Physical therapy documentation is where a lot of good clinicians burn out. The evaluation, the plan of care, the progress report every tenth visit, the recertification, the daily note that has to prove skilled care to a payer looking for a reason to deny. AI can draft every one of those from your findings, in your clinic's format, in the time it takes to walk to the next patient.
The limits matter more here than in most jobs. A chat assistant cannot examine a patient, cannot pick or progress an exercise, and does not know your state's practice act or the payer's current policy. It will write a confident justification for a plan it has never seen work, and it will invent an outcome-measure cutoff score if you let it.
And it cannot hold protected health information. Patient names, dates of service, videos of a patient walking, and the other HIPAA identifiers stay inside your EMR and any documentation tools your practice has licensed under a Business Associate Agreement. General assistants get the fictional case, the de-identified pattern, the payer's public policy, and the handout that needs a rewrite.
Quick wins this week
- Paste your clinic's evaluation template and a fictional patient and ask for a model evaluation with measurable, functional, time-bound goals, then compare the language to your last three evals.
- Rewrite the instructions for a home exercise program at a sixth-grade reading level, with the 'stop and call us if' section your patients actually read.
- Draft an appeal for a denied claim using the payer's own written policy and a de-identified summary of the skilled need and the progress data.
- Turn a research article on a technique you use into a one-paragraph patient explanation and a three-bullet summary for your team, then check both against the paper.
- Build a continuing education plan around your state's renewal requirements and the topics you keep postponing.
What AI can do for physical therapists, task by task
Evaluations and plans of care from your findings
Dictate or type your de-identified subjective and objective findings in shorthand and ask for the evaluation in your template, with goals that are measurable, functional, and time-bound. Check for findings it added that you did not assess: models fill in range-of-motion numbers and strength grades to make a note look complete. The evaluation complexity code (97161 to 97163) has to be supported by what you documented about history, examination, and presentation, so choose it yourself.
Daily notes and progress reports that show skilled care
Medicare wants a progress report at least every ten treatment days and a plan of care recertified at least every 90 days, and every payer wants to see why a therapist was needed rather than a printed exercise sheet. Paste a fictional note and ask what a reviewer would question, then ask for a rewrite that states the skilled decision you made that visit. Watch for cloned language across notes; identical paragraphs are an audit flag, and a model given the same template will produce them.
Home exercise programs and patient handouts
Keep the exercises and the images in a licensed HEP platform such as MedBridge or Physitrack, and use the assistant to rewrite the instructions in plain language, explain why each exercise matters to the patient's goal, and add the stop rules. You choose the exercises and the dosage; the model formats. Verify that every set, rep, and hold you entered came back unchanged, and never use generated images for exercise instruction, because the anatomy is unreliable.
Denials, appeals, and prior authorizations
Give the model the denial reason, the payer's published policy, a de-identified summary of the skilled need, and the outcome-measure trend, and ask for a letter that addresses each policy criterion in order. Know the rules the model does not: the KX modifier once a patient passes the annual therapy threshold, and the Jimmo settlement, which established that Medicare coverage does not depend on improvement when skilled care is needed to maintain function. Verify every citation.
Payer rules, the practice act, and research questions
Upload the therapy sections of the Medicare Benefit Policy Manual, your state practice act, and the payer policies you fight most into NotebookLM and ask questions that must be answered from those documents with the passage shown. For clinical evidence, use Perplexity to find candidate studies, then read the paper; summaries overstate findings. Check the date on every policy, because the model cannot tell you what changed this year.
Patient communication, scheduling, and the front desk
Draft the plan-of-care explanation, the missed-visit follow-up, the discharge letter, and the cash-pay estimate in plain language, with fictional details in a general tool and merge fields added in your EMR. Self-pay patients are owed a Good Faith Estimate under the No Surprises Act, so keep that template accurate to your fee schedule. Check tone; a model's default is either stiff or oddly cheerful, and neither sounds like your clinic.
Mentoring new grads, PTAs, and students
Ask for fictional cases at a stated level with clinical-reasoning questions, a competency checklist for a new hire, or a supervision log template that matches your state's PTA supervision rules, which you paste in because they vary. Everything is fictional by design, which makes this one of the safest uses. Have a senior clinician review scenarios before they count toward a formal competency.
Prompts for physical therapists
Replace the bracketed placeholders, paste into any chat assistant, and iterate on the result.
Evaluation draft from de-identified findings
Act as an experienced outpatient physical therapist who writes clear, defensible evaluations. Region or diagnosis: [BODY REGION OR REFERRING DIAGNOSIS]. De-identified subjective and objective findings, in my shorthand: [DE-IDENTIFIED FINDINGS]. Outcome measure and score: [OUTCOME MEASURE AND SCORE]. Planned frequency and duration: [FREQUENCY AND DURATION]. My template: [PASTE YOUR EVALUATION TEMPLATE]. Write the assessment, the clinical impression, and three to five goals that are measurable, functional, and time-bound, in my template. Do not add any measurement, test result, or history item I did not provide. At the end, list everything you inferred or assumed so I can correct it. Do not select an evaluation complexity code; I will do that.
Tip: The 'list everything you inferred' line is the safety net. Read that list before you read the note.
Skilled-care critique of a daily note
You are a Medicare documentation reviewer for outpatient therapy. Here is a fictional daily note with no real patient data: [PASTE A FICTIONAL DAILY NOTE]. Identify every place a reviewer would question whether skilled physical therapy was needed that visit, whether the note shows progress toward the plan-of-care goals, and whether the treatment matches the billed codes [CPT CODES BILLED]. Then rewrite the note in the same length so the skilled decision I made that visit is explicit. Do not add any treatment, finding, or time that was not in the original.
Tip: Run this on a few notes from your own template (rewritten as fictional) and you will see the same weak spots repeat. Fix the template, not only the note.
Home exercise program instructions in plain language
Act as a physical therapist writing for patients. Exercises and dosage I have already chosen: [PASTE EXERCISE NAMES WITH SETS, REPS, HOLDS, AND FREQUENCY]. The patient's main goal in their own words: [PATIENT GOAL]. Reading level: [READING LEVEL]. For each exercise write two or three plain sentences on how to do it, one sentence on why it matters for the goal, and one common mistake to avoid. Keep every number exactly as I gave it. Add a short 'stop and call the clinic if' section and a line about normal soreness versus pain that should not be pushed through. No images, no medical claims beyond what I provided.
Tip: Paste the result into your HEP platform next to the licensed images. Check every number twice; models transpose sets and reps.
Appeal letter built from the payer's policy
You are a physical therapist appealing a denied claim. Denial reason as stated by the payer: [DENIAL REASON OR CODE TEXT]. Payer policy, pasted from the published document: [PASTE THE PAYER'S POLICY]. De-identified summary of the condition, the skilled interventions, and why they required a therapist: [DE-IDENTIFIED SUMMARY]. Progress data: [OUTCOME MEASURE SCORES OVER TIME]. Write an appeal under 450 words that takes each relevant criterion in the policy, quotes it, and shows how the summary meets it. Cite nothing outside the policy and the summary. If a criterion is not clearly met, tell me separately. Formal tone, no criticism of the payer.
Tip: Finish with real dates and identifiers only inside your EMR or an approved tool. The chatbot never sees the actual patient.
Plan-of-care conversation script
Act as a physical therapist who is good at explaining the plan in the first visit. Condition in general terms: [CONDITION]. Plan: [FREQUENCY AND DURATION] with [MAIN INTERVENTIONS IN PLAIN TERMS]. The patient's concern: [PATIENT CONCERN, E.G. COST, TIME, OR DOUBT THAT IT WILL WORK]. Write a 90-second explanation at an eighth-grade reading level that covers what we will do, why this frequency, what they do at home, and how we will measure progress. Then give one teach-back question and two short answers to the concern. No outcome promises, no clinical claims beyond what I gave you.
Tip: Practice it out loud once. Then cut anything you would never say.
Want a prompt for something else? Use the Prompt Builder.
Skills to build
Knowing what PHI looks like in a therapy clinic
Why: Dates of service, medical record numbers, and full-face photos are on HIPAA's Safe Harbor identifier list, and a gait video or a before-and-after clip is identifying even with the name cropped out.
How: Keep the identifier list next to your documentation station. Any video, photo, or note with a real patient stays in the EMR or an approved tool, and every prompt to a general assistant starts from a fictional version.
Feeding the model your findings, not asking it for findings
Why: A model given a diagnosis and a template will fabricate a plausible exam. A model given your actual measurements and told to add nothing will structure them well.
How: Dictate findings in your own shorthand first, then prompt. Add 'do not add any measurement I did not give you' to every documentation prompt.
Reading for the invented number
Why: The dangerous errors in generated notes are the ones that look right: a strength grade you did not test, a range-of-motion value it filled in, a goal it wrote for a deficit you never measured.
How: Compare the draft to your raw findings rather than reading it on its own. Anything in the note that is not in your findings gets deleted or measured next visit.
Grounding prompts in payer policy and the practice act
Why: Coverage rules, supervision requirements, and direct-access limits vary by payer and state, and the model's memory of them is stale or blended.
How: Keep a folder with your state practice act, the Medicare therapy manual sections, and the policies of your top payers. Paste the relevant section into the prompt instead of asking the model what the rule is.
Learning your EMR's AI features and their review step
Why: Therapy EMRs are adding ambient documentation and note drafting under their own BAAs, and these tools drop findings and misattribute statements just like their hospital cousins.
How: Take the vendor's training, learn the consent script for recording (all-party consent states require it), and read every generated note against the visit before you sign it.
Tools worth knowing
Microsoft Copilot
AI inside Word, Excel, Outlook, and Teams, with your company's data protections.
Claude
A careful writing and analysis assistant that shines on long documents.
ChatGPT
The general-purpose AI assistant most of your coworkers already use.
NotebookLM
A research notebook that only answers from the sources you give it, with citations.
Canva
Design for non-designers, with Magic Studio AI for text, images, video, and resizing.
Cautions for physical therapists
Consumer chatbots have no Business Associate Agreement, and a prompt with a name, a date of service, a photo, or a video is a disclosure of PHI. Never paste confidential patient data into a consumer AI tool unless your practice has approved that tool and signed the agreements. Work from fictional or fully de-identified cases, and finish anything with real detail inside the EMR or an approved tool.
Exercise selection, progression, manual therapy choices, red-flag screening, and discharge decisions come from your examination and your reasoning. A model has not seen the patient move, does not know today's evidence, and will justify whatever plan you describe. If it gives you an idea on a fictional case, verify it in a current source before it reaches a treatment room.
A generated note that includes a test you did not do, minutes you did not spend, or cloned language from last visit is a compliance problem the moment it supports a billed code. Medicare and commercial audits look for exactly those patterns. Read every draft against what happened, keep timed-code minutes honest, and sign only what is true.
A model's knowledge stops at a cutoff, and it blends Medicare rules with commercial ones and one state's direct-access or PTA-supervision rules with another's. Paste the current policy or statute into the prompt, check the effective date, and confirm anything that affects billing with your compliance contact or your state chapter.
AI-generated exercise images routinely show wrong joint positions and impossible anatomy, so use licensed HEP libraries for every picture. Text has its own trap: a handout can read as friendly and still miss the one warning that matters, so compare generated instructions to your protocol line by line before they go home with a patient.
Your 30-day plan
- Week 1: Ask your practice owner, compliance contact, or EMR vendor which AI tools are approved and what may be pasted into each. Read the HIPAA identifier list. Use a general assistant for two non-patient tasks: a CEU plan and a new-hire checklist.
- Week 2: Rewrite three of your most-used home exercise program instruction sets at a sixth-grade reading level and load them into your HEP platform. Draft one evaluation from a fictional case in your template and compare it to your real ones.
- Week 3: Build an appeal from a payer's published policy using a de-identified summary and verify every citation. Load your state practice act and the Medicare therapy sections into NotebookLM for a week of lookups.
- Week 4: If your EMR has ambient documentation or note drafting, complete the training, learn the consent script, and use it for a week while reading every note before signing. Note the error types you catch.
- End of month: Write down the three tasks where AI saved real time, the one place it was confidently wrong, and what you want to standardize across the clinic.
Frequently asked questions
Will AI replace physical therapists?
Can I use ChatGPT to write my PT notes?
Is it okay to use AI for Medicare documentation?
Can AI create a home exercise program?
Do I need patient consent to use an AI scribe in the clinic?
Terms used on this page
Related roles
- AI for PhysiciansThe documentation and inbox load that follows you home is where AI pays off first. It drafts; you decide, verify, and sign, and nothing with PHI leaves the systems your organization has covered under a Business Associate Agreement.
- AI for NursesAI can take the writing, summarizing, and studying load off your shift without ever touching a patient record, as long as you know where the HIPAA line is and stay on the right side of it.
- AI for Healthcare AdministratorsAdministrators drown in documents: policies, payer correspondence, board decks, survey prep, budget narratives. AI drafts and summarizes all of it, if you keep PHI inside covered systems and set the rules for your organization before someone else sets them for you.
- AI for Medical Coders and BillersAI can read a note faster than you can, and it will also cheerfully assign a code that was deleted two years ago. Used well, it speeds up the reading, the denial letters, and the research; the code selection and the compliance stay yours.