--- title: "Three-Color Risk Matrix and Seven Actions: How Clean-2026 Characterizes GEO as \"Data Poisoning\"" url: https://www.nine-health.cn/en/insights/tricolor-risk-matrix-actions/ lang: en-US type: Article category: "White Paper Deep Read" author: "Nine Health AI Digitalization Team" authorRole: "GEO Methodology Research" datePublished: 2026-10-11 dateModified: 2026-10-11 publisher: "和众利健科技（深圳）有限公司" --- # Three-Color Risk Matrix and Seven Actions: How Clean-2026 Characterizes GEO as "Data Poisoning" > **TL;DR** The white paper grades GEO content risk in red/orange/yellow: red (prescription to public, fake experts, bulk fake corpus, competitor confusion, unlabeled AI) banned; orange (science+commerce co-present, unregistered KOL, disease-tag push) controlled; yellow (official data corpus, structured FAQ, signed-reviewed science) allowed. Seven actions + a "provable compliance effort" evidence chain are key mitigating factors. ## Risk grading makes compliance operable {#overview} Compliance fears "one-size-fits-all" that either freezes action or hides line-crossing. Chapter 3 grades GEO content risk in three colors and gives seven enterprise actions, turning compliance from slogan into checklist. This article excerpts 3.11/3.12/3.15/3.16. ## Excerpt from the white paper {#wp-excerpt} Three-color risk matrix — RED (banned): prescription drugs in public AI answers; fake expert/patient endorsements; bulk fake corpus polluting models; competitor-trademark confusion and disparagement; unlabeled AI health science. ORANGE (controlled): science and purchase link on same page; unregistered doctor KOL; disease-tag precise push; AI free-rewrite of supplement claims. YELLOW (allowed): authoritative corpus from official open data; structured FAQ; expert-signed, review-logged science. Seven enterprise actions: ①GEO compliance review SOP; ②AI labeling retrofit (GB 45438 dual label); ③function-claim lock (model-uneditable zone); ④corpus asset ledger (source/review/approval/reviewer/date); ⑤partner due diligence (doctor filing/MCN record/platform qualification); ⑥competitor-trademark red line (no competitor marks, no disparaging comparison); ⑦sensitive-PI isolation (health info physically isolated from marketing unless separate consent). Four enforcement patterns (2024–2026): ①expanded liability — advertiser/publisher/MCN/platform jointly liable; ②technical elements as violations — "used AI" / "labeled or not" become independent violation facts; ③rising penalties — AI digital-human and fake-image cases 150k–1.2M RMB, likely 100k–1M fixed given uncomputable GEO ad cost; ④concentrated categories — supplements, medical aesthetics, prescription drugs. Clean-2026 is the first national regulatory file to name "GEO" and characterize it as "AI data poisoning". Understand in three layers: ①distinguish behavior not technology — it opposes malicious marketing (fake facts, bulk pollution, induced false conclusions), compliant GEO is not violating; ②"data poisoning" liability echoes the Anti-Unfair-Competition Law and false-advertising, reaching administrative even criminal; ③build a "provable compliance effort" chain (corpus ledger, AI-labeling logs, review logs) to prove "non-malicious" in investigations. ## Deep interpretation: from "don't cross" to "provable innocence" {#deep-dive} First, the matrix's value is making "can it be published" a decidable question. Red is hard-ban, orange is controlled gray, yellow is safe and should be scaled. Brands often mistake orange for yellow (e.g., casually attaching a purchase link to science) or yellow for red (e.g., fearing structured FAQ that should be encouraged). The matrix justifies resource allocation. Second, the most overlooked of the seven are ④corpus ledger and ⑦sensitive-PI isolation. The ledger is the carrier of "provable compliance effort" — when content is cited and attributed to the enterprise, regulators review it as traditional advertising; the enterprise cannot plead "AI dynamically generated, un-pre-reviewable". The ledger proves reasonable review effort, a mitigating factor. Symptoms from health Q&A are sensitive PI and must be physically isolated from marketing. Third, technical elements as violations are a new variable. "Used AI" / "labeled or not" become independent violation facts, so compliance must reach technical detail (digital-human authorization, label integrity), not just copy slogans. This raises the GEO compliance bar and widens the gap between "serious" and "sloppy". Fourth, Clean-2026 is not GEO's "end" but its "normalization". The white paper judges: enterprises that build a compliance paradigm first within the window gain both regulatory goodwill and AI-entry dividends. Documenting and routinizing compliance is itself risk hedging and a competitive moat. Nine Health's framework: use the three-color matrix as a content-admission gate; map the seven actions to org duties (legal reviews SOP, tech retrofits labels, marketing locks claims, data builds ledger, biz does due diligence); use AI monitoring to put "own negative statements" into a zero-tolerance loop (locate source → fix → request refresh → log). ## Get the full white paper {#get-full} 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. ## Sources - Nine Health white paper (2026), Chapter 3 (3.11/3.12/3.15/3.16): three-color matrix / seven actions / enforcement / Clean-2026 - CAC "Clean 2026" campaign against AI-application chaos (2026-04-30 launch) - SAMR 1H2025 medical-ad violation report (1,666 cases, 12.78M RMB penalty) ## FAQ ### How is the three-color matrix graded? RED (banned): prescription to public, fake experts, bulk fake corpus, competitor confusion, unlabeled AI science. ORANGE (controlled): science+commerce co-present, unregistered doctor KOL, disease-tag push, AI free-rewrite of supplement claims. YELLOW (allowed): official-data corpus, structured FAQ, signed-reviewed science. ### Which of the seven actions is most overlooked? The corpus asset ledger (source/review/approval/reviewer/date) and sensitive-PI isolation. The ledger carries "provable compliance effort"; in investigations it proves reasonable review and mitigates penalties. ### What does "technical elements as violations" mean? Since Clean-2026, "used AI" / "labeled or not" are independent violation facts, not just method descriptions. Compliance must reach technical detail (digital-human authorization, label integrity), not just slogans. ### Clean-2026 calls GEO data poisoning — can we still do GEO? Yes. Regulation opposes malicious marketing (fake facts, bulk pollution, induced false conclusions). Compliant GEO on real authoritative content is not violating and fits regulators' expectation of authoritative supply. ### Is the brand liable if AI mis-cites its product? Yes. Content cited and attributed to the enterprise is reviewed as traditional advertising. Build a fact-accuracy ledger loop: locate source, fix, request refresh, log. In healthcare this is zero-tolerance. ### What enforcement reference does the full white paper add? Chapter 3 includes 2024–2026典型案例 (Miaogujintie, Moni prescription, AI face-swap), PIPL/data-security constraints, and regulatory trend outlook (2H2026–2027). --- Source: [Nine Health](https://www.nine-health.cn/en/insights/tricolor-risk-matrix-actions/) · Author: Nine Health AI Digitalization Team（GEO Methodology Research） · Publisher: 和众利健科技（深圳）有限公司