years — before falling to 199 million yuan in the second quarter of 2026. Total revenue peaked at 4.199 billion yuan in 2023 and shrank to 2.749 billion by 2025, and the company stopped disclosing monthly active users in 2025, swapping user-scale for content-and-payment as its narrative. Even so, the second quarter brought a tentative bottom: revenue of 690 million yuan, up 5.9% from the prior quarter, and an adjusted net profit of 6.9 million yuan for the first half of 2026 — a second straight stretch in the black. Advertising, as Zhihu’s own executives conceded on the earnings call, had become a structural problem rather than a cyclical one. The reason is the shape of the attention itself: users come to Zhihu to search for an answer and leave, averaging 41 minutes a day against Bilibili’s 113, and attention-pricing pays by the scroll, not the query. Zhihu tried the scale game once, burning past 2 billion yuan a year on sales and marketing to buy traffic it could not keep, then pulled back to 1.25 billion by 2025. It also pruned low-quality commercial content, cutting its share of recommended-feed exposure by more than forty percent from the start of 2024.
One Asset, Four Paydays
The paid-content engine moved in the opposite direction. Yan Yan Story grew author copyright income 620% year over year, pushed cumulative copyright sales past one thousand titles, and reported second-quarter deal volume up 105% from the prior quarter, according to figures the unit released alongside the earnings. Zhihu had worked out an arithmetic its archives never allowed before: a story can be sold once as a subscription, again as a short-drama adaptation, again as a comic or AI-comic adaptation, and again as a film or television license. One asset, four paydays. The company’s IP arm, paired with Tinghuadao Studio, shipped seven premium short dramas and three AI comics — each topping 900 million views — and Yan Yan ranked among the top IP suppliers for AI comics on TikTok’s domestic market and in the top three on the Hongguo platform in the first half of 2026. The most prolific writer on the platform, Mango Jia La, sold eighteen copyrights in six months. For the writers themselves, the economics flipped almost overnight: a single author could now clear hundreds of thousands of yuan in a half-year from licensing alone, a career path that did not exist in the community’s first decade.
Owning the Inventory
The deeper logic is a shift in how internet content gets priced. Advertising prices attention: you are paid by the exposure, the click, the impression, which is why a search-and-leave platform could never out-earn one engineered for scrolling. Content assets price knowledge, and the same material can be resold across formats and markets. A recruitment specialist at SunTzu Recruit described the change as structural, not cyclical. “Zhihu stopped renting its traffic and started owning its inventory,” the specialist said. The inventory is vast — 993 million pieces of content from 81.5 million creators — and Zhihu is now selling it twice over. Its expert-data arm, SoTALab, has assembled more than 200,000 credentialed specialists, from doctors and lawyers to investment-bank analysts and engineers, who build training data, benchmarks, and reasoning traces for large language models. The comparison the company reaches for is Reddit, which sells user-discussion data to AI companies under a deal worth roughly $60 million a year with Google alone. Zhihu’s bet is that expert data is scarcer, and worth far more, than discussion data — the difference between a general annotator paid a few dollars per item and a PhD-level specialist paid hundreds.
A New Hiring Bill
Here’s the thing: a company that changes what it sells has to change who it hires. Zhihu’s drift from an advertising-and-sales organization toward a content-asset organization is quietly rewriting its recruiting needs — and those of every platform watching it. The roles in shortest supply are content-IP operators, short-drama and AI-comic producers, copyright managers, and AI-data specialists who understand both domain expertise and model evaluation. These are young jobs with almost no pre-existing talent pool in China’s cross-border content market. A senior consultant at SunTzu Recruit, who has run searches for content platforms, put the shortage bluntly: the ideal candidate has to read a balance sheet and a storyboard, and those two people are rarely the same person. SunTzu Recruitment has watched one Beijing-based content platform spend nine months trying to hire a content-IP operations director with overseas distribution experience; the cross-border requirement, the firm’s practice lead noted, eliminates most domestic candidates while the short-drama boom pulls the remainder toward production studios. A partner at SunTzu Recruitment who oversees the media and content practice observed that these searches now routinely run twice as long as a conventional operations hire. The three-way requirement — content judgment, copyright fluency, and model-evaluation literacy — is hard to find in a single resume, and the people who hold it are being pulled in three directions at once: the short-drama industry is bidding up directors and producers, while AI labs compete for the same scarce pool of domain experts who can both practice medicine and score a model’s output. That said, the squeeze is not uniform. Some platforms are buying their way around it, acquiring studios outright rather than hiring executives to build them, which shifts demand upstream toward M&A and content-strategy leaders instead.
The Second Act
The content-asset model is spreading beyond Zhihu, and with it a new executive-search category is taking shape. SunTzu Recruit expects the competition for content-IP operators, short-drama producers, and AI-data specialists to intensify as more internet platforms convert their archives into assets and push their international ambitions through cross-border distribution. The platforms that staff this transition early — hiring leaders who understand both intellectual property and model training — will be the ones that turn fifteen years of accumulated content into a genuine second act. The rest will keep renting attention at a price that keeps falling.
Sources: Zhihu Q2 2026 earnings and Yan Yan Story copyright figures (August 2026); Huxiu, “Zhihu is becoming a content-asset company” (August 2026); Zhihu 2023–2025 annual reports; Reddit’s reported data-licensing agreement with Google.
Comments are closed