AI is writing the doctor's notes. The time it saves is the least of it.

Next time you see a GP, you might notice something new. Before you start, the doctor asks whether you mind them using a note-taking tool, and props a phone up on the desk. Then they do something you may not be used to. They look at you for the whole appointment.

The phone is an AI scribe. It listens to the conversation and drafts the notes, and often the letters too, for the doctor to check and sign off. The NHS calls it ambient voice technology. Most doctors just call it the scribe.

It’s arriving quickly. In July, NHS England in the Midlands signed a deal covering 1,239 GP practices and around 70,000 clinicians across 15 trusts, the biggest yet. Four trusts in south-west London are rolling one out to 20,000 clinicians over four years. And in a survey of 598 UK GPs, 40% said they already used one.

That makes the scribe one of the first places AI has properly arrived in an everyday job, at scale, with the results written down. So what do the results say?

The obvious place to start

Every job page on this site lists what AI could take on and what still needs a person. For a general practitioner, the first thing on the AI list is “notes, letters and coding”. The things that stay include “physical examination” and “noticing what the patient didn’t come in to say”.

The GP’s job scores 71 out of 100 here, which means most of the work still needs a person. But the notes were always the obvious target. They’re words, there are a lot of them, and they’re written after every single appointment. A scribe is AI doing exactly the task our scores say it can do, and nothing else.

In the lab, it cut note-writing by two thirds

Last month researchers in Toronto put nine doctors through simulated appointments, with and without a scribe. With one, the time they spent documenting during the appointment fell by 69%. They stopped typing while the patient talked and checked the note afterwards instead.

That’s the version of the story you’ll hear from the people selling them. The real world is more modest.

In real clinics, about two minutes an hour

The biggest real-world study so far was published in JAMA in April. It followed clinicians at five large US health systems, including Mass General Brigham, Yale and UCSF, and compared the 1,809 who took up a scribe with colleagues who didn’t.

The ones with a scribe spent 16 fewer minutes on notes for every eight hours of appointments. That’s two minutes an hour. Their total time in the patient record fell by 13 minutes, and they saw about one extra patient a fortnight. The time doctors spend on the record after hours, the notes finished at the kitchen table, didn’t change.

Two minutes an hour

Minutes of note-writing saved for every eight hours of appointments, at five US health systems

Everyone with one 16 min

Heavy users 27 min

Part of the gap is that plenty of doctors barely used it. Only 32% of those who took it up used it for at least half their appointments. The ones who did saved 27 minutes of note-writing per eight hours, which is more like it.

Kaiser Permanente in California tells the same story. Its doctors used a scribe in 2.5 million appointments over 15 months, and the headline was 15,791 hours saved, the equivalent of 1,794 working days. That sounds enormous. Spread across 2.5 million appointments, it’s about 23 seconds each.

Why so much less than the lab? Some of it is that the notes still have to be read. NHS England’s rules say every note a scribe drafts must be checked by a clinician before it goes into your record, so some of the time saved writing gets spent reading. Some appointments don’t suit it, and some patients say no. And no simulation lab has ever had a phone that wouldn’t stop ringing.

What it gives back isn’t really time

If the scribe saves so little time, why are doctors so keen on it? Because the time was never the main thing.

In a study of 263 clinicians across six US health systems, the share reporting burnout fell from 52% to 39% after a month with a scribe. They also reported less mental strain, less writing up after hours and more attention on the patient in front of them. It was a before-and-after survey with no comparison group, so treat the exact size of the drop with care. But it’s a lot of relief from a tool that saves two minutes an hour.

Less burnt out, a month later

Clinicians reporting burnout, before and after 30 days with an AI scribe

Before 52%

After a month 39%

The NHS has its own evidence. A trial led by Great Ormond Street Hospital and paid for by NHS England covered more than 17,000 appointments at nine London sites, from GP surgeries and mental health services to A&E and ambulance crews. Time spent directly with patients went up by 23.5%. Appointments got 8% shorter. In A&E, doctors saw 13.4% more patients a shift. And 92% of patients agreed to the scribe being used.

That last number matters, because the scribe changes the appointment for the patient as much as for the doctor. You stop talking to the side of someone’s head.

This is the pattern you’d expect from the way this site scores jobs. AI takes one task, the notes, and leaves the rest. The parts of a GP’s day that stay with the doctor, the examination, the judgement, noticing what you didn’t say, get a bit more of the doctor’s attention.

What it gets wrong

A scribe isn’t a stenographer. It summarises, and summarising means choosing what to leave out.

That’s where most of its mistakes are. When researchers in Toronto tested six scribes on the same staged appointments, every one of them left things out, and none of them invented anything. The notes were good, they concluded, but checking them was still essential.

A review from the University of Edinburgh last month went further. Across the 27 papers it looked at, scribes tended to keep the clinical facts and lose the patient’s own account: how it felt, what they were worried about, the things that show in a face rather than in words. The authors warn of “real risks that the patients’ stories are lost”, and they found that the way patients are asked for consent is still largely improvised.

That’s why the rules matter. Under NHS England’s guidance, patients have to be told, they can say no, and a clinician has to check every note before it goes into the record. Each practice also needs a clinical safety officer to watch for the risks, a role that’s on our list of jobs AI is creating.

The jobs that are mostly notes

Look down the health jobs on this site and the notes are at or near the top of almost every list of things AI could take on.

Job	Our score	First thing AI could take on	Still needs a person
General practitioner	71	Notes, letters and coding	Noticing what the patient didn’t come in to say
Specialist doctor	73	Clinical notes, letters and records	Difficult conversations and clinical judgement
Registered nurse	75	Care plans and documentation	Noticing when a patient changes
Ambulance worker	79	Patient records written up from voice notes	Keeping calm in chaotic scenes
Health care assistant	86	Recording observations and care notes	Comforting people who are frightened or in pain
Medical secretary	47	Booking appointments, then records and letters	Talking to patients
Medical records technician	48	Clinical coding, then transcribing records	Checking accuracy

For doctors, nurses, paramedics and healthcare assistants, the notes are one task among many, and the rest is hands, judgement and people. That’s why those jobs score between 71 and 86.

The jobs built around the record are different. Medical secretaries score 47 and medical records technicians 48. For them, letters, coding and transcription aren’t one slice of the day. They’re a big part of it.

Nobody has counted yet what scribes are doing to those jobs, and we’d be wary of anyone who says they have. Our guess is that the first change looks like the one in our picture: the letter drafts itself, and the secretary becomes the person who checks it, chases it, and knows which consultant always forgets the follow-up date. Our page for medical secretaries lists talking to patients, supporting clinicians and running the practice office as the parts that stay. Those were always the hard parts.

Beyond the GP’s room

Doctors got scribes first, but they’re spreading. Hertfordshire Community NHS Trust is rolling one out to 1,000 clinicians after a pilot in its community and mental health services, and expects it to be used in around 250,000 appointments a year.

For nurses the shape of it is the same. “Care plans and documentation” is first on our list of what AI could take on for a registered nurse. “Noticing when a patient changes” stays with the nurse, and on a ward at three in the morning that’s the part that matters.

What it means for you

If you’re a patient, you should be asked before a scribe is used, and you can say no. The doctor still has to check the note. If something matters to you, say it plainly and ask for it to go in, because the thing a scribe is most likely to drop is the part that wasn’t about symptoms. You can usually read your GP record in the NHS App, and you can ask for a mistake to be put right.

If you’re a clinician, the evidence says the benefit comes to those who use it consistently, and the risk comes from trusting it without reading it. Both of those are habits.

If your job is the record, the work that’s going is the typing. The work that’s staying is knowing what the record is for. Find your job to see how it breaks down task by task.

AI scribes are the clearest look so far at what happens when AI arrives in a job that’s mostly about people. It took the typing, which is exactly what our scores said it could do. What it gave back was a little time and a lot of attention, and that turns out to be the part of the appointment that was always the point.