Aipedia: Doctors and medical practices

AI-drafted clinic letters and referrals

Aipedia, by the Apex broking team · Last reviewed 2026-10-27

Doctors can let AI draft clinic, referral and discharge letters, but must check them before they go, because the GMC treats the content as the doctor's own communication. An error in a letter travels to every clinician and patient who relies on it.

Part of: AI in doctors and medical practices

In short

AI tools now turn consultations and records into clinic letters, referrals and discharge summaries in seconds. GMP Domain 4 says information you communicate must be accurate, not false or misleading, and checked first, and paras 69 to 70 require accurate records showing who created them. Typical failures are wrong drugs or doses, the wrong side, invented findings and missing safety-netting. When a GP or another clinician relies on a wrong letter, the claim may name both, and indemnity usually depends on where the work was done and whether a human reviewed the output.

How AI drafts letters

Ambient scribes often produce a clinic or referral letter alongside the note. Other tools draft discharge summaries from the hospital record, or generate referral letters from structured templates and the GP record. Some practices use generative AI to rewrite letters into plain English for patients. The appeal is clear: letters that once waited days for dictation and typing can be sent the same day.

The duties that apply

Domain 4: Good medical practice requires information you communicate to be "accurate, not false or misleading" and checked first, and confidentiality in written and digital communications.
Records (paras 69 and 70): letters filed in the record must be clear, accurate and contemporaneous, and show who created them.
Delegation (para 66): if a colleague or administrator sends AI-drafted letters for you, be confident they have the knowledge and training to spot errors.
Incident reporting (para 73): report errors from AI tools that cause or nearly cause harm.

Where the letter comes from an ambient scribe, NHS England's scribing guidance requires users to review and approve outputs and organisations to reinforce practitioners' ongoing responsibility to review and revise them.

What can go wrong

Illustrative scenario: a consultant sees a patient and decides to reduce a heart medication. The AI-drafted clinic letter states the old dose as the new one and lists a cancelled investigation as booked. The consultant signs a batch of letters at the end of the clinic without reading each in full. The GP updates the prescription from the letter, and the patient continues on the wrong dose until they are admitted.

Other common errors include the wrong side for a procedure or lesion, allergies missed or attributed to the wrong drug, a family member's history recorded as the patient's, and safety-netting advice the doctor never gave. Because letters move care between teams, one error can be repeated in prescriptions, coding and later referrals.

Confidentiality is a separate risk. A letter drafted with a public generative AI tool, rather than an approved system, may send the patient's details to a third party, contrary to the GMC's confidentiality guidance and data protection law. Letters rewritten for patients in plain English can also lose important caveats, such as when to seek urgent help, so the patient version needs the same check as the clinical one.

Who pays when someone relies on the letter

The receiving clinician is entitled to rely on a specialist's letter in many situations, but may also be criticised for not spotting an obvious error. A claim may therefore name both the author and the recipient. For NHS work, CNST covers trusts and CNSGP covers NHS general practice, so the schemes usually deal with apportionment. For private consultants, the doctor's own indemnity usually responds, subject to the terms. If the letter came from a faulty tool, the deploying organisation may struggle to recover from the supplier: NHS England warns liability may default to the provider, and one defence organisation's published position excludes system defects and product liability. A notification of circumstances may be needed once an error is found.

Practical controls

Read every AI-drafted letter in full before signing; do not batch-approve.
Check medicines, doses, laterality, allergies and planned investigations against the record.
Make sure the letter reflects what you actually examined and said.
Mark the letter with who reviewed it.
Set a process for correcting a sent letter and telling recipients.
Audit a sample of AI letters for errors.

How Apex helps

Apex arranges indemnity for doctors and clinics, including private work outside NHS schemes. A broker reviews your cover and how your policy treats AI-drafted correspondence; see what to document about AI use.

Sources

Frequently asked

Do I have to read every AI-drafted letter?

Yes. GMP Domain 4 requires information you communicate to be accurate and checked first. A letter you sign is your communication, and NHS England guidance for scribes says practitioners have an ongoing responsibility to review and revise outputs.

Who is liable if a GP acts on a wrong AI-drafted letter?

Possibly both the author and the GP, depending on whether the error should have been obvious. For NHS work, CNST and CNSGP usually handle the claims. For private work, the author's own indemnity usually responds, subject to its terms.

Should letters say AI was used?

There is no specific GMC rule requiring this, but para 70 says records should show who created them, and the GMC expects doctors to discuss AI use with patients. Showing who reviewed and approved the letter is good practice.

Check your indemnity covers how you use AI

Tell us about your practice and the work you do. A broker will review how your current cover treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.

Start an indemnity quote Request a callback

Apex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.