OH Advisor

AI Transparency Note

Version 1.2 · September 2026 · OH Specialists Hub Ltd · Replaces Version 1.1 (May 2026)

1. What is OH Advisor?

OH Advisor is an AI-powered occupational health guidance platform developed and operated by OH Specialists Hub Ltd. Its AI advisor, Susan, is purpose-built for UK-based occupational health practitioners and HR professionals. There are two products:

• OH Advisor HR — for HR professionals and line managers. Guidance on fit note interpretation, referral decisions, reasonable adjustments, relevant legislation and return-to-work processes.
• OH Advisor Clinical — for qualified occupational health practitioners (nurses, advisors and physicians). Clinical-depth guidance, report writing assistance, clinical note structuring, phased return planning and document generation.

OH Advisor is a guidance and drafting support tool. It does not make clinical decisions, diagnose conditions, prescribe treatment, or replace a qualified practitioner’s professional assessment.

2. Who Operates It?

OH Advisor is operated by OH Specialists Hub Ltd, a UK company. It was developed by its director, Craig Page RGN DipOH, a registered occupational health nurse with over 30 years of clinical experience, who holds a Diploma in Occupational Health and is registered with the Nursing and Midwifery Council (NMC).

OH Advisor was developed to extend access to high-quality occupational health guidance for practitioners and employers who may not have immediate access to specialist support. For queries, concerns or feedback, contact craig@ohspecialistshub.co.uk.

3. How Does the System Work?

3.1 Retrieval-Augmented Generation (RAG)

OH Advisor uses a Retrieval-Augmented Generation (RAG) architecture. When a user asks a question, the system first searches a curated occupational health knowledge base for relevant content, then uses that content to inform and ground its response, rather than relying solely on the general knowledge of a language model. This is specifically designed to reduce the risk of plausible but inaccurate responses (“hallucination”) by anchoring answers in source documentation.

3.2 Large Language Model (LLM)

The language model powering OH Advisor is Claude, developed by Anthropic, accessed through the Anthropic API. The model currently in use is Claude Sonnet (claude-sonnet-4-6). The model may be updated as newer versions become available; this note will be updated when that happens.

OH Advisor does not use internet search to generate responses. Responses are grounded in the curated knowledge base described in Section 4. Anthropic does not use data submitted through its API to train its models. More information about Claude and Anthropic’s safety practices is available at www.anthropic.com.

3.3 Embeddings and Vector Search

Questions are converted into numerical representations (embeddings) by an embedding model (nomic-embed-text) running on our own server. These are used to search the knowledge base for relevant content. No question data is sent to any external embedding service.

4. The Knowledge Base

4.1 What is in the knowledge base?

The knowledge base is a curated collection of occupational health documentation compiled by Craig Page, comprising several thousand indexed content sections drawn from:

• UK occupational health legislation and statutory guidance (e.g. Equality Act 2010, Health and Safety at Work etc. Act 1974, RIDDOR, UK GDPR)
• Health and Safety Executive (HSE) guidance
• Faculty of Occupational Medicine (FOM) guidance and good practice documents
• Relevant NICE guidelines
• ACAS guidance on absence management, reasonable adjustments and return to work
• DVLA Assessing Fitness to Drive (November 2025 edition)
• NMC Code (2018)
• ICO guidance on workers’ health data and UK GDPR
• Ethics guidance for occupational health practice
• Clinical reference texts, subject to the licensing position described in Section 4.3
• A proprietary corpus of over 500 fully anonymised occupational health reports, prepared for retrieval, which provides structure, tone and real-world practice knowledge that does not exist in published form elsewhere

The knowledge base does not contain patient records or identifiable clinical data. Where Susan refers to an example from the anonymised corpus (for example, “Case 344”), this is an anonymised reference, not a real patient record.

4.2 Currency and maintenance

The knowledge base is reviewed and updated periodically by Craig Page when legislation changes, new guidance is published, or content is identified as outdated. The target review frequency is every six months, with ad hoc updates after significant legislative or guidance changes. Users should always verify critical legislative references against current statutory sources; Susan will direct users to authoritative sources (HSE, NICE, Gov.uk, ACAS, FOM) for verification.

4.3 Copyright and licensing

Content has been selected on the basis of open access availability, Crown Copyright, or use for professional reference purposes. No subscription-only databases are included.

Some clinical reference publications are currently subject to licensing review with the relevant publishers and professional bodies. Depending on the outcome, they will be retained under licence, replaced, or removed. Where the licensing position of a publication has been unclear, it has been removed pending clarification. OH Specialists Hub Ltd is committed to ensuring all knowledge base content is properly licensed and will update this note accordingly.

Where responses include external links, these direct users to the original source. OH Advisor does not reproduce substantial portions of third-party copyrighted content.

5. Data Privacy and Security

5.1 Infrastructure and hosting

OH Advisor runs on our own dedicated server in DigitalOcean’s London data centre, alongside the knowledge base and embedding model. Apart from the Anthropic API, which generates each response, no other external service processes your questions. All connections are encrypted, and access requires an account with two-step sign-in (password plus a one-time email code).

5.2 What data is processed and what is kept

5.3 What you may share

OH Advisor HR: keep questions anonymous. Do not enter employee names, dates of birth, NI or NHS numbers or other identifiers, or detailed medical information about an identifiable employee. Describe situations generically (e.g. “a 45-year-old warehouse employee”).

OH Advisor Clinical: qualified practitioners may include case-specific clinical information where it is needed for the case. It is protected by encryption, UK hosting, access controls and audit logging. Practitioners remain responsible, as the healthcare professional, for having a lawful basis to use their clients’ or patients’ information and for telling them that they use AI-assisted tools. We recommend including only the information the question needs, and leaving out names and direct identifiers where they are not needed.

5.4 GDPR

OH Specialists Hub Ltd processes personal data in accordance with UK GDPR. Our full Privacy Policy is available at ohadvisor.pro/privacy.html, and our Cookie Policy at ohadvisor.pro/cookies.html. For data subject requests or privacy concerns, contact dpa@ohspecialistshub.co.uk.

5.5 Quality review of stored questions

Stored questions can be read only by our lead clinician, a registered nurse bound by professional confidentiality, and every access is recorded in an audit log. They are used solely to check the accuracy and safety of Susan’s answers and to improve the knowledge base and guidance. They are never shared with employers or any other party, never used for marketing, and never used to train AI models. Users can ask for their stored questions to be deleted at any time.

6. Limitations and Appropriate Use

6.1 What OH Advisor is not

• A clinical decision-making tool for individual patient care
• A replacement for a qualified occupational health assessment
• A source of definitive legal advice
• A diagnostic tool
• A substitute for professional training or clinical supervision

Responses are AI-assisted guidance based on the knowledge base and should always be reviewed in the context of the individual case by a qualified professional.

6.2 Accuracy and reliability

While the RAG architecture significantly reduces the risk of fabricated responses, no AI system can guarantee complete accuracy. Users should:

• Treat all outputs as a starting point for professional judgement, not a final answer
• Verify legislative references against current statutory sources
• Apply their own clinical or professional expertise to the guidance provided
• Seek specialist advice where cases are complex, high-risk, or outside their competence

6.3 MHRA classification

OH Advisor does not perform autonomous clinical decision-making. It is a drafting and guidance support tool for qualified practitioners. All clinical outputs — including report drafts, structured notes and phased return plans — require review, amendment and sign-off by the practitioner using the tool. The classification of AI-assisted tools under MHRA medical device regulations is an evolving area; OH Specialists Hub Ltd monitors regulatory developments and will update this note as guidance is clarified.

6.4 Fairness and bias

The knowledge base has been curated to reflect current UK occupational health legislation, guidance and professional standards, prioritising evidence-based, authoritative sources. Responses are generated in the context of UK employment law and clinical practice; users working with employees whose circumstances may not be reflected in mainstream guidance should apply additional professional judgement. The knowledge base is reviewed for representational balance as part of its periodic review.

7. Human Oversight

OH Advisor operates under continuous human oversight. Every interaction requires a person to initiate, review and apply the guidance. The platform does not take autonomous actions, send communications, access external systems, or operate without direct user engagement.

Susan’s instructions explicitly direct her not to make clinical diagnoses, not to infer conditions not stated by the user, and to flag gaps and uncertainties rather than fill them with assumptions. Where a question falls outside the platform’s competence or knowledge base, Susan is instructed to say so and direct the user to specialist input. Stored questions are reviewed by our lead clinician to monitor the quality and safety of answers (Section 5.5).

8. Governance and Compliance

9. Contact and Feedback

We welcome feedback on OH Advisor from occupational health professionals, HR practitioners and users. To report a concern, inaccuracy or governance query, email craig@ohspecialistshub.co.uk.

This transparency note is reviewed at least annually, and after any significant change to the platform’s architecture, knowledge base, data handling or regulatory context.