Start with a work product
Ask participants to bring a task rather than a favourite tool: a meeting brief, a training outline, a policy summary or a management report. Use fictional or approved non-sensitive inputs. Agree what a useful output should contain before the exercise.
Use a two-hour practical session
Spend 15 minutes on limitations and data boundaries, 20 on a guided demonstration, 45 on a department exercise, 25 on peer review and 15 on an action plan. This is a suggested format, not a report of completed training.
Teach a reusable instruction pattern
State the task, audience, source material, required format and constraints. Ask the tool to identify missing information rather than invent it. Compare the response with the source and edit it as a professional work product.
Make review visible
Ask another participant to check factual accuracy, omissions, tone and confidentiality. Keep a record of correction effort. A quick first draft has little value if verification takes longer than the original task.
Leave with one pilot
Identify the task owner, approved data, review checklist and a small evaluation period. Compare quality and total effort with the baseline. Stop or redesign workflows that create unacceptable errors.
A checklist to use with your team
- One defined departmental task
- Approved or fictional practice material
- A reusable instruction template
- A reviewer checklist
- A named owner and pilot measures
Further reading: WHO guidance on AI ethics and governance.
Continue with healthcare digital transformation or AI for healthcare.