NHS England has put 19 AI scribe suppliers on a national registry. And in the study behind it, clinicians got 23.5% more direct time with patients (yes, from a note-taker).

That's the good news.

The awkward part: nobody has properly worked out what AI in healthcare does to the people who are still learning how to write a note...

What's actually happening with AI scribes in the NHS

AI scribes (the NHS calls it ambient voice technology) listen to a consultation and draft the clinical note for the clinician to review.

The numbers so far:

  1. NHS England launched a self-certified registry of 19 suppliers in January, backed by a GOSH DRIVE study across 9 London sites and over 17,000 patient encounters
  2. That study found 23.5% more direct patient interaction time, 8.2% shorter appointments and 13.4% more patients seen per A&E shift
  3. A JAMA study of 8,581 US clinicians found scribe adopters spent 16 fewer minutes on documentation per 8 hours of scheduled patient care (about 10% less)

Diagnostics is moving too.

Last week twelve US health systems formed a Diagnostic AI Consortium with Aidoc, covering nearly 20 million patients a year. First results are due in 2027.

So this isn't a pilot anymore. It's becoming the job.

Where the evidence on AI scribes gets thin

→ the same JAMA study found no significant change in after-hours EHR time. The minutes saved didn't buy anyone their evening back → residents were among the groups with bigger gains, but so few used scribes that the authors couldn't judge the effect on learning (and called for research on exactly that) → a February review checked the AMA, ACGME, AAMC and others and found no resident-specific guidance on AI documentation at all

In the UK, the BMA called the draft NHS workforce plan a "massive, dangerous gamble" on AI back in May.

Their line: "A chat bot cannot look after a patient."

Honestly both things are true.

The scribe saves time. It also quietly removes the part of training where you learn to think on paper.

What medical school won't teach you about AI scribes

Writing the note was never just admin. It's where a junior:

  1. decides what actually mattered out of 15 minutes of talking
  2. commits to a differential
  3. notices the question they forgot to ask

bad: open the AI draft, tweak two words, sign good: form your own assessment first, then read the draft and hunt for the gaps

The review authors even suggest limiting AI for sections like the assessment and plan.

That's the bit you have to protect yourself, because no curriculum guidance is doing it for you yet (our AI toolkit for medical students can help).

A critical-thinking framework for AI in clinical work

Five questions before you sign anything an AI drafted:

  1. What would I have written? Jot your one-line impression and plan before you open the draft. If you can't, that's the real problem, not the tool
  2. What's missing? AI notes read fluently, which makes gaps harder to see. Check the negatives, the red flags you asked about and anything the patient said in their own words
  3. Who said what? Did the patient say it, or you, or the relative in the corner? Make sure the note attributes it correctly
  4. Who owns this once I sign? You do. NHS England describes these tools as drafting notes "for clinicians to review". Your signature is the accountability
  5. Who was the evidence tested on? The JAMA data came from academic centres only, and the consortium hasn't reported yet. Ask whether the tool was checked on patients like yours

(nurses and students on placement: the same five work for you, just swap "sign" for "hand over")

What to do this month

  1. If you're on a placement that uses a scribe, ask to compare one raw transcript with the final note. Spot what got dropped
  2. Write the assessment and plan yourself for your next ten cases, before any AI or senior input
  3. Read the NHS England announcement and your trust's own guidance on patient consent and clinician review
  4. Ask one registrar or senior nurse how they check AI drafts. Their checklist will teach you more than the vendor demo

so TLDR: AI scribes are saving real time, but the note was always where juniors learnt to think, so protect that part on purpose.

Which part of a clinical note would you never let an AI write first?

Sources

Get the next issue by email

theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.