In a clinic with one or two practitioners, the consultation is the easy part. Writing it up afterwards is what eats the evening, and it's the first thing to slip when the diary fills. Software that listens to the call and drafts the note moves that work off your plate, as long as it's set up properly.
This article covers what these tools do, what the write-up is costing you now, and what has to be in place before you record a patient. It also covers where the technology gets things wrong, and what to ask a supplier before you pay for anything.
Key takeaways
- 01Count the write-up, not the call: a 20 minute consultation usually carries 8 to 12 minutes of notes behind it, and in a 2 person clinic nobody else is going to do it.
- 02You still own the note: the software drafts, you read it and sign it, and the responsibility for what's in the record never moves.
- 03Do the paperwork before the first recording: a data protection impact assessment, a written contract with the supplier, and an updated privacy notice all come before you press record.
- 04Ask properly and accept no: tell the patient what's being recorded and what happens to it, and make sure refusing changes nothing about the care they get.
- 05Check doses and anything historical: transcription mishears numbers and drug names, and it can turn a patient's hypothetical into a stated fact.
- 06Clinic Pro keeps it in one place: video consultations, AI transcription, drafted notes, prescriptions and the patient record sit in the same system, so nothing gets copied between tools.
What does an AI scribe actually do?
An AI scribe listens to a consultation, writes down what was said, and turns that into a structured clinical note. NHS England refers to this category of software as ambient scribing products, and the word ambient is there because the software sits in the background of the appointment rather than being something you operate during it.
Three things happen in sequence. The audio from the call is converted into text, so you get a transcript of who said what. A language model then reads that transcript and produces a summary in whatever format you've asked for, which for most clinics means history, examination, assessment and plan, or whatever structure your service already uses. Finally it pulls out the things that need doing, so the follow-up date, the test to arrange and the dose change come out as a list rather than being buried in a paragraph.
What comes out is a draft. Somebody still has to read it, correct anything that's wrong, and sign it off as the clinical record, and NHS England's guidance is explicit that users must review and approve the output before anything else happens with it. That review step is the whole safety argument for using these tools at all, and it's the part clinics are tempted to skip once they start trusting the output.
The time saving comes from changing what you're doing during the appointment as much as after it. A practitioner who isn't typing can look at the patient, and a practitioner who isn't holding six things in their head until the call ends tends to ask better questions.
What is the admin after each consultation really costing you?
The appointment ends and the work doesn't. For a solo prescriber or a 2 person clinic there's nobody else to hand it to, so every minute of write-up comes out of either clinical time or your own evening.
Measure it before you decide whether any of this is worth it. Time yourself writing up 5 consultations properly, including the follow-up letter and anything you had to look up, then take the average and multiply by the number of appointments in your week. Most people doing this exercise for the first time find the write-up runs somewhere between a third and half the length of the consultation itself, which means a full clinic day carries roughly another hour behind it.
| What happens after a 20 minute consultation | Doing the write-up yourself | With the call transcribed and drafted |
|---|---|---|
| The clinical note | Typed up afterwards, often hours later | A draft waiting for you when you leave the call |
| The action points | Held in your head until you get to them | Pulled out of what was actually said |
| The follow-up letter | A separate job that gets postponed | Drafted from the same transcript |
| When it's finished | Evenings, and whatever's left on Saturday | Read, corrected and signed in a few minutes |
There's a quality cost as well as a time one, and it's the part people underrate. A note written 6 hours after the appointment is written from memory, and memory drops the specifics first. The detail that turns out to matter 8 months later is exactly the kind of thing that doesn't survive the gap between the consultation and the write-up.
Delay also creates the backlog that private clinics get caught by. Notes that haven't been written yet are notes that can't be answered from, so a query about a patient waits until you've caught up, which is one of the admin tasks that quietly stops a clinic growing. For an independent prescriber whose income is tied to the hours they can see patients in, that hour of typing is the difference between 2 more appointments a day and none, which is the arithmetic behind how to make money as an independent prescriber.
What does the workflow look like from call to finished note?
The point of the setup is that nothing gets copied from one place to another. Every time information has to be moved between systems by hand, you've reintroduced the problem you were trying to solve.
A workable sequence for a small clinic runs like this:
- The patient books and completes their questions beforehand. Their answers are already attached to the appointment, so the consultation starts with context rather than with history-taking, which is the same argument behind forms that flag concerns before the appointment.
- You tell the patient the call will be recorded and note their answer. This happens at the start, in plain language, before anything is captured.
- The consultation runs as normal. Nobody is typing, and the software isn't part of the conversation. The wider requirements for running video appointments safely are covered in online consultations: setup, compliance and workflow.
- The transcript and the drafted note appear when the call ends. The draft is attached to that patient and that appointment rather than landing in a separate inbox.
- You read it, fix what's wrong, and sign it. This is the step that makes it a clinical record.
- The actions come out as actions. The prescription, the follow-up and the referral are generated from the same record rather than being rewritten somewhere else.
Where this breaks down in practice is step 6. Plenty of clinics buy a transcription tool, get a very good note out of it, and then paste that note into whatever system holds their patient records. That paste is a copy of sensitive data moving between two places, it's a chance to put the note on the wrong patient, and it's the reason the time saving never quite feels as big as it looked in the demo. The transcription has to live where the patient record lives, which is the broader point behind why a clinical website needs a workflow underneath it.
What makes a generated note safe to keep?
A generated note is safe to keep when a clinician has read it against what they remember and taken responsibility for it. Nothing else makes it safe, and no amount of accuracy in the software changes that.
Build the checking into the same sitting as the consultation. A draft you review 2 minutes after the call ends gets checked against a fresh memory, and a draft you review on Friday for a Tuesday appointment gets skimmed. That's automation bias, and it's one of the specific risks NHS England names in its guidance, because the better the output usually is, the less carefully people read it.
Give the clinician an easy way to correct things, and expect corrections. If editing the note is awkward, people will either accept output they shouldn't or stop using the tool, and both of those are worse than typing it themselves. Keep a record of what was changed as well, because being able to show that a human reviewed and amended the output is the evidence that your process works.
One structural decision matters more than it looks. Decide whether the transcript itself is part of the record or working material you delete once the note is approved. Keeping full transcripts of every consultation indefinitely means holding a great deal of sensitive data you have no clinical reason to keep, and it's one more thing to disclose if a patient asks for everything you hold on them. The same thinking applies to everything else you store, which is why moving consultations off paper tends to improve record keeping rather than just speed it up.
What do you need in place before you record anything?
Recording a consultation creates health data about an identifiable person, which is special category data under UK GDPR and the most tightly controlled information you'll ever handle. The paperwork comes first, and it's less work than it sounds for a clinic of 2 people.
A data protection impact assessment
Write down what you're processing, why, what could go wrong and what you've done about it. The ICO's guidance on impact assessments covers when one is required, and new technology processing health data sits squarely in the territory where you should be doing one. NHS England's guidance treats it as a baseline requirement before an ambient scribing product goes live. For a small clinic this is a few pages, not a project.
A written contract with whoever processes the data
The supplier is processing patient data on your behalf, so you need that in writing, with what they can do with it, how long they hold it, and what happens when you leave. The single question worth asking before anything else is whether your recordings or transcripts are used to improve their models, because if the answer is yes then your patients' consultations have left your control for good.
A privacy notice that says what you're doing
Patients are entitled to know that consultations are recorded, what the recording is used for, how long it's kept and who can see it. Put it in the privacy notice on your website, and say it out loud at the start of the call rather than relying on somebody having read it.
The honest caveat on all of this is that NHS England's guidance is written for NHS organisations in England, and some of it assumes a scale a 2 person clinic doesn't have. You're not going to appoint a clinical safety officer. What does transfer is the shape of it: assess the risk, write it down, contract properly, tell patients, and review every output. If the CQC looks at how you deployed new technology, that's the kind of evidence they'll want to see.
How should you ask the patient for permission?
Ask at the start, in ordinary words, and make it easy to say no. A patient who feels they've been recorded without a proper explanation has a complaint, and it's one that's very hard to answer afterwards.
Cover four things and it takes about 15 seconds. What's being recorded, what it's for, what happens to it afterwards, and that they can say no. Something close to "I use software that listens to our call and writes my notes up for me, the recording is deleted once I've checked the notes, and I'm happy to turn it off if you'd rather" does the whole job without sounding like a disclaimer.
Then record what they said. A field on the appointment is enough, and it means you can show the conversation happened. Patients who've agreed once shouldn't have to be asked every time, but they should be able to change their mind, and that needs to be as easy as telling you.
If somebody says no, the consultation runs exactly as it would have done and you write the note yourself. Care can't be conditional on agreeing, and being visibly relaxed about it is what makes most patients comfortable saying yes. Worth watching whether particular groups of patients decline more often than others, because that tells you something about how you're asking.
Where does AI transcription get things wrong?
The failures are predictable, which is what makes them manageable. They cluster in a few places, and knowing where to look turns the review from a full re-read into a targeted check.
| What the software gets wrong | How it shows up in the note | What to check before you sign |
|---|---|---|
| Mishearing numbers and drug names | A plausible but wrong dose, written as confidently as a right one | Every dose, strength and medication name |
| Treating something hypothetical as a fact | "Patient has had a blood clot" when they said "if I'd had a blood clot" | Anything that reads as patient history |
| Accents, dialects and English as a second language | Sentences garbled or missing entirely | That the summary matches the appointment you remember |
| Filling a gap with something that fits | A symptom that sounds right for the presentation but nobody mentioned | Anything you don't recognise from the call |
Hypotheticals turned into facts is the failure worth dwelling on, because it's specific to how these tools work and it's genuinely dangerous. NHS England's guidance names contextual misunderstanding directly, giving the example of a patient's hypothetical statement being recorded as a confirmed diagnosis. A transcript is a flat record of words, and the software summarising it doesn't always carry the "if" through into the note. Anything that asserts a diagnosis or a history needs your eyes on it.
Accuracy also varies by who's speaking, and that's a fairness problem rather than just a technical one. Performance drops with strong regional accents, with speakers whose first language isn't English, and with speech disorders. If your patients include people in any of those groups, and they will, you're going to be doing more correcting for some patients than others, and you should know that's happening rather than discovering it later.
One more thing to watch for at the point of purchase. If the product does anything beyond writing down what was said, like suggesting a diagnosis or proposing next steps, it may meet the definition of a medical device and fall under MHRA regulation. That's a very different position from a tool that produces documentation, and the supplier should be able to tell you straight away which one they are.
What should you ask a supplier before you buy?
A supplier who's done the work can answer 6 questions immediately, and those answers tell you more than any demo will. The ones who haven't done the work will offer to come back to you.
| What to ask the supplier | Why the answer decides it |
|---|---|
| Are our recordings or transcripts used to train your models? | If yes, your patients' consultations have permanently left your control |
| Where is the data processed and where is it stored? | Processing outside the UK needs a transfer safeguard written into your contract |
| How long is the audio kept, and can we set that ourselves? | Holding recordings longer than you need is the easiest failure to avoid |
| Will you give us your data protection and clinical safety documentation? | A supplier who can't produce it hasn't done the assessment |
| Does the product suggest diagnoses or next steps? | That can make it a regulated medical device, which changes your obligations |
| How does it perform with different accents and dialects? | A supplier who says accuracy is uniform either hasn't measured it or isn't telling you |
Add one practical test that no answer substitutes for. Run the thing on your own consultations for a fortnight before you commit, with your actual patients, your actual accents and your actual vocabulary, and compare the notes you get against the notes you'd have written. A tool that handles general practice well can struggle with the specific vocabulary of a weight management or ADHD service.
Ask how the product fits what you already use, too. A transcription tool that doesn't connect to wherever your patient records live creates a copy-and-paste step that undoes a good share of the benefit, which is the recurring argument in why you need more than a website to run an online clinic.
Run consultations and notes in one place with Clinic Pro
A small clinic that buys video calling from one supplier, transcription from another and patient records from a third ends up moving sensitive data between three systems by hand. We built all of it into one platform so there's nothing to move.
- Video consultations built into the platform. The appointment, the patient record and the call are the same thing, so there's no separate meeting tool to manage. See clinical admin and patient records.
- AI transcription with the notes drafted for you. The call is transcribed and a structured note is produced against that patient, ready for you to read, correct and sign.
- Questions answered before the appointment. Patients complete your medical questions on their phone beforehand, so the consultation starts with their answers already in front of you. See digital intake forms.
- Prescriptions and labels from the same record. Once the note's approved, the prescription and dispensing label come from the record rather than being typed somewhere else. See the platform.
- An audit trail behind all of it. Every amendment and approval is timestamped, which is what you'll want if anybody ever asks how a note was produced.
If you're writing up consultations in the evening and want that time back, get in touch and we'll show you how the consultation and the record fit together.
The note should be finished before the call is
The practitioners who get the most out of this aren't the ones who trust the software most. They're the ones who've set it up so the draft arrives attached to the right patient, and who read it while the appointment is still fresh.
Everything else is paperwork you do once. An impact assessment, a contract, a line in your privacy notice and a sentence you say at the start of each call. None of it takes long, and all of it is far easier before the first recording than after a patient asks what you did with theirs. If you're setting the clinic up from scratch, it belongs in the build rather than being added later, which is the case made in how to start an online clinic in the UK.
Time your own write-up this week and see what the number is. If it's an hour a day, that's a day a fortnight you're currently spending on typing, and it's the cheapest hour in your clinic to get back.
