Direct answer
PelvicEx uses AI to assist research organization, claim outlining, drafting, editing, localization, code generation, and automated validation. AI output is never treated as a source, author, clinician, or professional reviewer. Each clinical page must trace substantive claims to eligible evidence, use original prose, state uncertainty, and preserve safety boundaries. Automated checks support consistency, but they cannot diagnose, prescribe, or establish that content suits a particular person.
This policy describes content production and the limits of automated assistance. It does not describe user-specific medical decision making.
Where AI may assist
AI may help inventory roadmap topics, identify overlap, structure a source-to-claim outline, compare metadata, produce an initial draft, adapt tone, generate original abstract SVG code, propose related links, and check whether required sections exist. It may also help flag duplicate passages, unsupported product claims, restricted source identifiers, stale translations, and broken routes.
These uses are workflow accelerators. The system can still misunderstand a source, flatten disagreement, invent a citation, overgeneralize a population, or produce fluent but unsafe advice. Passing an automated test means only that the tested rule passed; it is not evidence of clinical correctness.
Source eligibility comes first
Before AI-assisted synthesis, a source receives a usage classification. Open-synthesis material may support ordinary original synthesis under its terms and public accessibility. Research-only material may inform fact checking from a lawfully available private copy, while public citations point to the official publisher or bibliographic record. Bibliographic-only records support identification but not substantive extraction. Excluded-AI sources are not used for AI-assisted content when terms or permission do not allow it.
Uncertainty defaults to exclusion. A prestigious guideline is not automatically eligible. PelvicEx should prefer legitimate publisher, DOI, PubMed, government, or professional-organization pages and should not route readers to unofficial ebook hosts. Attached commercial books remain private research material; they are never committed, redistributed, or linked from an article.
Draft from claims, not passages
The workflow begins with a fact outline that records the intended population, claim, source, evidence boundary, safety implication, and content level. Drafting then starts from that outline rather than rewriting source paragraphs sentence by sentence. This reduces the risk of close paraphrase and makes unsupported transitions easier to see.
PelvicEx uses no copied source figures, screenshots, tables, exercise diagrams, or long quotations. Illustrations are repository-owned abstractions or original simplified schematics that must not trace a commercial source. An originality audit compares pages with one another; source-level review is still needed because inter-page uniqueness cannot detect close copying from a book.
Citation and factual checks
Every evidence identifier in frontmatter must resolve to the source catalog. Reader-facing citations should appear near substantive claim clusters, and a complete Sources section should provide a legitimate public destination. Clinical articles normally combine a relevant textbook or foundational record with a current guideline, consensus document, systematic review, or primary study when each materially improves the page.
An editor or fact checker should ask whether the cited source studied the same population, intervention, comparator, outcome, and setting; whether the design can support the wording; and whether newer evidence changes the conclusion. The official record for Greenhalgh and Dijkstra’s evidence-appraisal text informs that discipline without licensing copied prose.
Medical and safety boundaries
AI-assisted pages cannot diagnose weakness, overactivity, prolapse, infection, retention, nerve dysfunction, or the cause of pain or leakage. They should avoid universal dose, treatment clearance, and outcome promises. Specialist and safety topics need explicit stop rules, scope limits, and routes to appropriate professional or urgent care.
The project does not ask an AI system to approve a postoperative plan, interpret an individual symptom history, or override a clinician. A page can describe general evidence while remaining unsuitable as a self-guided programme for a particular person.
Human accountability and disclosure
The ICMJE AI publishing recommendations explain why AI cannot accept authorship responsibility. PelvicEx therefore does not attribute a guide to AI as though it were a person or present machine assistance as human authorship or clinical authority.
The WHO ethics and governance guidance for AI in health highlights transparency, accountability, inclusion, and risk governance. PelvicEx applies those principles conservatively: disclose material AI use, protect sensitive information, test for accessibility and localization failures, record evidence, and make limits visible to readers.
Localization and change control
AI may produce a first localized adaptation, but translation is not accepted as a mechanical word swap. Technical terms, reading time, metadata, inclusive wording, caution, source links, and care-routing language must be checked in context. When English meaning changes, each affected translation is adapted and checked again before publication.
Source hashes connect each translation to the English version it reflects. They do not prove linguistic or clinical quality. Arabic directionality, long titles, script rendering, and locale-specific navigation require additional technical checks, while professional claims must remain identical in strength across languages.
Sources
- International Committee of Medical Journal Editors: Use of Artificial Intelligence in Publishing.
- World Health Organization: Ethics and Governance of Artificial Intelligence for Health.
- Greenhalgh and Dijkstra: How to Read a Paper, official Wiley record.
- PelvicEx evidence catalog, production manifests, validation scripts, and guide metadata, inspected 1 September 2026. Internal first-party disclosure.