AI for Physical Therapists

Updated January 2026 | 9 min read

You're seeing twelve patients today. Each one is at a different stage in their treatment plan. The rotator cuff repair patient is four weeks post-op and ready to start resistance exercises. The runner with IT band syndrome has been doing foam rolling but still can't run more than two miles. The post-stroke patient improved their gait pattern last week and you want to build on that progress.

You remember most of this. You check your EMR notes before each patient walks in. But the details that matter—the specific exercise that caused pain last time, the home environment challenges the elderly patient mentioned, the work deadline that's making the office worker skip their exercises—those details aren't always in the chart.

Standard AI tools don't help with this continuity. ChatGPT can explain exercises or look up treatment protocols, but it doesn't remember that this specific patient can't do lunges because of knee crepitus, or that the last progression you tried didn't work because their apartment doesn't have room for resistance bands.

Exercise Progression That Builds on History

Every patient progresses differently. The protocol says to advance from partial weight bearing to full weight bearing at six weeks. But this patient had complications. That patient exceeded expectations. Another patient is technically meeting the timeline but compensating with poor movement patterns.

You adjust progressions based on what you observe. You tried eccentric calf raises last week, but the patient had increased pain for three days. You modified to bilateral heel raises instead, and those worked well. This week you're ready to progress again, but you need to remember that eccentric loading didn't go well.

Your EMR might note "patient reported pain," but it doesn't capture the decision tree. AI with memory tracks not just what you prescribed, but what worked, what didn't, and why you made each modification. When you're planning this week's progression, you have the full context.

Patient-Specific Modifications

The textbook shows exercises with ideal form. Your patients have limitations. One has limited shoulder flexion from an old injury. Another has balance issues that require holding onto a counter. A third has arthritis in their hands that makes gripping equipment difficult.

You develop modifications for each patient. Use a towel instead of a resistance band for the arthritis patient. Do single-leg stance exercises near a wall for the balance patient. Adapt the shoulder exercise to work within available range of motion.

These modifications aren't standard—they're specific to each patient's combination of limitations and goals. AI that remembers these adaptations means you don't have to recreate them every session. You build on what already works.

Home Exercise Program Compliance Patterns

You prescribe home exercises. Some patients do them religiously. Some do them occasionally. Some don't do them at all but say they did.

You learn who's compliant and who isn't. More important, you learn why. The morning person does exercises before work without fail. The parent of three young kids can't find time during the day but will do exercises during kids' screen time. The chronic pain patient does exercises when pain is moderate but skips them when pain is high or low.

Knowing these patterns changes how you structure home programs. Give the busy parent three exercises instead of eight. Make sure the chronic pain patient has a modified "bad day" version of their program. Schedule the morning person's progressions to start on Monday so they have the weekend to prepare.

Without memory, you're prescribing generic programs and hoping for compliance. With memory, you're designing programs around each patient's actual life and behavior patterns.

Pain Patterns and Movement Compensations

A patient reports knee pain. Where exactly? Is it sharp or dull? Does it happen during the exercise or after? Is it the same pain as last week or different? Did it show up during loaded exercises, unloaded exercises, or daily activities?

These details matter for diagnosis and treatment decisions. Pain during loaded knee extension suggests different issues than pain after exercise. Sharp anterior knee pain has different implications than dull posterior pain.

You observe compensations too. The patient is hiking their shoulder during arm raises. They're shifting weight to one leg during squats. Their knee is collapsing inward during step-downs. You cue corrections. Some compensations resolve. Others persist.

Tracking these patterns over time shows you what's improving and what needs more focus. AI with memory holds the detail that your EMR doesn't capture—the specific compensation pattern, the cueing that worked, the exercise modifications that reduced unwanted movement.

Insurance Authorization and Documentation

Insurance companies authorize a specific number of visits. Some patients have 20 visits covered. Others have 12. Some plans require reauthorization at six visits with progress documentation. Others need functional outcome measures submitted at evaluation and discharge.

You track authorized visits remaining. You know which patients are approaching authorization limits and need progress notes submitted. You remember which insurance companies accept phone requests versus requiring written documentation.

Medicare has specific documentation requirements. You need to show skilled service, not maintenance. You document medical necessity. You note why this treatment requires a physical therapist's expertise rather than a home exercise program.

This administrative knowledge is part of running a practice, but it's not what you learned in PT school. AI that remembers insurance requirements per patient and per plan keeps you compliant without constant reference to policy manuals.

Surgical Protocol Variations by Surgeon

Post-surgical protocols vary by surgeon. Dr. Anderson wants rotator cuff patients in passive range of motion only for six weeks. Dr. Martinez starts active-assisted ROM at four weeks. Dr. Kim allows active ROM at three weeks if the tear was small.

These aren't just preference differences—they're based on surgical technique. Dr. Anderson uses a different repair construct. Dr. Martinez's patients have better initial tissue quality because of his patient selection. Dr. Kim's approach involves a different rehabilitation philosophy.

You need to know these differences. When a post-op patient comes in, you check who did the surgery, not just what surgery they had. AI with memory holds surgeon-specific protocols. You don't have to look up Dr. Anderson's preferences for the tenth time this year. You know them.

Complication Patterns by Procedure

Some procedures have predictable challenges. ACL reconstruction patients often struggle with quad activation. Total knee patients frequently have difficulty with terminal extension. Ankle fracture patients commonly develop stiffness that's hard to resolve.

You've seen these patterns enough times to anticipate them. You start quad activation exercises early for ACL patients. You're aggressive about terminal extension work after total knee. You get ankle fracture patients moving early to prevent stiffness.

AI that remembers these patterns—both the published ones and the ones you've observed—helps you design better initial treatment plans. You're proactive about common complications instead of reactive when they appear.

Outcome Measure Tracking

You administer outcome measures at evaluation, reassessment, and discharge. Lower Extremity Functional Scale. QuickDASH. Oswestry Disability Index. Numeric Pain Rating Scale. Global Rating of Change.

These scores matter for insurance, for demonstrating progress, and for your own assessment of treatment effectiveness. But they're snapshots. A score of 45 on the LEFS means something different at week two versus week eight. A patient reporting 4/10 pain is different if last week was 7/10 versus if last week was 3/10.

Your EMR stores these scores. It might even graph them. What it doesn't do is connect the scores to what was happening in treatment. The LEFS score jumped up the week you added plyometric training. The QuickDASH score plateaued despite continued therapy—but that was when the patient went back to work and stopped doing home exercises.

AI with memory connects outcome data to treatment decisions and life context. You see not just that progress happened, but what drove it or stalled it.

Clinic Logistics and Scheduling Considerations

Some patients need 45-minute sessions to get through their program. Others do well with 30 minutes. Some need to be scheduled first thing in the morning before they stiffen up. Others prefer end-of-day appointments after work.

You have equipment constraints too. If you schedule three patients with similar protocols back-to-back, you don't have enough equipment. If you book two patients who both need the parallel bars at the same time, someone has to wait.

This knowledge helps you schedule efficiently. AI that remembers each patient's time requirements and equipment needs lets you plan better. You reduce wait times, use equipment efficiently, and keep sessions running on schedule.

Return to Sport or Work Requirements

A patient's goal is to return to soccer. Another needs to lift 50-pound boxes for their warehouse job. A third wants to hike without knee pain. These goals determine your treatment approach.

You test sport-specific or work-specific activities as patients progress. The soccer player needs to pass agility drills, cutting maneuvers, and ball handling without pain or instability. The warehouse worker needs to demonstrate safe lifting mechanics and adequate endurance. The hiker needs to manage inclines and declines with good form.

You document when these milestones are met. You also document when patients are pushing to return before they're ready. The athlete wants to play this weekend but hasn't passed your functional tests. The worker's employer is pressuring them to return but they can't perform the job safely yet.

AI with memory tracks both the functional requirements and the patient's readiness. You have clear criteria for return-to-activity decisions, not just guesswork.

The Setup for Physical Therapy Practice

One markdown file. CLAUDE.md sits in a synced folder on your work device. Inside that file: patient-specific progressions, home exercise compliance patterns, surgeon protocols, insurance requirements, outcome measure tracking, equipment and scheduling notes.

Every time you use Claude Code, it reads that file first. The AI knows your patients' histories, your treatment approaches, the modifications you've developed. You don't re-explain why this patient can't do standard progressions. You don't look up Dr. Anderson's post-op protocol again.

When you learn something new—a progression that works well, a compensation pattern that predicts outcomes, an insurance quirk—you add it to the file. Your clinical knowledge grows with your experience.

No EMR integration. No database. No technical setup. One file gives AI persistent memory across every session.