You didn’t become an occupational therapist to write notes until 7 p.m. But that’s the reality for a lot of OTs — the “second shift” of charting after the last client leaves, clicking through EMR fields, turning shorthand into proper progress notes, rewriting goals into something a parent can follow. Surveys keep finding that OTs spend at least as much time on documentation as they do on actual treatment.
So here’s a genuinely interesting development: a 2026 peer-reviewed study (Lee & Park, Digital Health) compared occupational therapy notes written by licensed OTs against notes drafted by ChatGPT across 15 standardized cases. Both OTs and patients/caregivers rated them. The AI-drafted notes scored significantly higher on completeness, correctness, and concordance — and on every measure of perceived empathy (all with p < 0.001). The humans still won on consistency between raters, which matters. But the headline is hard to ignore: for the drafting part of documentation, a well-prompted AI is now genuinely good.
The key word is drafting. This post shows you how to turn five bullet points into a clean progress note in about five minutes — while keeping the two things that have to stay yours: your clinical judgment, and your patient’s privacy.
Read this first: the de-identify rule
Before any workflow, the one non-negotiable: never paste identifiable patient information into consumer ChatGPT. No names, no dates of birth, no anything that could point to a specific person. Plain ChatGPT is not a HIPAA-covered tool, and inputs can be stored or used to improve the model.
This isn’t optional caution — it’s the line every OT online draws, and the one your licensing body (AOTA in the U.S., and equivalents elsewhere) holds you to: you remain responsible for accuracy and for protecting patient data. The fix is simple and it’s the whole trick to using AI safely in a clinical setting:
Write your notes in shorthand with no identifiers. Let AI structure the language. You add back the clinical judgment and keep the record.
So instead of “Jacob, DOB 3/14/19, struggled with…”, you type “pediatric client, fine-motor session, difficulty with…” The AI never needs to know who the person is to help you write a better paragraph about the session.
The 5-minute progress note
Step 1 — Brain-dump the session in de-identified bullets. Right after the session (or from your scribbles), type rough notes with zero identifiers:
- pediatric client, school-based OT, fine-motor goals
- worked on scissor skills + pencil grip, 30 min
- mod verbal cues for grip; min assist for cutting curved lines
- improved from last week (was max assist on cutting)
- tolerated full session, engaged with preferred Lego activity
- next: continue grip strengthening, introduce dynamic tripod
Step 2 — Ask for the structure you use. Paste this prompt:
Turn these session notes into a professional SOAP-format
progress note for an occupational therapy record. Keep clinical
terminology. Do not invent any detail I didn't provide. Flag
anything that seems incomplete.
That last sentence matters — it tells the model not to hallucinate assist levels or scores you didn’t give it, which is the main risk.
Step 3 — Edit for clinical accuracy, then it’s your note. Read every line. Did it state an assist level correctly? Did it add a detail you didn’t mention? Fix it. The AI gave you a clean first draft; the clinical reasoning, the corrections, and the sign-off are yours. That’s the “human in the loop” rule every OT lives by — and it’s exactly what the research says makes this safe: AI as a draft generator under clinician oversight, not a replacement for your judgment.
The other big time-saver: parent-friendly home programs
This might be the single best use. You finish an eval, you know exactly what the home program should be — and then you spend 20 minutes translating clinical language into something a caregiver will actually follow. Hand that to AI:
Rewrite this home program in warm, plain language for a parent,
at about a 6th-grade reading level. Keep it encouraging and
specific. Here are the activities: [de-identified bullets]
Pediatric OTs online swear by this — one trick that makes the rounds is asking for the home exercises wrapped in a theme the kid loves (“make this a Minecraft-themed routine”). Engagement goes up, and you didn’t lose an evening to it.
What this means for you
School-based OTs: IEP season and quarterly reports are the crunch. Build de-identified templates for your most common goal areas now, and the drafting collapses from hours to minutes.
Pediatric / clinic OTs: The parent-handout translation is your highest-value use. It’s repetitive, it’s plain-language work, and it’s exactly what AI does well — as long as the clinical content came from you.
Home-health / mobile OTs: Many already voice-record between visits. Dictate de-identified observations, then have AI structure them. Just confirm your recording tool and your AI tool both keep data out of training.
The skeptic: Fair. Start with the lowest-risk task — turning your own de-identified bullets into cleaner prose — and judge the output yourself. You’re not outsourcing the thinking; you’re outsourcing the typing.
What it can’t do
- It can’t make a clinical judgment. Assist levels, goal selection, whether a child is progressing — that’s you. The AI only knows what you tell it.
- It can’t be trusted with PHI. Consumer ChatGPT is not HIPAA-compliant. De-identify, every time, no exceptions.
- It can’t catch its own invented details. It will sometimes add a plausible-sounding fact you never gave it. Reading every line isn’t optional.
- It can’t replace the relationship. The study found AI notes read as more empathetic — but empathy in a record isn’t the same as the trust you build in the room. That stays human.
- It isn’t a substitute for your board’s rules. Check AOTA guidance and your state’s documentation standards.
The bottom line
The 2026 research settled the “is the writing good enough?” question — for drafting, it is. What it didn’t change is the part that was always yours: the clinical reasoning, and the duty to protect your patient. Keep identifiers out, keep judgment in, read every line — and documentation stops being the reason you’re still at your desk at 7 p.m.
Want the full safe workflow, including the HIPAA side done properly? AI Therapy Notes: The HIPAA-Safe Workflow walks through de-identification and note-drafting step by step, and AI for Healthcare Workers covers the broader picture for clinical settings.
Sources
- Lee & Park — AI in occupational therapy documentation: chatbot vs occupational therapists (Digital Health, PMC)
- JMIR AI — A language model for pediatric occupational therapy documentation (2026)
- Değerli et al. — Comparing ChatGPT and OT approaches in assessment and intervention (SAGE, 2026)
- TheraPlatform — AI for occupational therapy documentation
