China already finished the SEO-to-GEO war. Here’s the after-photo.
One market already lived through the whole AI-search transition the West is still arguing about. Here’s what it looks like in numbers — every figure attributed.
Why we watch China for a living now
Our Policy Radar recently started watching Chinese first-party sources — Baidu’s webmaster docs, ERNIE’s changelogs, ByteDance’s crawler pages. Not because we sell in China. Because China is the only major market where the AI-search war is essentially OVER, and the after-photo tells the rest of us what’s coming.
Think about their starting conditions: no Google to defend old habits, a billion mobile-first users, and domestic AI apps pushed by the biggest platforms on earth. China didn’t transition to AI search gently. It jumped.
The after-photo, in numbers (all attributed)
- Industry analysis in 2026 reports that around 70% of Chinese search queries now end zero-click (ChinaSEO industry reporting, 2026; no public primary link).
- AI agents — Doubao, ERNIE Bot, DeepSeek, Qwen, Kimi — accounted for roughly 34% of product discovery queries by April 2026, up from 8% in 2024 (The Silicon Review, May 2026; no public primary link).
- Baidu’s ERNIE-generated answers sit on top of results for a large share of informational queries — reported at ~40% of queries answered directly, with organic click-through down roughly 18% in one quarter (Baidu investor-relations data cited by SEO agencies working in China; no public primary link).
- ByteDance’s Doubao is reported around a quarter-billion monthly users; Baidu embeds ERNIE directly into an app with 700M+ monthly users, one tap from classic search (KrASIA, February 2026).
Read that list twice. Every anxiety the Western SEO industry has about 2027 — China’s market already metabolized it.
What the winners there actually did
The case that should be pinned above every marketer’s desk: a beauty brand’s plain FAQ page became ERNIE Bot’s primary cited source for skincare-routine answers — reported 35% sales lift, zero extra ad spend (The Silicon Review, May 2026; no public primary link). Not a campaign. A page of honest answers, structured so a machine could lift them.
And the practice that grew around wins like that has a name we’re formally stealing for the glossary: entity seeding. Stop obsessing over what people type; plant consistent, verifiable facts about who you are — on your site, in your structured data, across surfaces the AI trusts — so when someone asks an agent “which clinic is gentle with anxious kids?”, the machine can confidently say YOUR name. Chinese practitioners treat keywords as the old war and the entity as the new one. They are right.
The three lessons worth importing today
- Answer pages are revenue infrastructure now. The FAQ-to-sales case is the whole AEO argument with a profit figure attached. If your best answers live in a salesperson’s head instead of on a liftable page, you are invisible to the channel that’s coming.
- Entity consistency beats content volume. Chinese GEO practice obsesses over the same thing our lockout and capsule studies measure: can machines verify who you are, everywhere, identically? One contradiction and the agent names your competitor.
- The platform answer-layer eats first. When ERNIE answers on the page, the blue links below starve — exactly the CTR collapse now measured in Japan and arriving everywhere. The defense isn’t ranking harder. It’s being the source the answer quotes.
The honest footnotes
China’s numbers travel with caveats: metrics come from trackers and industry reports with ranges, not audited registries; zero-click definitions vary; and China’s walled ecosystem means some dynamics (super-app search, platform content quotas) won’t map one-to-one onto open-web markets. We cite ranges and sources so you can judge. And a linkability confession: the KrASIA figures above are linked to the article itself; the Silicon Review, ChinaSEO and Baidu investor-relations numbers circulate in industry reporting without a public primary dataset we could link — treat those as directional signals, not audit-grade data. If a primary source surfaces, we will link it here. What doesn’t need a caveat: the direction. The agent ate the query box there first. It is eating everywhere.
— The SynapseIN Research Desk · edited & approved by Manas Panda
Honesty notice: All statistics attributed inline to their published sources (industry reporting 2026); none are SynapseIN measurements. Our related first-party data: the AI Lockout study and the Machine Time Capsule, raw data public. The Entity Seeding definition lives in our glossary — etymology honestly credited to Chinese GEO practice. Where no public primary link exists, the honest footnotes above say so.
More from the bench: What is GEO? · GEO entity engineering · The AI Lockout study · Glossary: entity seeding
Questions about the after-photo
What share of product discovery do AI agents have in China?
About 34% by April 2026, up from 8% in 2024, per industry reporting (The Silicon Review, May 2026). No public primary dataset stands behind that figure; we flag it as directional, not audit-grade.
What is entity seeding?
The practice of planting consistent, verifiable facts about who you are — on your site, in your structured data, across surfaces AI agents trust — so a machine can confidently say your name. The term is credited to Chinese GEO practice.
Does China’s experience apply to Western markets?
The direction, yes; the specifics, partially. Walled-garden dynamics (super-app search, platform content quotas) don’t map one-to-one onto open-web markets, but the agent-first shift does.
Want to know what a machine can say about you?
Every Chinese winner started with the same uncreative first step: a radically honest look at what an AI agent can verify — and quote — about the business today. That lens is free.