The problem
After an incident or near miss, the write-up gets put off because it takes an hour and a half. If you have 2 events a month and each write-up takes 90 minutes, that's 3 hours a month.
What the AI does
The GP lead jots down what happened, with no names, dates of birth or NHS numbers. Paste the notes into ChatGPT, Claude or Gemini and ask for a structured write-up: what happened, contributing factors, learning and actions. The GP lead checks the facts, decides the learning and actions, and files it. Decisions about patients and duty of candour stay with the clinical team.
A worked example
A private GP clinic in Cheltenham has about 2 events a month. The clinical lead writes bullets from memory and an AI assistant structures them.
Before
90 minutes per write-up, 3 hours a month, often delayed
After
About 45 minutes to check and finalise each, 1.5 hours a month, done the same week
What it's worth
About 1.5 hours a month saved (45 minutes on each of 2 write-ups), worth around £150 if a GP's time is worth £100 an hour.
How to set it up
- Write the facts as bullets with no names, dates of birth or NHS numbers.
- Paste them into your AI assistant and ask for a structured write-up with contributing factors and proposed actions.
- Check every fact, and decide the learning and actions yourself.
- File the write-up with the patient identifiers kept in your clinical system, not the AI tool.
- Review actions at your next team meeting.
What goes wrong
- Anonymising properly matters: a rare condition or date can still identify someone. Strip detail you don't need.
- An assistant can invent causes or reassure too quickly. The learning must come from the people involved.
- Duty of candour decisions are for the clinical team, not the AI.
- A neat write-up is not the same as learning. Make sure actions happen.