Osteopath AI: Notes, Bookings and Letters Under GOsC Guidance

Yes, an osteopath can use AI for case notes, booking replies and patient letters, provided you stay accountable for every output, tell patients what the tool does, check where their data goes, and review each draft before it is filed or sent. The General Osteopathic Council's interim guidance asks for exactly that.

The guidance does not add new rules. It explains how the Osteopathic Practice Standards you already work to apply when AI is involved, and it labels itself "interim" because the technology keeps moving. The practical consequence is that you do not need a new compliance project. You need a short, written routine for each AI job and the habit of following it.

Follow me on Instagram@sagnikteaches

What the guidance says, translated into practice jobs

The guidance page (read for this tutorial in November 2026) groups its expectations under accountability, AI literacy, transparency and consent, and patient safety and confidentiality. It links each to numbered standards: B1 to B3 for knowledge and competence, A1 to A4 for communication and consent, C2 for records, D4 and D5 for complaints and confidentiality. Here is how those themes land on the three jobs in the title.

Connect on LinkedInSagnik Bhattacharya

Notice what is missing: diagnosis. The guidance mentions AI in diagnosis and preventative medicine as wider healthcare uses, and says a tool with a medical purpose may need regulatory approval. Everything below stays on the administrative side of the line, where you are saving typing time and not outsourcing judgement.

Subscribe on YouTube@sagnikteaches

Step one: decide which job to automate first

Practices that try to switch on notes, chatbot and letter drafting in the same month tend to abandon all three. Pick one using time and risk. Say a solo osteopath sees 24 patients a week and spends about eight minutes writing up each one. That is roughly three hours and twenty minutes of notes weekly. Booking enquiries might take another two hours, and letters (GP updates, insurer reports, referral replies) about 90 minutes.

Notes save the most time and carry the most risk, which is why they go last. Bookings are the gentlest way to learn how an AI tool behaves, and the habits you form there (read before you send, keep a log) carry straight across.

Notes: running an AI scribe the way the guidance expects

An AI scribe listens to the consultation, transcribes it and drafts a structured note. Two scribes a manual-therapy practice is likely to meet are Jane and Heidi. Jane is practice-management software with a built-in scribe: its pricing page, checked in November 2026, lists plans from $59 a month (Balance, capped at 20 appointments a month) and $79 a month (Practice, one full-time practitioner included), five free AI Scribe notes a month on every plan, and unlimited notes at $15 a month for each practitioner who opts in. Heidi is a standalone scribe with a free tier (unlimited transcription and standard templates, with a limited allowance of advanced templates) and paid Clinician and Practice plans whose prices are not published: the page offers a 14-day trial and a "talk to us" button instead. Both change their plans often. Treat these as two examples to test, not recommendations; the osteopath software checklist covers how to compare the rest.

Before the first recorded appointment

1. Read the vendor's terms and privacy policy. The guidance names three questions: where is patient data stored, is it used to train the tool, and what do the terms of use allow. It also makes you responsible for checking that the tool has adequate cyber security, so ask the vendor what protects the data (encryption, access controls, two-factor login). Write the answers on one page and keep it.
2. Check the audio setting. Many scribes can discard the recording once the note is produced. A vendor's own statement matters here, and it can change: one clinical note-taker switched new users to keeping de-identified transcripts by default in June 2026, so re-check settings after any product update.
3. Update your privacy notice. The guidance gives transcription software as the example of something to reference. Add a sentence naming the tool, what it does with the audio and how long anything is kept.
4. Write the patient wording (see the script below).
5. Tell your insurer and ask whether they need a declaration. Keep the reply.
6. Confirm your registration with the data-protection regulator is current, as the guidance reminds you.

The sixty-second consent script

"Before we start: I use a tool that listens to our conversation and
drafts my notes, so I can look at you instead of the keyboard.
The audio is [deleted straight after the note is made / kept for X days].
Only I and [the tool provider, as our data processor] can see it.
It doesn't make any treatment decisions; I check and sign every note.
Is that all right? You can say no at any point and I'll write the
notes by hand, and your treatment won't change."

Record the answer in the notes ("AI scribe explained; patient agreed") so standard A4 (valid consent, recorded as appropriate) is evidenced. If a patient says no, switch the scribe off for that appointment. A practice where saying no is awkward has a consent problem, not a scribe problem.

What a scribe gets wrong, and the edit you make

Here is an invented, illustrative excerpt from a scribe draft after a lower back appointment, with the points a careful reviewer would fix.

Draft: "Pt reports sharp pain L5 region radiating to left leg to the
knee since lifting boxes on Tuesday. No red flags. Plan: soft tissue,
HVLA to lumbar spine, advised rest for two weeks."

In the session the patient actually said the pain "sometimes goes towards the knee", you did not use HVLA (a high-velocity thrust technique) because of how irritable the area was, and you advised relative rest for a few days. The draft made the symptom firmer, invented a technique and quadrupled the rest period. "No red flags" is the most dangerous line, because the scribe may have inferred it from silence. Your checklist for every note: symptoms and their distribution, the red-flag questions you really asked, techniques you really used, advice and its timeframe, follow-up. Anything the draft states that you cannot recall saying gets deleted, not softened.

Sign off the same day. The guidance links records to standard C2 (comprehensive, accurate, legible, prompt), and an unreviewed note drifting for a week undermines "accurate" and "prompt" at once. If the audio is not retained, your edited note is the record, so the edit is the clinical act.

Bookings: an assistant that stays inside its lane

Booking is where AI is least risky and most visible to patients. The tool can be a scheduling feature in your practice software, a website chat assistant, or simply ChatGPT, Claude or Gemini drafting replies you copy into your inbox. The guidance's transparency theme means a chat assistant must say it is automated and must route anything clinical to you. See the clinic chatbot data and consent checklist for the vendor questions.

Rules to give any booking assistant

You answer enquiries for an osteopathy practice. You are an automated
assistant and say so in your first message.
You may: give opening hours, fees and the cancellation policy as
written below; explain what a first appointment involves; offer
available slots shown to you; take a name and phone number for a
call-back.
You must not: give advice about symptoms, say whether osteopathy will
help a condition, suggest a diagnosis, or promise outcomes. If a
message mentions chest pain, numbness in the groin or legs, loss of
bladder or bowel control, a recent fall or accident, a high temperature,
or sudden severe pain, reply: "Please contact your GP or emergency
services now" and flag it for the osteopath.
If unsure, say you'll ask the osteopath to call back.

That last block is a safety rule, not decoration: a chat window is where a patient with a serious problem might type it. Test it with ten awkward messages before launch. Illustrative tests: "My leg has gone numb and I can't feel when I wee", "Can you fix my son's scoliosis?", "Do you do home visits?" (answer only if you do), "Cheaper than the other osteopath?" and "Can I bring my own recording device?".

Drafting replies by hand-off

If you do not want a live chatbot, ask an assistant to draft replies to the emails you receive, using a template pack you wrote. Strip names and symptoms from what you paste in, or use a business plan whose terms exclude training on your content (ChatGPT Business, Claude Team and Gemini in Workspace plans don't train on business content by default, per the vendors; check your own admin settings anyway).

Reply to this enquiry using only the facts in the FACTS block.
Warm, plain English, under 90 words, no clinical advice.
FACTS: first appointment 45 minutes, [fee]; Tuesday to Saturday;
cancel 24 hours ahead or pay 50%; the form is at [link].
ENQUIRY: "Hi, I've had neck pain for two weeks after a long drive.
Can I get in this week?"

Illustrative output: "Hi [first name], thanks for getting in touch and sorry to hear about your neck. First appointments take 45 minutes and we see patients Tuesday to Saturday. Here is the booking link: [link]. If anything changes before you come, such as new numbness or weakness, please phone us. Our cancellation policy is 24 hours' notice." The fix you would make: the model added "sorry to hear about your neck", which is fine, but check it did not add "we can certainly help with that", which promises a result.

The reminder side is simpler still. Appointment reminders and rebooking prompts are routine messages that should read the same each time; how a clinic can use AI to cut missed appointments sets out the timing.

Letters: drafting from a template, signing as the author

Osteopaths write GP update letters, insurer reports, referral letters to other practitioners and patient summaries. A letter is read by someone who was not in the room, so errors travel further than in a note. The workable pattern is a template with locked wording and blanks for AI to fill from your own note, then a human read.

A filled-in example

Say a patient has consented to a summary letter to their GP after six weeks of treatment. You give the assistant your signed-off note excerpts (identifiers removed), not the raw transcript:

Write a letter to a GP from an osteopath. Use only the facts below.
Do not add findings, diagnoses or recommendations that are not here.
Structure: reason for letter; presenting complaint and duration;
treatment given and number of sessions; response; current plan;
what the GP might watch for. Under 220 words. End "Yours sincerely".
FACTS: 41-year-old patient; 3 months of recurrent low back pain;
6 sessions of soft tissue work and exercise advice; pain down from
7/10 to 3/10; discharged to self-management; advised to return if
pain spreads below the knee or sleep is disturbed.

Check the output against the facts line by line. You are looking for invented extras: "no neurological deficit", "MRI not indicated" or a medication suggestion. Those read naturally and are precisely the sort of hallucination the guidance warns about. Add the patient's name, date of birth and address in your own document afterwards, so the identifying details never pass through the tool.

Where letters should not be drafted by AI

• Complaint responses. Standard D4 expects a complaints policy and a human voice; draft them yourself, and take advice if a claim is possible.
• Reports for legal or insurance proceedings, where you are giving an opinion over your own signature.
• Anything breaking bad news, or referring a patient urgently for suspected serious pathology. Write it yourself, phone if it is urgent.

A written policy that fits on one page

Because the guidance is about being able to explain what you do, a one-page internal policy is the cheapest evidence you have thought it through. A filled-in version for a two-osteopath practice might read:

The "incidents" row matters because the guidance says you should know how to raise concerns with a provider if an output could cause harm, and patients should know how to raise concerns with you. Add a line to your complaints information saying so.

Auditing ten notes after the first month

Do an audit instead of trusting the feeling that it is going well. Pull ten AI-assisted notes at random and compare each with what you remember and with your appointment diary. Count how many needed substantive clinical edits (not spelling). Illustratively, if four of ten needed a corrected finding or technique, the tool is not ready for unsupervised use and you should keep it as a first-draft aid only, or change settings and templates.

Track three numbers: minutes per note before and after, edits per note, and patient objections to recording. A scribe that saves two minutes per note but needs four minutes of correction costs you time. Compare it with plain dictation into your software. The small private clinic scribe cost check gives a way of deciding whether the subscription earns its place.

The same audit applies to booking replies: read a week of drafts, count the ones where you changed meaning, and look for any promise of an outcome. For letters, ask a colleague to read two and flag any statement they cannot trace back to the notes.

A quick sum: is the scribe worth it for one practitioner?

Take the earlier figures: 24 patients a week, eight minutes of notes each, 192 minutes in all. Suppose a scribe cuts first-draft time to three minutes of review and correction per patient. That is 72 minutes, a saving of 120 minutes a week, or about 100 hours over 50 working weeks. If your time is worth $60 an hour, that is $6,000 a year against a subscription of, say, $15 to $100 a month. The sum looks lopsided, so test the assumption: if your audit shows five minutes of correction per note, the saving falls to 72 minutes a week, and a patient who declines recording costs you the full eight minutes. The subscription is rarely the real cost. Your review time is.

Two awkward situations and how to handle them

A patient who says yes, then looks uneasy. Watch for hesitation and offer the switch-off again: "I can turn it off now if you'd prefer." Note the change. The guidance asks you to keep the patient central to decisions, and consent that was given reluctantly is weak evidence.

A patient who asks where the recording goes. You should be able to answer in two sentences without reading the vendor's policy aloud: where the data is stored, whether it is used for training, and how long it is kept. If you cannot answer, you have not finished the vendor checks. Go back to the one-page summary and complete it before the next recorded appointment.

Common ways practices slip against the guidance

• The free tool "just for letters". Pasting a patient's history into a free chatbot is the clearest breach the guidance names. Fix: business-grade tool, or placeholders.
• Consent given once, never revisited. A vendor changes how it handles transcripts, your privacy notice stays the same. Fix: dated review every six months.
• Cloned wording. Templates that copy "no red flags" into every note make the record less accurate, not more. Fix: red-flag line left blank for you to complete.
• Staff using their own tools. A receptionist who finds a handy assistant and pastes in a patient email has just moved data you never assessed. Fix: name the approved tools in the policy and tell staff why.
• Assuming a regulated product is approved for your use. The guidance asks you to check regulatory approval where relevant, for example for AI-powered medical devices. A scribe is not the same thing as a tool that interprets images or suggests diagnoses.

A four-week rollout for a small practice

1. Week 1. Read the guidance page fully. Write the one-page policy. Ask the insurer. Update the privacy notice.
2. Week 2. Start booking-reply drafting with human approval. Test ten awkward messages. Log changes.
3. Week 3. Add letter templates with placeholders. Run two letters past a colleague.
4. Week 4. Trial the scribe on five consenting patients, review every note the same day, and run the ten-note audit at the end of the following week before deciding.

If you would rather map this against your own diary and software with someone, that is what my AI implementation consultation covers; see the contact page. Whichever route you take, the guidance is your yardstick: you stay accountable, patients know what is happening, their data stays where you can explain it, and a human reads before anything is filed or sent. For the wider picture of AI across the profession, the AI guide collects the related tutorials.

Osteopath AI questions practices ask next

Do I need a separate consent form for an AI scribe?

GOsC's guidance asks you to discuss AI use in plain language and to consider consent, and gives transcription software as an example to mention in your privacy notice. A short verbal check at the start of each appointment, recorded in the notes, plus the privacy notice entry, is a common way to meet that. Ask your insurer and data-protection adviser if you want a written form.

Can I paste patient details into free ChatGPT to write a letter?

The guidance says sharing confidential patient data with free public AI tools is likely to breach confidentiality. Use a business-grade tool whose terms exclude training on your data, or write letters from a template with placeholders and add identifiers yourself afterwards.

Does the guidance ban AI for clinical decisions?

It keeps every clinical decision with you and says responsibility cannot be passed to a tool, its developer or an employer. Drafting notes, replies and letters is administrative support. Using AI to suggest a diagnosis or treatment plan is a different risk, and a tool may then need medical-device approval.

Should I tell my insurer I use AI?

GOsC says to consider whether your insurer requires a declaration about AI use. Check your policy wording and ask the insurer in writing, then keep the reply with your practice records.

Further reads

• How Physiotherapists Use AI to Write Treatment Notes Faster — How a neighbouring profession handles AI-drafted treatment notes.
• AI Scribe for Physiotherapists in the UK: A Consent and GDPR Checklist — A parallel consent and records checklist from a neighbouring profession.
• 10 Admin Tasks a Small Clinic Can Hand to AI This Month — Ten more admin jobs a small clinic can hand over.
• AI Scribe for Chiropractors: Chiro Tools vs Heidi and Jane — Scribe options compared for a similar manual-therapy practice.
• How to Roll Out an AI Scribe Without Losing Patient Trust — Keeping patient trust while a scribe goes live.
• What Is an AI Scribe and How Does It Work in a Consultation? — The mechanism behind a scribe, if it is new to you.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: General Osteopathic Council interim guidance on artificial intelligence (practice guidance page, read November 2026); Jane and Heidi pricing pages (November 2026); SimplePractice transcript-retention FAQ; OpenAI, Anthropic and Google Workspace business privacy pages.