‘GPs are adopting AI faster than we understand it’
As AI tools rapidly spread through primary care, Dr Brighton Chireka argues that GPs must guide the technology, rather than quietly conform to it
Recently, I spoke at Pulse LIVE London on artificial intelligence in general practice. What stayed with me most was not the technology. It was the room.
I asked two questions. First: Who has used AI in clinical practice in the past week? Almost every hand went up. Second: Who has had any formal training in how these tools work, what they can get wrong, or where responsibility sits when they do? Almost every hand went down. That gap between adoption and understanding is the most important clinical issue facing general practice right now.
AI is no longer a future concept. It is already in our consulting rooms, transcribing conversations, generating clinical notes, supporting triage. Most of us are using it. Far fewer of us understand it well enough to use it safely.
Much of the debate focuses on accuracy, liability, and safety. These matter. But there is a quieter risk that deserves more attention. We may be losing our presence in the consultation without noticing.
A colleague described an experience that felt both subtle and unsettling: reviewing notes that were accurate and comprehensive yet did not feel like their own. This matters more than we might realise.
Clinical note writing in general practice is not administration. It is thinking. It is where we synthesise complexity, prioritise risk, and construct our understanding of a patient. When that process is handed to an algorithm, we are not simply saving time. We may be handing over part of our clinical reasoning with it.
I was reminded of this in my own consulting room. A patient I had known for years came in for what appeared to be a routine follow-up. The AI summary was tidy and clinically complete. But something in his manner told me this was not a routine visit.
I put the screen aside and asked how he really was. What followed was one of the most important conversations I have had in recent memory – one no algorithm could have anticipated. It was only possible because I was fully present.
Consider what this looks like in practice: A patient attends complaining of tiredness. During the consultation she mentions that things at home have been difficult and she has not been herself for months. The clinician senses something important in her averted gaze and change in tone.
In a traditionally documented consultation, that instinct finds its way into the record: low mood, possible depression, social stressors, follow-up. With an ambient scribe running, the note captures what was said but not what was sensed. The note is accurate, but incomplete in a way only the clinician in the room would recognise.
Efficiency is real, but it is different from care. If the time AI saves us is used only to see more patients in the same hour, we risk compressing consultations to the point where nuance is missed, safeguarding signals are lost, and the therapeutic relationship – the very reason patients come to us – risks being hollowed out. The question is not how much faster AI can make us. It is what we choose to do with the time it gives back.
One principle must be non-negotiable: AI should be assistive, not autonomous. It supports clinical thinking. It does not replace it. Regardless of how sophisticated these tools become, the responsibility, the accountability and the standard of care remain entirely with the GP.
To help navigate this in practice, I shared a simple framework with the session – the H.U.M.A.N. framework:
- Care must remain human-led
- We must understand the limits of AI
- Do not delegate moral responsibility to an algorithm
- Use of AI must be appropriate – know when to switch it off
- Nuance and relationship matter
This is not a technical checklist. It is a clinical and ethical commitment.
For GPs wondering where to start, three practical steps make an immediate difference.
First, read every AI-generated note before you file it. Published studies have found significant error rates in ambient scribe outputs, including some with potential for patient harm. Your signature is your professional endorsement of its accuracy.
Second, ask your vendor the questions that matter: How was this technology trained, on whose data, and how does it perform for patients who are not native English speakers or have strong regional accents? These are clinical governance questions and deserve clear answers.
Third, protect the sensitive consultation – domestic abuse disclosures, mental health crises, safeguarding concerns. Know when to switch the technology off.
What was clear from the conversations I had with people after the conference is that GPs are not resistant to AI. They are curious, and open to it. The question they are asking is: How do we use this safely? And the answer is to become intelligent customers – not passive users, but clinicians who understand what these tools are doing, question their outputs and take genuine ownership of the care they inform.
AI will only become more present in general practice. The defining question is not whether we use it – it is whether we lead it or let it lead us.
The responsibility is still ours and the opportunity is significant. GPs must use AI to reclaim time for the relationship between doctor and patient, not to replace it. Because AI should not make general practice less human. It should help us become more human than ever before.
Dr Brighton Chireka is a GP, AI consultant and founder of the DOCBEECEE Leadership Academy. He works at the intersection of frontline clinical practice, compassionate leadership and responsible AI in health and social care.
Continue your learning by registering for our upcoming Pulse Virtual Events. These events are designed for GPs and primary care professionals seeking practical, CPD-accredited clinical updates and expert-led insights they can apply straight into everyday consultations. Taking place in May, we’ll be focusing on Chronic Conditions, Dermatology and Diabetes, and in June, Women’s Health, Urology and Dermatology.
We’ll also be travelling to Birmingham, Newcastle, Cardiff, Glasgow, Belfast and Liverpool this year, delivering our Pulse LIVE Events. Book your free place today.
Have you got a view you want to share with Pulse?
We’re always open to first-hand pieces and opinions from GPs.
Email your piece for consideration to be published on our site.
Related Articles
READERS' COMMENTS [3]
Please note, only GPs are permitted to add comments to articles


Like the notes of caution.
In the eyes of AI Bros, GPs are mere delivery devices transferring the waveform “beauty” of their unique experience and management of the probability, complexity and uncertainty in patient’s verbal and physical presentations….into “dead” data particles.
Thus (mind-bogglingly, for GP colleagues are some of the most intelligent and empathic people I’ve come across, even those whose political views I disagree with..) assisting in their own erasure…..Handing off to LLMs and a future AI personal assistant “you can call me Al, your friendly GP”!
An excellent, slightly technical, balanced book by influential AI scholars (Princeton Prof) Narayanan and Kapoor, “AI Snake Oil”, critiques the use limits and also the hype around AI, worth a read.
Us using AI as notes scribe is actually training the AI model to replace us. The data in and of itself is secure, but allowing AI to organise history, examination, differential diagnosis and management essentially does our job. When we read and edit the transcript we are training the feedback loop. Artists, writers, music have had their intellectual property stolen. IT workers are being forced to use AI to do their jobs and then correct the output; literally they are being used to train the AI functionality that will replace them and we are doing this voluntarily. Unbelievable. Think you’ve read the small print and it won’t. The model has to us this feedback to improve its outcomes or it isn’t AI. We are not surgeons or dentists or anaesthetists. The consultation is our tool and our job in toto. Examination. Yeah they are making AI stethoscopes too.
Using AI scribes affects how we consult; in order for the AI to organise better our consultations have to be done in a more linear pattern like a computer type process rather than human thought which is far more complex and includes lateral thinking, intuition, truly innovative thinking, and is more web like. This will change how we consult and the quality of our work making consultations more linear, less innovative and less nuanced.
The AI scribes don’t save time. They create long, unfiltered readouts, guff that I can’t be bothered to read. This reduces the major function of patient notes which is to communicate with other health professionals and instead I find myself scrolling through outpatient letter that used to have a few distinct points and are now full of dross. The two functions of patient comms is to communicate with other professionals and for medicolegal reasons. My consult of ‘cough 1/12; no red flags: no blood, no sob, no increased wob, list of obs, creps left base, doxy, f/u 2/52 unless fully settled’ takes 2s to write 2s to read and covers the function of said notes. My referral letters are similarly brief. Probably helps whoever is trawling through them. Takes longer to read through the guff the AI scribe churns out to then edit and approve it. Sometimes doctors don’t notice the errors and mistakes are made. A patient locally was writing about it on facebook because they read their incorrect notes and complained. Why are doctors jumping to use a tool that hasn’t been tested, isn’t needed, is training the AI to replace us and has been shown to have risks and affect quality of consultations for the worst. Because we don’t slow down and think and are like a flock of sheep that’s why. Wake up folks.
I don’t use an AI scribe, I don’t use AI in my work at all. I used it once to write a journalistic article on the NHS. The text sounded good. Then I looked up the sources which the AI text does list. The worst bunch of irrelevant random sources I have ever seen. No human would use them.. AI just scrapes the internet. You are all mad if you use it. Look up GP notebook. Tried, tested and reliable. And get off the AI scribe. I’m not saying it will replace us with a good alternative. But it’ll probably do better than an ARRS at some point and AI triage will beat 111 pathways and then my friend you are done. We will be reading through AI work to provide oversight and goodbye profession. You don’t think governments will care the AI replacement doesn’t do such a good job as the real person. It’s cheaper so they won’t care. It’ll drive up diagnostics, unnecessary referrals, etc just like ARRS do now. But hey ho. Public money will be going to tech bros, private diagnostics, private hospitals and they can cut out us dopes. Wake up folks
\having trained years ago in patient centred care I find guidelines and templates have taken over. We spend a lot of time noting negative findings as if every case will end up being read by lawyers. If you get sued you are judged by the small print you missed. Guidelines can be very complex and hard to remember and drug interactions are a mine field. I would hope that AI could deal with all this small print stuff leaving us to be the human input with empathy and psychological insight … just like the training my era of GP had, and I do believe we were a lot happier and the public were more supportive and respectful.