A survey of 1,000 NHS staff, paid for by AI scribe company Heidi, found 73% said AI helps most with drafting referrals and letters. That doesn't surprise me. Anyone who's finished a clinic and looked at the pile of letters still to write gets it. Writing them is the easy part. What matters is what goes into the tool, and what you do with what comes out.

Before the letter carries your name

Three checks
before you sign.

01What did I put in?

Check what went into the tool — and whether that use was permitted.

02Can I trace every fact?

Compare the draft against the record: doses, dates, allergies and plans.

03Would I sign it?

Your name, your judgement. Rewrite anything you can't stand behind.

A screenshot-friendly reminder, drawn from the three checks in this article.

The one rule

Nothing that could identify a patient goes into a public AI tool. Everything else is about how you work.

Drafting a clinic letter from anonymised notes is fine. So is restructuring a discharge summary from your own bullet points, or rewriting an explanation so a patient at reading age 11 can follow it. That's where the time saving is. What you never do is paste Mrs Smith's name, NHS number and history into ChatGPT to "make it sound more professional." That's not a workflow, it's a breach.

What it's good for

Getting the structure right. "Turn these bullet points into a GP letter with history, examination and plan." AI helps with this, but you control what goes in.

Making it readable. Tightening a paragraph that rambles, or rewriting a clinic explanation so a patient can actually follow it. Writing at reading age 11 isn't dumbing down. It's making information easier to understand.

The routine stuff. The twentieth referral letter of the week doesn't need your creativity. It needs to be accurate, complete and sent.

What it can't do

It doesn't know what's true about your patient. It can't check the record. And it can't take responsibility for the letter. Your name goes at the bottom, so the judgement has to be yours.

It also can't fix a bad process. If discharge summaries take hours because the information is spread across five systems, AI just helps you write faster inside a broken system. Keep that in mind before you count the time you've saved.

Three checks before you sign

Before anything AI has drafted goes near a record:

  1. What did I put in? If it included anything that could identify a patient and the tool was a public one, stop. Delete the chat and start again properly.

  2. Can I trace every clinical fact? Read the draft against the record, line by line. Doses, dates, allergies, plans. Every one needs to come from the notes, not from the AI sounding sure of itself.

  3. Would I sign this as my own? Because you are. If a line makes you hesitate, rewrite that line yourself.

If you want the fuller version of this habit, the Clinician's AI Safety Check is nine questions to run before you use AI at work. And if you want better first drafts, the AI Prompting Guide shows you how to ask for them. Both are free.

Before you rely on this

Using AI to draft a letter doesn't change your record keeping duties. Your regulator's standards, your Trust's documentation policy and your professional accountability all still apply. Check what your Trust's AI policy allows, because it varies a lot between Trusts.

One last tip. The biggest time savings come from drafts you reuse. A good discharge summary layout, sorted once and saved as a template, is better than starting from scratch every time. Build yourself a library.

Sources & further reading

About this answer

Written by
Juwon Akinyande · occupational therapist and founder of Clinically Intelligent
Last reviewed
October 2026
Next review due
April 2027
What's changed
Initial draft · no corrections recorded. Review history will appear here after publication.

Independent educational resource: not an NHS policy document, a substitute for local organisational guidance, or individual clinical advice.

There’s more to Clinically Intelligent.

We explore how AI can make healthcare work better — and how healthcare professionals can help shape what comes next.

Explore Clinically Intelligent