How to Use Claude as a Medical Writer: The Nora Skill Guide

Medical writing is a field where one careless sentence can stall a regulatory submission or mislead a patient. It is also drowning in volume: the same trial data has to become a regulatory summary, a journal abstract, a clinician brief, and a plain-language patient leaflet, each in a different register. Used carefully, Claude as a medical writer can take the first-draft burden off that workload. Used carelessly, a generic chatbot invents statistics — which in this field is not a quirk but a hazard.

Nora is the careful version. She is a medical-writing persona you load once into Claude, ChatGPT, or any AI chat, and two rules govern everything she does: she writes for a defined audience and document type, and she never fabricates data. Where a figure is missing she marks it [DATA TO BE PROVIDED] rather than inventing one. Every output is a first draft for qualified review — never a finished record.

Why generic AI medical writing is risky

An unguided model will confidently produce clinical numbers, citations, and event rates that were never in the data. It writes in one register regardless of reader, so a regulatory summary and a patient leaflet come out sounding the same. And it ignores the structures these documents must follow — ICH formats, SmPC and PIL requirements, CONSORT for trials — while implying the result is ready to submit. In a regulated field, each of those is a real problem, not a stylistic preference.

What changes with the Nora skill

Nora asks first: who is the audience, what document type, which jurisdiction. A clinical study report, a journal manuscript, a GP reference, and a patient information sheet are four different jobs, and she treats them that way — matching register, depth, and structure to each. She works only from the data you provide, flags every point that needs clinical or regulatory verification, and keeps the writing scientifically precise without drifting into claims the evidence does not support.

What it actually produces

Drafts across the full span of medical writing: clinical study report sections, SmPC and patient information leaflet drafts, study protocol synopses, journal manuscript sections, conference abstracts, plain-language summaries of trials, and HTA submission narratives. Each one is structured to the relevant standard and handed back as a draft, with placeholders where data or verification is required.

How to get the most out of it

State the audience, the document type, and the jurisdiction before you ask for anything. Give it the real data — it will not invent it, so a draft is only as complete as the inputs. Treat the output as a genuine first draft that saves hours of structuring and phrasing, then route it through your qualified scientific, medical, and regulatory review before anything is submitted or published.

Who this is for

Medical writers, regulatory and clinical teams in pharma and CROs, medical communications agencies, and academic researchers writing up studies. It works with Claude, ChatGPT, or any AI chat that accepts a system prompt. For adjacent clinical documentation, the health & medical skills collection covers consent forms, clinical summaries, and patient materials — each a focused assistant rather than a general chatbot. All outputs require qualified review and are not medical advice.