AI-SAFE-GP Study

  • Research type

    Research Study

  • Full title

    The AI-SAFE-GP study: Evaluating Artificial Intelligence (AI) to SAFEly improve General Practitioner (GP) documentation and communication in complex consultations.

  • IRAS ID

    359605

  • Contact name

    Peter Edwards

  • Contact email

    peter.edwards@bristol.ac.uk

  • Sponsor organisation

    University of Bristol

  • Duration of Study in the UK

    2 years, 0 months, 1 days

  • Research summary

    Background
    In 2023, GPs in England carried out 160 million appointments. After each one, the GP writes a summary in the patient’s medical record. These summaries are important, but they’re often rushed, missing key details, or contain mistakes.
    New computer tools called “AI-scribes” can write these summaries automatically. Companies say they save time and improve quality, but these tools haven’t been properly tested by independent UK researchers.

    AI might also help GPs give better advice to patients about when to seek more help, called “safety-netting advice”. This advice is meant to help patients know what to look out for, but it only happens in about two out of three consultations. It’s rarely written down, even though patients prefer clear, written instructions.

    Aim
    We want to find out if AI can safely improve GP summaries and written advice using real and actor-based consultations.

    What we’ll do
    1. Testing AI in GP consultations
    Ten GPs will each take part in six realistic consultations with actors playing patients. We’ll compare how long it takes them to write notes themselves versus editing AI-made notes. We’ll also ask them what they think about using AI.

    2. Checking the quality of summaries
    We’ll compare what the GP and AI wrote down with what was actually said. We’ll do this for both the actor consultations and 92 real recorded GP telephone consultations collected in a previous study.

    3. Improving written advice
    We’ll work with patients to design a safety-netting advice form, then test how well AI can fill it in based on what was said in consultations.

    Patient involvement
    Patients helped design this study. A group of 8–10 patients will continue to support and advise us throughout the project.

    Sharing what we learn
    We’ll share our findings through short videos, social media, talks, and journal articles.

  • REC name

    South West - Frenchay Research Ethics Committee

  • REC reference

    25/SW/0108

  • Date of REC Opinion

    19 Sep 2025

  • REC opinion

    Favourable Opinion