Source Engineering: L0–L5 Authority Hierarchy, Empirical Data, and Source KPIs

The source tier decides citation fate

If Chapter 2 is "how the engine reads", Chapter 4 is "content must stand in the right tier". In the generative era, whether a brand is cited largely depends on which L0–L5 tier its content occupies. This article excerpts 4.1/4.2/4.3/4.9: the hierarchy, empirical data and operable KPIs.

Excerpt from the white paper

L0–L5 China healthcare authority hierarchy: L0 national official (NHC/NMPA, extremely high citation, passive only, not co-buildable); L1 societies/guidelines (CMA/CSCO, high, partly co-buildable); L2 core/databases (Cnki/PubMed, high but paywalled); L3 authoritative platforms (Tencent Medical/Baidu Health/Dingxiang/MSD Manual, high and operable, co-buildable); L4 pro media/think tanks (DXY/Yimaitong/IQVIA, mid-high); L5 wiki/UGC (Baidu Baike/Zhihu/Xiaohongshu/Douyin, high but uneven, highest risk).

Key insight: L3 is the only "high-weight and actively co-buildable" tier outside L0–L2 — MSD Manual and CMIA are observed in DeepSeek Top10. Tencent Medical entry customization and Baidu Health "Starfish" co-build are clear L3 entry paths for pharma.

Xinbang AI Source Study (period 2025-12, 1.429M Q&A, 16.836M source URLs) (secondary): 3 of Doubao Top5 are ByteDance; Yuanbao ~10% dependent on WeChat; DeepSeek prefers Sohu/Baidu Baike/NetEase; cross-platform universal sources are Sohu/NetEase/Toutiao; 60% of Doubao Top10 are mobile domains.

Source cadence: weekly 2–3 science posts on L5 owned channels for freshness; monthly check L3 co-built entries (refresh lastReviewed); quarterly assess L0–L5 structure and fill weak tiers; semi-annually full review and retire stale content. KPIs: L1–L3 share among cited sources (target ≥60%), brand AI mention trend, fact accuracy (100%), entity-consistency deviations (target 0).

Deep interpretation: from "publishing content" to "building tiers"

First, L3 is the brand's most realistic lever. L0 is not co-buildable, L1 needs academic门槛, L2 is paywalled; only L3 combines "high weight" and "operable" — Tencent Medical customization and Baidu Health "Starfish" (100k doctors co-building 100M AIGC science entries) are clear entries. But the threshold is "build qualification first": a certified brand account is the ticket to video sources; without it, no amount of short videos enters the compliant pool.

Second, cross-platform universal sources are an underrated play. Xinbang shows Sohu/NetEase/Toutiao highly cited across all four AIs; such "high-weight community/wiki/short-video" sources lower AI comprehension cost via clear structure and high information density, hence preferred. Brands can ride universal sources beyond owned channels to raise citation probability without grinding a single ecosystem.

Third, L5 is a double-edged sword: lowest cost but highest risk. Unqualified professional science is explicitly banned, and AI "reads subtitles not pictures" — short videos must yield machine-extractable subtitle and conclusion text, or they go silent or get blocked in mobile quick-answer scenes. L5 needs separate admission rules (qualification + subtitle + scene).

Fourth, KPIs make source operation concrete. The white paper gives four quantifiable metrics, core being "L1–L3 share ≥60%" and "fact accuracy 100%" — the former measures source-structure health, the latter is the lifeline of healthcare. Run source operation on a weekly/monthly/quarterly/semi-annual cadence, not as a sprint.

Nine Health's framework: three-layer source operation — owned anchor (knowledge base + certified accounts), authority co-build (Tencent Medical / Baidu Health Starfish / L1 submissions, into L3), ecosystem distribution (per-ecosystem certified accounts + universal sources). Pair with source monitoring of "competitor voice in AI answers", whose purpose is always "strengthen own authority", never disparage rivals.

Get the full white paper

This article excerpts chapters from the Nine Health white paper "Compliance as Growth: GEO Compliance & Growth in Healthcare" (2026 edition). The full edition covers the GEO tech stack, the compliance foundation, source engineering, the E-E-A-T trust system, and the eight capability-building modules, with a compliance self-check list, a three-color risk matrix, and a phased roadmap.

To request the full white paper or a one-on-one brand GEO diagnostic, contact Nine Health: phone 18126259547 / WeCom (see "Contact Us" on the site) / email jianyi@nine-health.com. Our team responds within 24 hours.