Was That Restaurant Influencer Ever a Person — or Ever in the Restaurant? The Take Inside China

Late August 2026 trending discussion in China: batches of AI “digital human” restaurant-review videos produced by MCN agencies that never visited the venues, fabricated praise, and left merchants fielding the disappointment gap when customers showed up. Our WeChat column’s take ran under the headline “The restaurant blogger you follow may never have entered that restaurant.” This entry translates it and pins its enforcement claims to primary sources.

The numbers

• 169 influencers recruited by Shanghai Dexin Hospital; ¥200,000 fine (SAMR third-batch typical cases, 2025-09-23) — verified against the official release
• 186 fake-experience videos posted [unverified — count from Chinese coverage; the official summary names the 169 recruiters’ arrangement, not the video count]
• Hangzhou Intermediate Court: first national unfair-competition ruling against AI-written “seeding notes” — co-defendants jointly liable, ¥100,000 damages (Zhejiang Online/新华 coverage)
• <1 minute to generate one storefront image with the Jimeng tool (reporter’s test as cited)
• Ele.me cleanup: 31,000+ violating merchants retired; JD Takeaway approval pass rate 40% vs industry 70–90% (as cited from Zhengzhou regulatory meetings)

The take inside China

The essay is built as a pipeline teardown in four movements.

One nonexistent blogger. 乔安 posted daily-life content and restaurant recommendations until readers noticed every photo wore the same face at the same angle with the same smile and lighting that didn’t behave like light. The account then stamped “AI-generated content” on some images and admitted the person was entirely synthetic — the transparency was extracted by eyeballs, not volunteered. The detail the take pauses on: the restaurant was real, the table was real, the window light was real; only the person living there was not.

The batch factory. Below the individual case, the industry: merchants send in venue photos and footage; operators have AI write the full review — texture of the dishes, atmosphere of the room, personal feelings included; users order through the link; the account takes commission. Three things that used to require physical presence (visiting, tasting, feeling) now require none. On the image side: AI storefronts flooding delivery platforms with telltale uniformity, an agency gray market offering “one-take walkthrough videos” for shops with no location and brand-attachment schemes to pass review — when verification requires video, the intermediaries sell video.

The three-loss chain. Consumers buy fabricated experience and the discrepancy lands in their stomachs; honest creators producing two or three tested reviews a day compete with a species that never sleeps (“it’s not that bad money won — it’s that good money tired first”); honest merchants eat the review-section blame for a script they never hired. The take’s counterweight: digital humans doing labeled livestreams are legitimate cost reduction — its rule is the tool is neutral, impersonation is not.

The closing gap. Regulatory response is real but slow-motion against generative speed: the Hangzhou ruling names AI-written seeding notes as unfair competition; Shanghai’s ¥200K fine prices fabricated experience; platforms retire tens of thousands of merchants. The take’s closing image: the conveyor belt runs faster than the verdicts.

What the Chinese take left out

The 186-video count and the JD Takeaway approval-rate figure are sourced to secondary relay (Lei Technology via The Paper) and uncorroborated by primary releases in our check — tagged accordingly above. Also absent from the take: what PIPL and the Interim Provisions on Countering Unfair Competition Online would require of platforms hosting this content at intake, rather than after the fact.

Why it matters outside

This is the food-and-retail instance of a general economics: generation cost for “looks authentic” has hit the floor while verification cost stays on the ceiling. Every trust system that runs on photos-plus-first-person-voice — reviews, testimonials, UGC marketing, even “citizen footage” — inherits the same asymmetry, and the Chinese enforcement ledger (court ruling + ¥200K fine + platform purges) is currently the most documented public experiment in pricing it.

Sources

• Zhejiang Online: When ‘seeding notes’ are AI-generated (乔安 case, Hangzhou ruling)
• Shanghai regulator relaying SAMR typical cases: Dexin Hospital fined ¥200K
• People’s Daily Online: AI writers cloning seeding notes

Provenance & disclosure. Originally published in Chinese on our WeChat channel on 2026-08-28 (“你关注的探店博主，可能从来没进过那家店？”); drafted with AI assistance under human editorial direction. Translated to English on 2026-08-29 (AI-assisted, human-reviewed). The Dexin penalty (169 influencers, ¥200K) and the Hangzhou ruling were verified against official/regulator-relayed sources; the 186-video count and platform approval-rate figures remain secondary-sourced and carry [unverified] tags. This is translated commentary — not a SigPulse measurement. Our first-party measurements live in the dispatches and the /data/ ledger.