Gains:
- Ability to draft session notes in fast, consistent and complete SOAP format with AI
- Ability to maintain elements of documentation that meet legal, professional and reimbursement requirements
- Ability to verify the note produced by AI with the patient's real session and eliminate the risk of fabricated findings
Every physiotherapy session requires a recording. Session note; It ensures continuity of clinical follow-up, facilitates communication within the team, offers legal protection and is required for reimbursement (insurance/institution). But writing notes is a significant source of time and mental load for the physical therapist who sees many patients per day; It is often postponed until the end of the day and detail is lost. Artificial intelligence (AI) eases this burden by converting session information into a fast, consistent and complete note. But the biggest risk of AI in documentation is that it fabricates a finding that doesn't exist. In this unit you will learn how to write a secure, complete and verifiable SOAP note with AI.
Anatomy of the SOAP note
SOAP consists of four parts, each serving a different purpose:
- S — Subjective: The patient's own statement. Pain level, complaints, change in daily life, goal. Example: "The patient stated that morning stiffness had decreased and he still had pain when descending stairs (4/10)."
- O — Objective: Measurable, observable findings. ROM (range of motion), muscle strength (MMT), special tests, edema, applied treatment. Example: “Knee flexion ROM 110°, quadriceps MMT 4/5.”
- A — Assessment: Physiotherapist's clinical interpretation; progress, obstacles, clinical wisdom. This part is human judgment.
- P — Plan: Next step; schedule change, frequency, target, referral.
AI organizes this structure perfectly, but you must provide the content of each section. Particularly part A is your clinical mind.
Attention: The documentation is a legal document. If a finding fabricated by AI (a test not performed, a value not measured) enters the patient's official record, it creates serious clinical, legal and reimbursement problems. Rule: AI formats only the session information you transmit; does not add information.
Step by step: SOAP annotation with AI
- Take raw notes during/immediately after the session. Short keywords are enough; But let's be real.
- Anonymous and structured data. Transfer raw note to AI; identification.
- Make the rule "Only use what I give." Don't make up the missing field; Type "not specified".
- Receive draft in SOAP format. With the distinction of subjective/objective/evaluation/plan.
- Verify one-on-one. Compare each finding in the note with the actual session; If it's fake, delete it.
- Confirm assessment (A). The clinical interpretation is yours; Even if AI suggests, you have the final say.
- Check refund/legal areas. Are all required mandatory fields filled?
Three mini cases (in numbers)
Case 1 — Time savings. In one clinic, a physical therapist saw 12 patients per day and spent an average of 6 minutes on each note (~72 minutes per day). By having AI convert raw keynotes to SOAP, the time per note was reduced to 2 minutes (~24 minutes per day); The time saved was allocated to the patient. All notes were recorded after verification.
Case 2 — Fake findings. In the raw note, the physical therapist had only written "shoulder ROM increased, home program given." AI filled the note with unmeasured values such as “shoulder abduction 160°, external rotation 45°.” The physical therapist deleted these because he had not measured angles that session; it just wrote "observationally ROM increased." Lesson: AI fills the gap with numbers; verification is required.
Case 3 — Subjective/objective distinction. The AI put “patient feels better” in the objective section. The physical therapist moved this to the subjective (S) section because it was the patient's statement, not the measurement. He left in the objective section only the values he actually measured. The minor distinction preserved the clinical and legal value of the note.
Weak prompt / Strong prompt
Weak: "Write a SOAP note for this patient's knee session."
No information; AI produces a memo that seems reasonable but is full of fabricated findings.
Strong: "Convert my raw session note below into SOAP format. Use ONLY the information I provided; do not add any values I did not measure, write 'not specified' in the missing field. Put the patient's statement at S, the measurement at O. Write the evaluation (A) section as a suggestion and I will approve. Raw note: [notes]."
Four copyable templates
Role: Clinical documentation assistant (you do not produce information).Task: Translate my raw session note into SOAP.Raw note: [keywords / memo]Rule: ONLY use what I give. Write "not specified" in the missing field, FITTING.S=patient statement, O=measurement/observation, A=suggestion (I will confirm), P=plan.
Task: Check this SOAP note for section correctness.Note: [draft]I want: Check if each statement is in the correct section (S/O/A/P).Check those placed incorrectly. For each number in the objective section, "is the source stated?" ask the question. Adding new information.
Task: Reimbursement/legal completeness check.Note: [SOAP draft]Required fields: [date, diagnosis, treatment administered, duration, response, plan, signature]I want: List missing mandatory fields. Don't make decisions, just point out the shortcomings.
Task: Make a progress summary from one-week session notes. Notes: [SOAP notes series] I want: Summarize the change in objective values over time (only with the numbers written in the notes). Don't make things up; If there is no note, write "no data".
Documentation risk control chart
Risk
symptom
precaution
fabricated finding
There is a value in the note that you did not give
"Just use what I give" rule + one-to-one verification
Department meddling
The patient's statement is in O
Check S/O/A/P discrimination
Leaving clinical interpretation to AI
Part A is independent of your mind
Get A recommendation, confirm it yourself
legal missing
Required field is empty
Completeness check tour
Personal data leak
Identity in raw note
Process by anonymization
Tip: Always complete the SOAP note in two rounds: the first round the AI formats, the second round you verify verbatim and confirm the clinical interpretation. This two-lap discipline maintains speed gains without compromising safety.
An often forgotten function of good documentation is to make clinical reasoning visible. A memo should reflect not only "what was done" but also "why it was done": why this program was chosen, why this change was made, what findings supported this decision. This strengthens both continuity of treatment (understands the logic when another physiotherapist takes over) and legal protection. AI can turn your messy justifications into a neat “evaluation” paragraph; but the rationale itself must come from your clinical mind. A blank, formulaic assessment (“patient progressing, continuing program”) is a poor mark; An assessment that reflects true clinical judgment makes the grade stronger. Use AI to beautify your justification, not to generate justification.
Voice recording, transcription and privacy
An increasingly common practice is to take notes by speaking during or immediately after the session and convert it to text (transcription) with AI. This further reduces the writing burden: the physiotherapist speaks the findings out loud, the AI translates it into a SOAP note. However, this convenience comes with a serious confidentiality responsibility. The patient's voice and spoken content are private data; The patient must know and approve that his speech will be recorded. The record should be kept in a secure, encrypted stream that is deleted when done. Additionally, transcription errors can turn into clinical errors: a misheard note may be passed, such as "right" instead of "left", "55 degrees" instead of "45 degrees". Therefore, the note produced from the transcription must be verified verbatim with the memory of the raw recording. Convenience is not a substitute for verification.
Tip: Always say the side (left/right) and numeric values twice in the voice memo and then check them separately in the written memo. It is precisely at these two points that transcription errors occur most frequently, and they are the most clinically critical.
Common mistakes
- Asking for a note without giving information. AI fills the gap with fabricated findings.
- Not making the "make it up" rule. Without explicit bans, AI fills in missing areas.
- Skipping in-person verification. Every finding must match the actual session.
- Leaving clinical interpretation to AI. Assessment (A) is the judgment of the physical therapist.
- Confusing subjective/objective. This distinction determines the clinical and legal value of the note.
- Processing personal data without anonymization. The raw note should not contain identification.
In summary
AI drafts SOAP notes quickly, consistently, and completely, dramatically reducing documentation burden. But the biggest risk is fabricated findings. Require use of only actual session information, verify each finding verbatim, preserve the subjective/objective distinction, self-validate clinical interpretation, and check for legal/reimbursement completeness. Two rounds of discipline (AI formats, you verify) ensures both speed and security.
Application task
Write a short raw note (keywords) of an actual session. Convert to SOAP with the first template, but deliberately leave out one measurement; Check if the AI is making it up. Check section correctness with the second template and check legal completeness with the third template. If you find a fitting finding, reinforce the rule.
checklist
- [ ] I only provided real, anonymous session information to the AI.
- [ ] I made the rule "Don't make up the missing field, write it as not specified".
- [ ] I verified every finding in the note verbatim with the actual session.
- [ ] I checked the subjective/objective/evaluation/plan distinction.
- [ ] Evaluation I approved part (A) myself.
- [ ] I have checked the legal and refund mandatory fields for completeness.