Key takeaways
- Treat every AI tool that touches patient data as a new processing activity: purpose, lawful basis, impact assessment and supplier contract first.
- Tools that influence clinical decisions are likely to be medical devices; ask for MHRA registration, marking and classification.
- For AI scribes used in individual care, implied consent under the common law can apply, but patients must be told the tool is used.
- CQC will look for procurement evidence, a hazard log, a clinical safety officer, human oversight and staff training.
AI is already in the consulting room
AI in practices is no longer a pilot topic. In a 2025 Nuffield Trust and RCGP survey of 2,108 GPs, 28% said they currently use AI tools in clinical practice[1]. Among those users, 57% use it for clinical documentation and note-taking and 44% for administrative tasks[2]. Dental practices and private clinics are adopting AI note-takers, image analysis, chat assistants and call handling just as quickly.
Patients expect care here. In the government's public attitudes tracker, healthcare was the area UK adults most often said needs careful management of AI to prevent negative outcomes, chosen by 29%[3]. Clinicians share the caution: among GPs not using AI, 89% were concerned about professional liability and medico-legal issues[4]. A clear, repeatable checklist lets a practice say yes to useful tools without guessing.
The checklist
1. Define the use and the data
Write down what the tool will do, for whom, which data it will receive (audio, transcripts, images, records, call details) and what it will produce. If you cannot describe the data flow in a paragraph, you are not ready to assess it.
2. Check its regulatory status
The MHRA says many software and AI products used in health and social care are regulated as medical devices[5]. CQC advises that a tool influencing clinical decision-making is likely to qualify, and that practices should ask for MHRA registration, UKCA or CE marking and the device classification[6].
3. Identify the lawful basis and confidentiality position
For UK GDPR, an NHS practice usually relies on its public task and the health care condition; a private clinic on contract or legitimate interests with the health care condition. Separately, the common law duty of confidentiality applies: CQC notes that an AI scribe used for individual care can rely on implied consent, but patients must be told it is being used[6].
4. Complete an impact assessment
UK GDPR requires a data protection impact assessment before processing likely to result in high risk, particularly with new technologies, covering the processing, its necessity and proportionality, the risks and the measures to address them[7]. AI processing health data almost always meets that bar.
5. Read the supplier terms
Confirm the supplier acts only on your instructions, does not use your patients' data to train its models without a lawful basis you have agreed, states where data is processed, and deletes recordings and transcripts on a defined schedule.
6. Meet clinical safety standards
When adopting digital technology for NHS care, practices need to follow DCB0160, nominate a clinical safety officer who is a registered senior clinician with suitable training, and keep a hazard log and risk assessment[6].
7. Use the procurement criteria
The Digital Technology Assessment Criteria help assess suppliers at procurement; if a tool comes from an NHS procurement list, CQC says it is reasonable to assume developer standards have been met, though you must still deploy it as intended[6].
8. Build in human oversight
Decide who checks AI outputs, how, and how you will audit it. CQC expects AI to be used as a support tool, not a replacement for human oversight[6].
9. Tell patients
Update your privacy notice, put a short notice in the waiting area and on the website, and give clinicians a sentence to say before using the tool.
10. Train staff and log problems
Train every user before go-live, and report errors to the supplier, through your incident process and, for medical devices, through the MHRA Yellow Card scheme.
Different tools, different questions
| Tool type | Data involved | Key questions |
|---|---|---|
| Ambient scribe or note-taker | Consultation audio, transcripts, draft notes | Is audio deleted after processing? Who checks the note? Is it a medical device in your use? |
| Online triage or symptom checker | Symptoms, history, contact details | Medical device class, clinical safety case, safe routes for urgent symptoms |
| Results or document processing | Lab results, letters, coding | Error rates, human review of every action, audit trail |
| General chatbot used by staff | Whatever staff paste in | Is patient data ever entered? Usually it must not be |
| Phone or chat assistant for practice information | Caller numbers, call audio, transcripts | Does it take health details? Where are recordings kept and for how long? |
| Image analysis | Clinical images | Medical device status, data retention, training use |
Non-clinical tools still need the basics. Even a phone assistant such as Ava, designed to answer general practice questions without asking for health details, should go through an impact assessment and a processing contract before any patient-specific feature is considered.
Telling patients without alarming them
Transparency is a legal duty and a trust exercise. CQC's position is that because AI technologies are new, practices need to tell people they are being used, which is about transparency and allowing people to object rather than asking for explicit consent[6]. A short, plain script works better than a long leaflet.
- Clinician, before a consultation: ‘I use a secure tool that listens and drafts my notes so I can focus on you. I check everything before it goes in your record. Are you happy for me to use it today?’
- Waiting room or website: what the tool does, that a clinician checks its output, where data is held, how long recordings are kept and how to say no.
- Privacy notice: the purpose, the supplier's role, the lawful basis, retention and the right to object.
- Record the patient's objection if they say no, and do not use the tool for that consultation.
Example: piloting an AI scribe
Example: Southfield Health Centre, a fictional practice, wants to pilot an ambient scribe with three GPs. The practice manager completes the checklist over four weeks. The supplier provides MHRA registration and the device classification for the intended use, a DCB0129 safety case, a processing contract that rules out training on practice data, and a commitment to delete audio within a defined short period after the note is produced.
Example: a partner who has completed digital clinical safety training becomes the clinical safety officer and keeps the hazard log. The impact assessment identifies the risk of an inaccurate note being accepted without checking; the mitigation is a rule that every note is read and edited before saving, audited weekly in the pilot. Posters and a privacy notice update go live a week before the first session. Two patients decline in the first month, and their objections are recorded.
What inspectors will look for
CQC's guidance on AI in GP services says its assessments will focus on the systems and processes that ensure safe and compliant use[6]. Keep a folder for each AI tool with the evidence below, and you will be able to answer an inspector's questions in minutes.
- Procurement evidence: DTAC, DCB0129 from the developer, MHRA registration where applicable.
- Your DCB0160 risk assessment, hazard log and the named clinical safety officer.
- The data protection impact assessment, processing contract and record of processing activities entry.
- Training records for every user.
- Audit results showing human oversight of outputs.
- Incident records and lessons shared.
- How you make sure patients who cannot or will not use digital routes are not disadvantaged.
Staff using AI on their own
The riskiest AI use in many practices is the unofficial kind: a clinician pasting a referral into a general chatbot, or a receptionist using a free tool to rewrite a complaint response. Make the rule simple: no patient-identifiable information goes into any AI tool the practice has not approved. Then make approved tools available, so the rule is easy to follow.
Start with an amnesty rather than an audit. Ask every clinician and member of staff which AI tools they have used for work in the past few months, including browser extensions and phone apps, and promise that the answers are for planning, not discipline. Put each tool on a simple register with its purpose, the data involved and a decision: approve with a contract, replace with an approved alternative, or stop. Review the register every quarter, because new tools appear quickly and staff will keep finding them.
Related reading: data processing agreements, suppliers storing data outside the UK, telling patients about AI, the information governance lead and identity checks on calls.
Frequently asked questions
Do we need patient consent to use an AI scribe?
For individual care, CQC says implied consent under the common law can apply, but you must tell patients the tool is used and let them object[6].
Is a DPIA required for AI tools?
Almost always, because AI processing of health data is likely to be high risk. UK GDPR requires one before such processing starts[7].
How do we know if an AI tool is a medical device?
Ask the supplier for MHRA registration, UKCA or CE marking and the classification for the intended use. Tools that influence clinical decisions are likely to be devices[6][5].
Can staff use free chatbots for admin?
Not with any patient-identifiable information. Provide approved tools with proper contracts instead.
References
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
GOV.UK (Department for Science, Innovation and Technology) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑
GOV.UK (Medicines and Healthcare products Regulatory Agency) (2025). Software and artificial intelligence (AI) as a medical device (opens in a new tab) ↑
Care Quality Commission (2025). GP mythbuster 109: Use of artificial intelligence (AI) in GP services (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2026). UK GDPR Article 35: Data protection impact assessment (opens in a new tab) ↑