Most people, including some inside healthcare, think of a clinical consultation as a simple conversation. The patient describes what is wrong. The clinician asks questions. There is an examination, perhaps imaging to review, a discussion of options, and then a plan. A conversation, with notes attached.
That description isn't wrong, exactly. It is just incomplete. It misses what is actually happening in the room, and it badly underestimates what has to happen afterwards.
I have been seeing patients for over 30 years. The longer I do this, the less a consultation looks to me only like a conversation and the more it looks like an event with consequences. I want to walk through what those consequences are, because they are the reason I started building Regenemm Healthcare, and they are the reason I am uneasy about a wave of AI tools that frame the consultation as a recording problem.
Every consultation is a decision-making event
When a patient walks into the clinic room, they rarely arrive with one clean question. They bring a story. Their symptoms have usually evolved over weeks, months, sometimes years. They are carrying imaging from one or two centres. They are carrying impressions from a prior clinician, from a family member who searched online, from their own theory about what is happening to their body.
My job in that time frame, often up to an hour, is to do several things at once.
I have to let the story come out in the patient's words before I push it into a template. I have to ask the questions that distinguish a structural problem from a functional one, an urgent presentation from a chronic one. I have to examine them and notice the small things: a subtly weaker dorsiflexion, a reflex that has disappeared, the way someone holds their neck when they sit down, how they stand or sit, what type of gait they have, whether they are in pain, and whether there are functional components that need to be understood. I have to read their imaging, often more than one study, and form a clinical impression that ties the story, the examination and the pictures together. I have to build a cohesive narrative.
And then I have to decide.
That decision is not only "do we operate." It is a layered set of decisions about what is going on, what to recommend, what to rule out, what to watch, what to refer for, and what to safety-net. Each of those decisions then has to be communicated back to the patient in language they can actually understand and act on, often when they are tired, anxious, and have just absorbed a lot of new information.
That is not only a conversation. That is a clinical event with significant weight attached. In software parlance, it is an exercise in state management.
All consultations are a documentation event
The moment I close the clinic room door behind the patient, a second job begins.
I owe the patient a record. I owe their GP or the referring clinician a letter. I owe the hospital, the insurer and any future clinician an account of what I considered and what I recommended. If I did anything procedural, I owe an operative or procedural note. If I prescribed something, I owe a medication entry that other clinicians can rely on later. If something was uncertain, I owe a record that captures the uncertainty honestly, rather than papering over it with confident-sounding language.
People sometimes assume that if you record the encounter, the documentation problem is solved. It isn't. A recording captures words. It does not capture clinical reasoning. It does not separate what the patient said from what I concluded. It does not generate a structured medication list. It does not produce a letter that a referring GP can read in sixty seconds and understand the plan.
A transcript is a starting point. The clinical record is the destination. They are not the same thing, and treating them as if they are is one of the biggest conceptual errors in healthcare AI right now.
Each consultation is a safety event
This is the part most people outside healthcare don't think about, and it is the part I think about most.
Every consultation produces facts that, if they are wrong or missing, can hurt someone.
If the laterality is wrong, left versus right, and that error survives into a referral or an operating list, a patient can be harmed. If a medication allergy isn't carried forward, a patient can be harmed. If a red flag like new bowel or bladder dysfunction in a patient with back pain isn't escalated, a patient can be harmed. If the patient understands the consultation differently from the way it has been documented, a patient can be harmed.
I take this seriously. Every clinician I respect takes it seriously. It is the reason I am sceptical of any system that promises to solve documentation by listening to the room and writing what it heard. The bar is not "did you transcribe accurately." The bar is "is the clinical record safe to act on."
I often tell my patients before surgery: "I am also your risk manager, and I will do everything in my power to minimise and mitigate risks for you and to you."
All consultations are an obligation event
When the consultation ends, I have generated obligations that don't disappear because I am running late for the next patient.
I have to make sure the patient leaves with something they can refer to. Patients forget quite a deal of what we discuss in a consultation. That is not a criticism. It is how memory works under stress. The encounter generates a need for a clear, written summary the patient can take home, share with family, and read again at three in the morning when they wake up worried.
I have to make sure the referrer is informed. A consultant letter is not a courtesy. It is the way primary care continues to manage the patient between specialist appointments. If that letter is late, generic, or wrong, the GP is being asked to do their job blindfolded.
I have to make sure follow-up actions are tracked. An MRI to be ordered. A review in six weeks. A referral to physiotherapy. A safety-net instruction for symptoms that should trigger an urgent presentation. If any of those actions falls through the cracks, the consultation has failed regardless of what was said in the room.
And I have to be able to defend the record. Years later, if a question is asked by the patient, by another clinician, by a regulator, or by a court, I have to be able to show what was discussed, what was decided, and why. That requires more than a transcript. It requires provenance: who wrote what, when, and after whose review.
What gets lost when we frame this as "a conversation"
When you frame the consultation as a conversation, you start designing tools that listen to the room. That is a useful thing to do, but it is not a complete thing to do.
A consultation framed as a conversation produces:
- a single document, usually a note
- written from the AI's interpretation of the audio
- with no separation between what the patient said and what I concluded
- with no provenance trail
- with no companion outputs for the GP, the patient or the follow-up system
- and frequently with the clinician asked to "review" something they had no part in drafting until the end
A consultation framed as a clinical event produces:
- a structured clinical note, drafted by AI and authored by the clinician after review
- a referrer letter, written for the actual reading audience and tone
- a patient summary, in language the patient can use at home
- a tracked set of follow-up actions
- an audit trail showing what the AI proposed, what the clinician changed, and what was finally approved
The second design assumes from the start that the clinician is the author of the record, that the AI is a drafting assistant, and that what leaves the consultation has to be safe in every direction it gets sent: to the patient, to the GP, into the medical record, and into the future.
That is a different problem from "transcribe the room," and it requires a different architecture. This is how we see it in Regenemm.
The architecture we want and commit to build from
These are the principles behind the way Regenemm is being built.
The consultation is treated as a source event. From that source event, multiple outputs are generated: a clinical note, a letter, a patient summary, a list of actions, and an audit trail. Each output is shaped for its actual reader. Each output is reviewable by the clinician. Each output is traceable back to the source. This is true state management.
The clinician remains the author. The AI is a drafting tool, not an oracle. The patient receives something they can hold onto. The system as a whole keeps a clean record of what was said, what was concluded, and what changed between the two.
That is what I mean by a governed clinical record. It is not a transcript with a tidy header. It is a layered, reviewable artefact that respects the weight of the encounter that produced it.
What I want clinicians to ask
If you are a clinician thinking about adopting an AI documentation tool, the question I would ask is not "can it transcribe." Almost all of them can, to varying levels of fidelity.
The questions I would ask are:
- Does it produce the outputs I actually owe at the end of a consultation, or only one of them?
- Does it preserve the difference between what the patient said and what I concluded?
- Does it let me review and correct before anything is finalised?
- Does it keep an honest record of what it generated and what I changed?
- Will it stand up if a colleague, an auditor or a court reads it in five years?
Those are the questions I ask of any tool I let near a real patient encounter. They are the questions I ask of the tools I am building.
Closing
A consultation deserves more than a transcript. It deserves a system that recognises the weight of what just happened in the room: the decisions, the obligations, the responsibilities, and the need to help the clinician carry that weight without dropping anything.
That is the problem I am trying to help solve. Not because AI is exciting, but because the documentation burden on clinicians is unsustainable, and because patients deserve records that are accurate, useful and safe long after the consultation has ended.
Related Regenemm workflow
If this is the clinical problem you recognise from your own practice, Regenemm Voice is being built to help clinicians convert real consultations into governed, reviewable clinical records, patient summaries and downstream care outputs.
Explore how Regenemm Voice turns consultations into governed clinical records
See how Regenemm Voice approaches AI clinical documentation software
Brendan O'Brien is Founder of Regenemm Healthcare and a practising neurosurgeon.
