How a GP Can Automate Prescriptions and Medical Reports with AI

Sophie, a general practitioner in Bordeaux, couldn't see the end of her evenings. After her last consultation, she would spend between 45 minutes and 1.5 hours writing reports, referral letters, and prescription renewals. That's around 2.5 hours of daily administrative work, the exact figure the French DREES (Health Statistics Directorate) assigns to 30% of a GP's working time in 2025. Since January 2026, Sophie does it in 20 minutes. The rest is automated.

The problem nobody measures

A GP sees 25 to 35 patients per day. Each consultation generates at least one document: a report, a prescription, a letter. Multiply by 220 working days and you get more than 6,000 documents per year, the vast majority of which follow identical structures from one patient to the next.

According to Gartner (2024), AI documentation tools reduce writing time by 65 to 80% when properly configured. McKinsey Health (2025) goes further: doctors who automate their administrative tasks see their patient capacity increase by 15 to 25% without hiring additional staff.

The paradox is that most GPs know these tools exist but have neither the time nor the support to set them up. This is exactly what the Diagnosis-Framing phase of the IMPACT method addresses: identify where time actually goes before proposing anything.

Sophie's story: GP in Bordeaux

Sophie, 38, practices in a group clinic in Bordeaux-Caudéran. She sees an average of 28 patients per day, the equivalent of 6 hours of face-to-face consultations. What she hadn't measured: the 2.5 hours of daily admin, one hour more than the national average according to the same DREES survey.

Together we identified three tasks accounting for 80% of her administrative time:

• Consultation reports (45 min/day on average)
• Prescription renewals for chronic patients (35 min/day)
• Referral letters to specialists and colleagues (30 min/day)

These three tasks share a key feature: they are repetitive, structured, and require little creative judgment. They are ideal for partial automation.

Step 1: dictate instead of type (estimated saving: 30 min/day)

The first and simplest transformation is structured voice dictation. At the end of each consultation, Sophie verbally dictates a 3-sentence summary: reason for visit, examination, decision. Cowork, the desktop automation tool I prioritize for healthcare professionals, transcribes automatically, formats to a pre-defined template, and generates a first draft report in under 20 seconds.

Sophie reviews, corrects if needed, and confirms with a click. What used to take 6 to 8 minutes per patient now takes 90 seconds.

Step 2: automate prescription renewals (estimated saving: 25 min/day)

For chronic patients (diabetes, hypertension, thyroid conditions...), prescriptions are renewed identically in 85% of cases. Cowork stores templates per patient and generates renewals automatically from a simple voice trigger. Sophie says "renewal Dupont" and the pre-filled prescription appears on her screen within 5 seconds.

The signature remains manual, which complies with legal requirements. What the AI does is prepare the document, not sign it.

Step 3: structure referral letters with Claude Sonnet 4.6 (estimated saving: 25 min/day)

Referral letters between doctors follow highly codified structures: clinical context, relevant history, reason for referral, tests already done. Claude Sonnet 4.6 excels at this type of structured drafting.

Sophie dictates clinical information in natural language, and Claude formats it according to a template validated with the group's specialists. The result is a complete, precise, professionally worded letter in under 45 seconds. She only reviews it to verify factual data, not to rewrite it.

Step 4: GDPR and HDS compliance (non-negotiable)

The question every doctor must ask before deploying anything: does my patient data stay under my control?

For Sophie, Cowork runs with a locally deployed model, with no data sent to external servers. No personally identifiable data leaves her clinic. This is the configuration I systematically recommend for healthcare professionals bound by medical confidentiality.

If you opt for Claude or ChatGPT directly, you must verify HDS certification and sign a GDPR sub-processing agreement before any deployment.

Results after 6 weeks

Sophie tracked her working hours herself, before and after:

• Daily administrative time: from 2.5 hours to 20 minutes, an 87% reduction
• Additional patient slots recovered per week: 4 to 5
• Report quality: colleagues noted her letters became more readable and structured, not less
• Total monthly cost (Cowork subscription, setup): around 45 euros

The time saved, valued at an average GP's hourly rate, represents 800 to 1,200 euros in recovered productivity per month, with return on investment from the first month.

What about you?

If you are a GP, nurse, physiotherapist, or any self-employed healthcare professional spending more than one hour per day on repetitive administrative tasks, this type of setup can be deployed in a single working day.

The TransformAudit (1,490 euros) identifies your three to five priority automation opportunities in one session and delivers an operational 90-day roadmap. Get in touch here.

FAQ

Is it legal to use AI to write medical reports?

Yes, as long as the doctor reviews and approves each report before transmission. The AI produces a draft; the signature remains the practitioner's responsibility.

Which AI model works best for medical writing?

Claude Sonnet 4.6 (Anthropic) gives the best results: restrained style, precise vocabulary, strong technical term handling. GPT-5.5 Instant also performs well for repetitive structured documents.

Do you need technical skills to set this up?

No. Cowork works like a standard desktop application, no coding required. Initial setup takes 2 to 3 hours. A doctor-secretary pair can deploy the system without IT support.

Is patient data secure?

This is the critical point. A local solution (Cowork with a locally deployed model) or one with HDS certification is mandatory. Never use ChatGPT or Claude directly with identifiable patient data without verifying compliance and signing a GDPR sub-processing agreement.