AI use policy

01 · Current disclosure

The content and product workflow is AI-assisted

AI tools have assisted source discovery, structured extraction, drafting, terminology organization, consistency checks, software development and test preparation. “China Care Desk Research Desk” describes that organizational workflow; it is not a named person or licensed professional.

02 · Permitted uses

Use AI for structured assistance, not final authority

• Generate bilingual search terms and locate candidate government, provider, insurer and regulator pages.
• Organize source notes into consistent fields such as publisher, URL, access date, jurisdiction and scope.
• Prepare plain-English drafts, check internal consistency and identify questions needing evidence.
• Assist terminology comparison while preserving useful Chinese names and flagging uncertain translation.
• Support code, accessibility checks, search structure, metadata and automated test preparation.

Candidate information produced by AI must be traced to an independent source. The AI response itself is not cited as evidence.

03 · Prohibited uses

AI must not make high-impact decisions for a reader

• Diagnose a condition, assess personal urgency, interpret a result, select treatment, recommend a medicine or determine a dose.
• Determine legal rights, liability, deadlines, evidence weight, compensation, settlement terms or litigation strategy.
• Invent or alter sources, quotations, statistics, provider services, reviewer identities, qualifications or approval records.
• Turn a provider's marketing claim into an independent quality ranking or guaranteed outcome.
• Automatically publish clinical, safety-sensitive or high-impact legal content without the required review and publication record.
• Use identifiable patient records, passport data, insurance identifiers or private case files as routine prompt material.

04 · Verification controls

Check the source, scope, date and meaning

The intended control sequence is: locate the responsible source, open the original page, record its publisher and date, extract the relevant claim, check jurisdiction and exceptions, compare conflicting material, and label what still requires direct confirmation.

Numbers, deadlines, eligibility rules, emergency routes, medicine terms, legal conclusions and clinical statements require heightened checking. If the controlling source cannot be found or the conflict cannot be resolved, the draft should say so, narrow the claim or remain unpublished.

The current corpus commonly has page-level source lists rather than complete inline claim-level citation coverage. That limitation is disclosed in the research methodology and remains part of ongoing source-improvement work.

05 · Human oversight and current gaps

A named human record is required where a gate calls for one

AI cannot satisfy a requirement for an accountable editor, clinician, bilingual medical editor or lawyer. A future review record should identify the person, relevant qualification, review scope, date, conflicts and material limitations.

Clinical and safety-sensitive material must not be described as professionally reviewed unless an appropriate current review record exists. Legal-process information remains general source-led information, not legal advice. Terms, privacy and governance documents are reviewed as the service and its capabilities change.

06 · Reader data and privacy

The current website has no reader-facing AI tool

China Care Desk does not currently offer a chatbot, symptom checker, document analysis tool, personalized recommendation engine or health-record upload. The application therefore does not intentionally send reader questions or medical files to an AI model.

Readers should not enter names, passport numbers, insurance numbers, diagnoses or medical records into site search. Internal research prompts should use public information and synthetic examples, not identifiable patient or private case data.

Before any production AI vendor or reader-facing feature is enabled, its product and privacy review must document purpose, necessity, vendor, data flow, storage, retention, access, security, model-use terms, applicable consent and cross-border or China-law implications.

07 · Future AI features

A chatbot would require a separate product and safety review

This policy does not authorize a future conversational medical assistant. Any reader-facing AI feature would need a defined low-risk scope, emergency handling, privacy assessment, age and vulnerability safeguards, professional evaluation, monitoring, rate limits, abuse controls, clear uncertainty language and a tested route to human or emergency help.

It must not present itself as a doctor, lawyer, hospital, insurer or government service, and it must not personalize diagnosis, treatment or legal strategy.

08 · Errors and accountability

Published material remains subject to source-based correction

“The AI said it” is not an acceptable explanation for an error. When an official source contradicts a published statement, the affected content should be corrected or withdrawn and its review record updated. AI-related errors follow the same priority as any other error, with possible immediate harm handled first.

Policy reviewed: 17 July 2026