Physiotherapists write treatment notes faster with AI in three ways: an ambient scribe that listens to the session and drafts a SOAP note, dictation into a scribe straight after the patient leaves, or typed shorthand that an AI expands into a full note. Each can save several minutes per patient, provided ranges, sides and exercise doses are checked before signing.
The twist for physios is that much of a session is silent or noisy. Hands-on treatment happens without talk, measurements get taken and remembered rather than said, and exercises happen in a gym full of other voices. Scribes designed around a doctor and patient talking across a desk hear the chat about the patient's weekend and miss the goniometer reading. The physios who get the most out of AI are the ones who say their findings out loud.
Why physio notes trip up tools built for doctors
A general practice consultation is mostly talk. A physiotherapy session is mostly doing. That difference shows up in the notes in four ways:
- The objective section carries the note. Range of motion in degrees, strength grades out of five, special tests, girth measurements, outcome measure scores. Get these wrong and the note is wrong, however well the history reads.
- Sides and levels matter everywhere. Right or left, L4 or L5, medial or lateral. A scribe that swaps one creates a note that contradicts the next.
- Exercise prescriptions are dosed. Three sets of twelve at a given load, twice a day, is a treatment. "Some squats" is not.
- Sessions move around. Couch, gym, treadmill, stairs. The microphone that was a metre away is suddenly ten metres away, next to someone else's session.
None of this rules AI out. It changes which method suits which part of the week.
Three ways to get a draft, and which suits which session
| Method | How it works | Best for | Main risk |
|---|---|---|---|
| Ambient scribe | Records the session, drafts the note from the conversation | Initial assessments in a treatment room, where the history is long | Missing silent findings; recording other patients in shared spaces |
| Post-session dictation | You speak a two-minute summary after the patient leaves; the scribe structures it | Follow-ups, gym sessions, home visits | Forgetting to mention something you did |
| Shorthand to note | You type your usual abbreviations; AI expands them into full sentences | Clinicians who already type fast notes; no microphone needed | AI expanding an abbreviation wrongly |
A common pattern in a busy clinic: ambient for new patients in private rooms, dictation for follow-ups and gym sessions, and shorthand on days when the clinician is behind and typing in the gaps.
An initial assessment, from narration to note
Here is an illustrative runner with right-sided knee pain, recorded with an ambient scribe. The clinician narrates the objective examination as she goes. An extract from the transcript:
"Standing, I'm watching you squat... pain comes on at about 60 degrees of knee bend, right side. Single-leg squat on the right, your knee drifts inwards, the left is better controlled. Lying down now: knee range is full both sides, no swelling. Pushing out against my hand, hip abduction is 4 out of 5 on the right, 5 out of 5 on the left. Your questionnaire score today is 58 out of 80."
The drafted objective section (illustrative):
O: Squat: R anterior knee pain onset ~60 deg knee flexion.
Single-leg squat: R dynamic knee valgus; L good control.
Knee ROM full and pain-free bilaterally. No effusion.
Hip abduction strength (0-5): R 4/5, L 5/5.
LEFS: 58/80.That is a useful draft, but two corrections were needed. The scribe wrote "pain-free" for the range of motion, which the clinician never said; she had said "full", and the end of flexion was actually uncomfortable. And the plan section recorded "running reduced", where she had said "running halved for two weeks, then build back by about 10% a week if pain stays at 3 out of 10 or below". The dose was the treatment, and the draft had flattened it. Both fixes took under a minute, against the ten or twelve minutes she used to spend writing an initial assessment from scratch.
Follow-up notes in a minute with shorthand
For follow-ups, many physios never switch on a recorder. They type the shorthand they have always used and let a chat assistant or their clinic software's AI turn it into a readable note. The prompt, with no patient identifiers in it:
Expand these physiotherapy follow-up notes into a SOAP note.
Keep every number, side and dose exactly as written.
Do not add findings, tests or advice that are not in the notes.
If an abbreviation is ambiguous, keep it as written and add it
to a list headed CHECK at the end.
Style: short clinical sentences, past tense, no patient name.
Notes:
FU3. R PFP. Pain 3/10 run, 1/10 stairs (was 5/10, 3/10).
Ran 3 x 20 min wk, no flare. SLS R valgus improved.
Hip abd R 4+/5. LEFS 66/80.
Rx: progressed side-lying abd to banded standing 3x12,
step-downs 20cm 3x10 R, cont wall sit 3x40s.
Plan: build run 10%/wk if pain <= 3/10. R/V 3/52.An illustrative output, with the one line that needed fixing:
S: Third follow-up for right patellofemoral pain. Pain 3/10 on running
and 1/10 on stairs (previously 5/10 and 3/10). Running 3 x 20 min
per week with no flare-ups.
O: Single-leg squat: right dynamic valgus improved.
Hip abduction strength right 4+/5. LEFS 66/80.
A: Improving.
P: Progressed side-lying hip abduction to banded standing hip
abduction 3 x 12. Step-downs from 20 cm step, 3 x 10 right.
Continue wall sit 3 x 40 s. Increase running by 10% per week
if pain stays at or below 3/10. Review in 3 months.
CHECK: none."R/V 3/52" means review in three weeks, not three months. The AI read "/52" wrongly and did not flag it, because it thought it knew. Clinic-specific abbreviations are exactly where shorthand expansion goes wrong, so add a line to the prompt listing your common ones: "3/52 = 3 weeks, 3/12 = 3 months, FU = follow-up, R/V = review". With that line in, the error stopped.
A template that makes the scribe say "not recorded"
Most scribes let you edit the note template, and a few minutes spent on it saves time on every note. The important change is to give the AI a way to admit a gap. Without one, a model asked to fill a "Red flags" heading will write something plausible. With one, it writes "[not recorded]" and you notice.
SUBJECTIVE: presenting complaint, onset, aggravating / easing factors,
24-hour pattern, previous episodes, goals.
RED FLAG SCREEN: list each item asked and the answer.
If not discussed, write [not recorded].
OBJECTIVE: each test or measure with side, value and unit.
Only include tests the clinician named aloud.
OUTCOME MEASURE: name, score and scale (e.g. LEFS 58/80).
If none, write [not recorded].
ASSESSMENT: clinician's stated impression only. Do not add one.
TREATMENT GIVEN TODAY: techniques and exercises, with doses.
HOME PROGRAMME: exercise, sets x reps or time, load, frequency,
progression rule.
PLAN: review interval, goals for next session, onward referral.
CONSENT: record that consent to assessment and treatment was given
only if the clinician said so; otherwise [not recorded].In practice, the first week with this template produces a lot of "[not recorded]" lines, and that is the point. An illustrative first-week note came back with "RED FLAG SCREEN: [not recorded]" for a new low back pain patient. The physio had asked the questions but had done it while walking the patient to the couch, out of range of the microphone. She now asks them seated, at the start, and the scribe records each question and answer. The template turned a silent gap into a visible one, which is the difference between a defensible note and a risky one.
The numbers AI gets wrong in physio notes
These are the fields to read twice. Each example is illustrative of the kind of slip that comes from speech recognition or over-eager drafting.
| Field | What went wrong | Example |
|---|---|---|
| Range of motion | Misheard number | "A hundred and ten" recorded as 100 degrees |
| Strength grades | Plus or minus dropped | 4+/5 recorded as 4/5, hiding the progress |
| Sets and reps | Order swapped | "Three by ten" recorded as 10 x 3 |
| Loads and heights | Units lost | "20 on the step" becomes "20 kg" instead of 20 cm |
| Outcome measures | Wrong denominator | LEFS 58 recorded as 58%, or an Oswestry score recorded out of 50 instead of as a percentage |
| Pain scores | Activity mixed up | 3/10 on running recorded against stairs |
| Laterality | Side swapped | Left shoulder in the subjective, right in the plan |
The practical defence is to say the unit and the side every time ("right hip abduction, four plus out of five"), and to scan the objective and plan sections for numbers before anything else.
Exercise prescriptions: record the dose, not the idea
Home exercise programmes are where notes and patient handouts meet. If the note says "hip strengthening", nobody can tell next week whether the patient progressed, and a colleague covering the review has to start again. The note should carry the exercise, the sets, the reps or time, the load or height, the frequency and the progression rule.
Before, as a scribe might draft it from a chatty session: "Given some hip exercises and told to keep them going."
After, once the physio has said it aloud: "Banded standing hip abduction, 3 x 12 each side, light band, daily. Progress to medium band when 3 x 12 is easy with good control."
Some clinics now have the AI draft the patient's exercise sheet from the same note, so the handout and the record match. That workflow is covered in creating home exercise programmes with AI.
Discharge summaries and letters to referrers
Once a course of treatment has six or eight notes, AI can turn the series into a discharge summary or a letter to the referring clinician far faster than writing one from scratch. Paste in the de-identified notes, or use the letter feature inside your scribe or clinic software, and ask for a one-page summary: presenting problem, key findings at assessment and discharge, treatment given, outcome measure change, advice given, and any onward recommendation.
Before, the kind of letter a tired clinician writes on a Friday: "Seen for knee pain x6. Improved. Discharged with HEP."
After, the AI's draft (illustrative) once the physio had checked it: "Thank you for referring this patient with right anterior knee pain. At assessment, pain came on at around 60 degrees of squat, with reduced right hip abduction strength (4/5) and a LEFS of 58/80. Over six sessions we worked on hip and quadriceps strength and a graded return to running. At discharge, pain is 1/10 on running, hip abduction strength is 5/5 and LEFS is 74/80. She is running three times a week and continuing a home programme. No further physiotherapy is planned; we have advised her to get back in touch if pain rises above 3/10 for more than a week."
The check that matters: every number in the letter should appear in a signed note. If the AI writes a discharge score you never recorded, it has invented it, and a referrer will take it at face value.
The copy-forward mistake that shows up weeks later
Many clinic systems let you start a follow-up note from the previous one, and an AI draft built on top of that inherits everything in it. It saves time and carries a quiet risk: findings from session one survive into session four even though they have changed. An illustrative case: a scribe carried "mild effusion right knee" forward into three follow-up notes after the swelling had gone. Nobody noticed until a colleague covering a session read "effusion" and held back a progression the patient was ready for.
If your notes start from the last one, treat every carried-forward finding as unverified until you have re-checked it in this session. Better still, set the template so the objective section starts empty for each follow-up and only fills with what you narrate today.
Tools physiotherapists are using
Three broad options, and the right one depends on your clinic software:
- AI built into practice software. Jane, for example, includes an AI Scribe with five free notes a month per practitioner and an unlimited option at about $15 a month per practitioner. Built-in tools write straight into the chart, which saves the copy-and-paste step and keeps notes inside a system you already have data terms for.
- Standalone clinical scribes. General clinical scribes such as Heidi, which has a free tier alongside paid plans, and a growing number of physio-focused products. Many offer physio templates. Check whether they integrate with your practice software or rely on copy and paste.
- General chat assistants. Useful for the shorthand method and for writing templates, on a plan whose data terms suit clinical work, with no identifiable patient details in the prompt.
Prices and free-tier limits change often, so check them on the vendor's own pricing page. For a wider comparison of what each category actually does, see AI tools for physiotherapy clinics.
Twelve patients a day: the note time, timed
An illustrative physiotherapist sees twelve patients a day: two initial assessments and ten follow-ups. Timed over a week before AI, her notes took about ten minutes per assessment and five per follow-up, so roughly 70 minutes a day, much of it after the last patient.
After a month using an ambient scribe for assessments and dictation for follow-ups, her timing looked like this:
- Initial assessments: about 4 minutes each to review and correct the draft (8 minutes a day).
- Follow-ups: about 90 seconds of dictation plus 1 minute of review each (25 minutes a day).
- Total: about 33 minutes a day, a saving of around 37 minutes.
Two things made the difference beyond the tool. She narrated measurements from week two onwards, which cut her correction time by more than half. And she dictated each follow-up before calling in the next patient, rather than batching them after the last patient, because a note dictated three patients later is a note half-remembered. Your own numbers will differ; the useful step is timing a week before and a week after.
Consent, shared spaces and patient data
Tell patients before recording and let them say no without fuss: "I use a tool that types up my notes from our conversation; I check everything, and the recording is deleted afterwards. Is that OK?" For gym sessions with other patients nearby, dictation after the session is safer than ambient recording, because the people at the next station have not agreed to anything.
For the shorthand method with a general chat assistant, the rule is no identifiers: no names, dates of birth, addresses, or details unusual enough to identify someone (a professional athlete's club, a rare condition in a small town). Techniques for stripping identifiers are in anonymising client data before using AI. Whatever the tool, get the vendor's data retention and training terms in writing, and ask your data-protection adviser if anything is unclear. How ambient scribes handle audio and transcripts is explained in what an AI scribe is and how it works.
Checks before you sign a physio note
- Every range, grade, score and dose matches what you measured or prescribed.
- Sides and levels are consistent from subjective to plan.
- Every test listed was actually performed; delete any that were not.
- Outcome measures carry the right scale ("LEFS 58/80", "ODI 34%").
- The plan carries the dose and the progression rule, not just the exercise name.
- Advice you gave on load, pacing or red flags is recorded.
- Abbreviations expanded correctly, especially review intervals.
Run that list for a month and you will know which items your tool gets wrong. For most physios it is two or three of them, and those become the lines you read first.
Questions physios ask about AI notes
Can I use a general chat assistant instead of a physio scribe?
For shorthand-to-note drafting, yes, as long as nothing identifies the patient and your plan's data terms suit clinical work. Leave out names, dates of birth and anything unusual enough to identify someone. For recording sessions, use a tool built for clinical notes with a proper data agreement, rather than a consumer chat app.
Will AI notes stand up if a patient complains?
A note is judged on its content, not on how it was typed. An AI-drafted note that you have checked, corrected and signed is your note. The risk is signing a draft that contains something you did not do or say. Keep the same standard as handwritten notes: accurate, contemporaneous, and specific about what was assessed and advised.
How long does it take to get used to narrating measurements?
Most clinicians find it awkward for a week and normal after two. Saying the measurement aloud as you take it also helps patients understand their progress, and many like hearing it. Start with ranges of motion and strength grades only, then add special tests and outcome measures once that habit sticks.
Do AI scribes work in a busy open-plan gym?
Less well. Background voices, music and equipment noise lower transcription accuracy, and the scribe may pick up another patient's conversation. For gym-based sessions, dictation straight afterwards in a quieter spot usually produces cleaner notes than ambient recording, and it avoids recording people who have not consented.
Further reads
- Marketing a Physio Clinic With AI: Posts, Tips, and Newsletters — Use AI for posts, tips and newsletters once notes are under control.
- AI Scribe Costs Compared: What Small Clinics Pay in 2026 — What small clinics actually pay for AI scribes.
- 10 Admin Tasks a Small Clinic Can Hand to AI This Month — Other clinic admin jobs to hand to AI this month.
- How to Roll Out an AI Scribe Without Losing Patient Trust — Bring a scribe into the clinic without unsettling patients.
- What to Ask Before Buying Any AI Tool for a Medical Practice — Questions to ask any vendor before a clinic signs up.
- Is an AI Scribe Worth It for a Small Private Clinic? — Whether the time saved justifies the cost for a small clinic.
- How Small Vet Practices Use AI to Cut Time on Clinical Notes — Vet AI scribes, consult by consult: narrating the exam, weight-based doses to check, multi-pet visits, discharge notes, and a three-vet practice's numbers.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Jane App AI Scribe guides (free and unlimited plans); Heidi pricing page; published scales for the Lower Extremity Functional Scale and Oswestry Disability Index.