Zhihu (知乎) 2025: tightening belts and running AI story sessions

Profit returned. Revenue did not.

Zhihu (知乎) achieved a milestone in 2025 — it reportedly posted its first full-year adjusted (Non‑GAAP) profit, an adjusted net income of roughly RMB 37.9 million — but the headline masks a much tougher reality: full‑year revenue slid to RMB 2.75 billion, down 23.6% year‑on‑year. Advertising fell sharply (‑32.3%) and membership revenue was also weaker (‑12.7%); monthly paying subscribers averaged about 13.5 million, down 10%. The company has leaned hard into cost cuts — employee expenses were down ~35% and R&D spending fell ~28% — and a one‑off non‑operating investment gain (reportedly about RMB 230 million from a fair‑value writedown reversal) helped the bottom line. Can profitability be sustained if users and ad buyers keep evaporating?

Embracing AI — search, creator tools and more stories

Management frames 2025 as a “structural inflection” toward an AI‑driven knowledge platform. CEO Zhou Yuan said Zhihu will solidify core businesses while accelerating AI commercialisation; CFO Wang Han had earlier signalled the company was prepared to tolerate slightly wider losses in pursuit of AI opportunities. The push has been concrete: Zhihu integrated DeepSeek‑R1 into search and creator workflows, and it has been reported that the company is training models on a proprietary corpus (reportedly ~50 million bilingual documents and some 870 million Q&A items) to power “Zhihu Direct Answers,” creator assistance and AI‑generated long‑form fiction for its paid Yanxuan (盐选) channels. That strategy lowers production costs and boosts content supply, but it also risks further eroding community tone and quality — and Q4 operating expense rose 15% year‑on‑year as the company balanced cuts with renewed investment.

Competition, community and geopolitical context

Zhihu’s pivot comes under intense competitive pressure. Fanqie Novel (番茄小说), part of ByteDance (字节跳动), dwarfs rivals on scale and has an early lead in AI writing and multi‑modal IP pipelines; Xiaohongshu (小红书) is siphoning high‑value Zhihu answers into short‑form commerce posts. Meanwhile Zhihu has stopped disclosing MAU, so outsiders cannot easily judge whether AI features will drive genuine traffic growth. Against a broader backdrop of US export controls and heightened focus on domestic AI stacks, Chinese platforms are racing to build local AI capabilities and commercial models — but questions remain: will Zhihu become another AI‑novel mill, or can it preserve a community moat rooted in higher‑quality knowledge exchange while finding new, sustainable monetisation routes?