Nine short sections, each with sample wording a private clinic can adapt.
Before a member of staff enters a patient's notes into an AI tool to draft a letter, they need to know whether that use is approved. Even with approval, a qualified person still needs to check the draft for errors before it is sent. Without written rules, a UK private clinic leaves staff to judge what information they can share and who needs to review the result.
Why Private Clinics Need a Written AI Policy
A written policy gives staff a shared rule for their use of AI, rather than leaving each person to decide whether a tool is safe. It also makes clear who must check the output.
These rules are already needed in clinical settings. In a 2024 survey reported by BMJ Group, 20% of 1,006 UK GPs reported using generative AI in clinical practice, despite a lack of formal guidance or clear work policies. Of those 205 users, 29% used the tools to generate documentation after appointments.
The policy should record the clinic’s decisions about approved uses. It should sit alongside assessments of how each tool uses patient information and what could go wrong.
The ICO’s guidance says most AI processing of personal data is likely to be high risk, requiring a data protection impact assessment. Even where a clinic judges the risk lower, it must document that decision. The clinic owner or clinical director remains responsible for ensuring these assessments are completed, even if someone else carries them out.
What an AI Policy Template for UK Clinics Should Include
| Suggested order | Section | What to settle first |
|---|---|---|
| 1 | Purpose and scope | Start here: name the policy owner before assigning the remaining work. |
| 2 | Patient information | Decide what information tools may receive before assessing them for approval. |
| 3 | Supplier checks | Use those data rules to assess each supplier. |
| 4 | Approved tools | Build the register once the assessments and supplier checks are complete. |
| 5 | Human oversight | Use the register to identify which outputs need checking and by whom. |
| 6 | Patient transparency | Base patient explanations on the uses actually approved. |
| 7 | Incident handling | Agree how staff report problems before training begins. |
| 8 | Staff training | Bring the approved tools, data rules, review duties and incident process together. |
| 9 | Review date | Set the date once the policy is ready to use. |
1. Set Out the Purpose and Scope
The clinic’s AI policy should cover everyone doing clinic work, including employees, locums and contractors. Its scope should include every AI tool used for that work, even free chatbots accessed on personal phones.
Name one policy owner to maintain the document and answer staff questions, with senior management retaining accountability for AI use.
Sample wording
“This policy sets rules for safe AI use and protects patient information. It applies to all employees, locums and contractors using AI for clinic work, on clinic or personal devices, through free or paid tools. [Name and role] maintains this policy and is the first contact for questions about its application.”
2. List the Approved AI Tools
Staff should use only tools on the clinic’s approved list, with each tool risk-assessed before approval. The CSP’s AI principles call for formal risk assessment before use and ongoing monitoring.
An AI usage policy template should include a register like the example below. A writing assistant approved for general clinic emails needs separate approval before staff use it to draft consultation notes.
| Tool | Approved use | Who may use it | Patient data allowed? | Date approved |
|---|---|---|---|---|
| Writing assistant | Draft general clinic information | Admin staff | No | [Date] |
| Clinical note assistant | Draft consultation notes | Trained practitioners | Only as assessed and approved | [Date] |
Sample wording
“Staff may use only the tools listed in the approved register, for the tasks and data specified. Each tool must pass a documented risk assessment before approval. Staff must request approval before using a new tool or changing how an existing tool is used.”
3. Decide What Patient Information Can Go Into AI Tools
The policy should restrict identifiable patient information to approved tools whose intended use has passed a data protection impact assessment (DPIA). It should prohibit entering that information into public AI chatbots.
These limits apply even when staff only want help rewriting a letter. Removing a name is not enough if other details could identify the patient.
May go in
- General clinic information, such as opening hours and publicly listed services.
- Patient information specifically permitted for an approved task, limited to what that task needs. For a letter rewrite, this may mean sharing a relevant extract, not the full patient record.
Must not go in
- Names, contact details, recordings or identifiable clinical notes entered into public chatbots.
- Case descriptions that could identify someone through their occupation, location or an unusual condition.
- Full patient records when only a small part is needed.
The ICO’s guidance sets out the DPIA requirements. For the assessment process, see the clinic AI data protection guide
Sample wording
“Staff must never enter identifiable patient information into public AI chatbots. Patient information may be used only for approved tasks in tools covered by a completed DPIA, within the agreed data limits. Staff must include only the information needed. If identification remains possible or permission is unclear, staff must check with the policy owner before proceeding.”
4. Require Human Oversight and Sign-Off
AI-drafted notes, letters and messages need approval before they enter a patient record or reach a patient or referrer. “Sign-off” means reading the full draft, checking it for accuracy, correcting errors and confirming that the final version is ready to use. The policy should require the reviewer’s name and approval date to be recorded alongside it. The treating practitioner remains responsible for clinical content and decisions.
The policy should specify who signs off each type of output:
- Clinical notes: The practitioner responsible for the consultation checks them against what was discussed and recorded, correcting anything missing or unsupported.
- Letters: The treating practitioner approves them before they are sent.
- Treatment suggestions: The treating practitioner reviews them against the patient’s history and assessment findings.
- Patient messages: Trained admin staff check routine messages; anything containing clinical advice goes to a practitioner.
- Live patient conversations: AI phone and web chat replies cannot receive individual approval, so the policy owner approves handover rules before launch. Afterwards, the owner or nominee samples conversations at set intervals and updates rules when problems arise.
BookedSolid’s Physica Physiotherapy case study describes staff answering first, with AI answering after three unanswered rings; callers can request a person at any point. For cancellations within four hours of the appointment, AI transfers to staff without completing the cancellation.
Sample wording
“Before AI-generated content is saved or sent, a staff member qualified for the task must read it, check its accuracy and correct any errors. They must resolve uncertainty before approving it, then record their name and approval date alongside the final version. Clinical content requires the treating practitioner’s approval. AI suggestions must not be treated as clinical decisions. Live AI conversations require policy-owner-approved handover rules, with patients able to reach a member of staff at any point.”
5. Tell Patients When AI Is Used
The clinic should explain when AI supports care or administration, both in its privacy notice and at the point of use. An AI policy for clinics should specify who explains this and when, such as before an AI tool records a consultation.
Sample policy wording
“Staff must explain relevant AI use in plain language before it begins, including its purpose and what patient information it processes. The clinic’s privacy notice must reflect current uses. Staff should direct questions or concerns to [name or role].”
Sample privacy notice wording
“The clinic uses [AI tool] to help with [specific task], processing [types of information] for this purpose. Patients can ask [contact] how it works and discuss any concerns.”
Replace the placeholders with the clinic’s approved uses before adding the wording to the privacy notice.
6. Check AI Suppliers Before Signing Up
The policy should require a documented check of every AI supplier before use and again at renewal. This part of an AI policy template should specify who reviews the evidence and approves the supplier.
Ask where patient data is stored, whether it is used to train AI models and who can access it. Check how long it is kept and how it is deleted, and request a data processing agreement and recognised security certification such as ISO 27001.
BookedSolid’s data protection guide lists the full set of questions to ask AI suppliers before signing.
Sample wording
“[Name or role] must document supplier checks before an AI tool is used and at each renewal. Unresolved concerns must be addressed before approval. Changes to the supplier’s terms or data handling trigger an earlier review.”
7. Train Staff on the AI Policy
Every staff member should receive training when they join and at each policy review, before using approved tools.
Staff training should cover:
- Applying the rules to a sample task, such as checking whether a tool is approved for drafting a patient letter.
- How to recognise errors, such as an AI scribe recording a symptom the patient never mentioned.
- Knowing when to seek help rather than accept an AI answer.
Across UK industries, ONS reported that only 11% of businesses with 10 or more employees said more than half their workforce had received AI-related training.
Sample wording
“All staff must complete AI policy training at induction and each review. Training must include practical examples of incorrect or harmful outputs. [Name or role] records completion and checks that staff can apply the rules.”
8. Set Out How to Report and Handle AI Incidents
The AI governance policy should give staff a clear process for harmful errors or patient information sent somewhere it should not go.
1. Stop the affected task
If an AI tool sends a letter to the wrong referrer, stop further automated sending while the cause is investigated. Staff should check recipients and send urgent letters manually. Address any immediate risk to patients.
2. Record what happened
Record the tool, time, task and information involved. Keep relevant evidence securely, without copying patient details into another unapproved system.
3. Tell the policy owner
Report the incident immediately to [name or role]. Name a backup contact for absences. If an AI error could lead to inappropriate treatment, alert the treating practitioner immediately too.
For personal data breaches requiring ICO notification, the clinic must report without undue delay and, where feasible, within 72 hours of becoming aware of the breach [source needed]. Staff must not wait for that deadline to raise the incident internally.
4. Assess harm and reporting duties
The policy owner should arrange a clinical safety review where needed and assess whether a personal data breach occurred, including any notification duties.
5. Fix the cause
Correct affected records or messages, investigate the cause and document safeguards before restarting.
Sample wording
“Staff must stop the affected task, preserve evidence securely and report AI incidents promptly. [Name or role] authorises restarting once safeguards are checked.”
9. Set a Review Date and Keep the Policy Current
Adding an AI scribe or changing how staff use an existing tool should prompt a policy review, even before the scheduled date. Set an annual review date too, so changes in guidance are not missed.
The CSP committed to reviewing its principles every six months. The policy owner should also check the HCPC for new AI guidance at each review.
Sample wording
“[Name or role] will review this policy by [date] and at least annually thereafter. Earlier reviews are required when the clinic adopts a new AI tool or relevant guidance changes. Each revision must record the changes and be shared with staff.”
Open this article pre-loaded into an AI assistant for a quick summary.
Frequently Asked Questions
Is an AI acceptable use policy the same as a clinic AI policy?
An AI acceptable use policy template covers permitted and prohibited staff behaviour, but a clinic AI policy has a wider scope. It also assigns responsibility for clinical oversight, patient transparency, supplier approval and incident handling. A clinic can combine both in one document, provided those responsibilities remain clear.
Is there an official UK AI policy template for private clinics?
There is no official UK AI policy template for private clinics. The CSP’s principles can inform a physiotherapy clinic’s policy. [source needed: confirm the current publication status of the HCPC’s AI guidance and state the date checked, replacing the earlier announcement link with the guidance if published.]
Is a DPIA needed before using AI tools?
For tools handling identifiable patient information, follow section 3’s rule: complete a DPIA before approving that use. The ICO says most AI uses involving personal data meet the high-risk threshold. A lower-risk use may not need a DPIA, but the clinic must record how it reached that decision. BookedSolid’s data protection guide explains the checks.
Putting the AI Policy Template to Work
The first version should reflect the tools staff actually use and the tasks they perform. If staff still have to guess whether a task is allowed, the policy needs more work.
Before approving the policy, the policy owner should work through a patient-letter task with staff: which tool is permitted, what information can go in and who checks the draft.
Read BookedSolid’s data protection guide for the assessment and supplier checks to document under section 6 of the policy before adoption.




