China’s autumn mobile-internet report landed this week, and the search line is the one non-Google SEOs should not shrug past. QuestMobile’s 2026 China Mobile Internet Autumn Report, published on 3 November and picked up by Jiemian the same day, says the search-engine category lost 984 million hours of monthly usage in September versus a year earlier. On the same chart, AIGC apps gained 978 million hours. Browsers lost 274 million hours. General information platforms lost 1.307 billion. Short drama and short video together contributed more than the entire net growth in user time.
That near-mirror between search’s loss and AIGC’s gain will be quoted as “users moved from Baidu to chatbots”. QuestMobile’s own data does not prove a one-for-one swap — industry totals do not track individual journeys — but it does show that the information-acquisition stack built around search and feeds is losing share of attention while AI-native apps are taking it. In August, QuestMobile also recorded a 21.7% year-on-year drop in monthly uses per user for search-engine apps, against a 67.1% rise for AI-native apps. Baidu, Meituan and Kuaishou all saw company-level user scale fall year on year in the same report window; ByteDance’s top three apps flipped from Douyin / Toutiao / Douyin Lite to Douyin / Hongguo short drama / Doubao.
Why the agent price war suddenly makes sense
This is the demand backdrop for the China agent-search API scramble we have been covering for a fortnight. On 20 October Tencent opened Yuanbao Search on the Sogou index. On 30 October Baidu cut its Enhanced Search AI API to a promotional ¥4.9 per 1,000 calls. Alibaba’s Bailian and ByteDance’s Volcano Ark have already wired networked search into model calls. A 5 November analysis on Jiemian (also carried in English via 36Kr) frames the shift cleanly: the buyer of search is no longer only a human at a results page. It is increasingly a model or agent procuring materials on someone’s behalf.
That changes the bargaining position. The assistant owns the user and the completed task. The search API often sees only a sub-query. Human time on search apps can fall while machine retrieval volume rises — or both can fall if agents lean on closed knowledge bases. QuestMobile cannot tell us which path China is on yet. It can tell us the human attention half of the ledger is already red.
What SEOs and publishers should do with this
Do not read −984 million hours as “Baidu SEO is dead”. Read it as a second scoreboard. Organic and paid visibility inside Baidu, Shenma and Sogou still matter for the humans who remain, and for the snippets agents ground on. But if your China plan still measures success only in human sessions from the search box, you are scoring last year’s game. Watch three series into Q4: QuestMobile’s next search-category hours print; Baidu’s own AI-search API call volumes once the ¥4.9 promotion runs (it caps at a million calls and ends 31 December); and whether Tencent’s WorkBuddy / ima stack discloses how much Yuanbao Search actually fires per completed office task. For site owners, the parallel to the Russian LiveInternet drop is obvious: human click curves are no longer a safe proxy for “being found”.
My take
The interesting number is not that AIGC gained almost exactly what search lost. It is that Baidu is pricing retrieval for machines in the same month humans are spending nearly a billion fewer hours a month in search apps. That is either the start of a wholesale channel that replaces the lost human time, or a race to the bottom while the real audience sits inside Doubao and Qwen. Primary sources: QuestMobile’s autumn report page, Jiemian’s 3 November traffic piece, and the 5 November “second war of search” analysis. Annotate September 2026 on every China search dashboard you still trust.