Minal Patel, PT, Director of Clinical Solutions, SPRY
LinkedIn: Minal Patel PT DPT, OCS
Brijraj Bhuptani, Co-founder and CEO
LinkedIn: Brijraj (Vaghani) Bhuptani
LinkedIn: SPRY PT
There is a moment every Physical Therapist knows.
Your patient is on the table. They are telling you something important about their pain, their progress, the thing they finally did last weekend that they hadn’t been able to do in months. And somewhere in the back of your mind, a quiet voice is already composing the note.
Not fully present. Not completely there. One foot in the room, one foot already at the computer.
This is what documentation has done to the profession. Not dramatically. Not all at once. Gradually, visit by visit, note by note, it has pulled therapists out of the room while they are still standing in it.
The AI documentation tools that arrived over the last several years promised to change this. Many of them delivered something useful: a faster draft, a cleaner starting point, some recovered time at the end of the day. For primary care and hospital medicine, ambient scribing was genuinely transformative.
In Physical Therapy, the same tools ran into something they were not built for.
Rehab therapy documentation is not primarily a transcription problem. It is a context problem. And that distinction is what the industry largely missed.
What the research is starting to show
A peer-reviewed study published in BMJ Digital Health and AI in September 2025 evaluated seven commercial AI scribes across clinical scenarios and found that none produced error-free summaries. Omissions dominated, accounting for 83.8 percent of all errors. More importantly, performance varied significantly by clinical scenario. Structured, straightforward consultations scored well. Complex, ambiguous ones broke down.
This matters because Physical Therapy is almost never the straightforward scenario.
As of August 2025, there are now 88 commercial AI scribe products, up from 40 just 16 months earlier. Most were built for general clinical use. Most were not designed with the specific documentation demands of outpatient rehab in mind.
Why PT documentation is different
A progress note in Physical Therapy does not stand alone. It exists to document change over time, relative to where the patient was at initial evaluation, across a full episode of care that might span three months and thirty visits. To write a defensible, clinically accurate progress note, the therapist needs to know the baseline range of motion measurements from evaluation, what functional goals were established and whether they are being met, what changed between last week and this week, and whether the trajectory of recovery supports continued care.
A generic AI scribe knows none of this. It meets the patient fresh at every session. It transcribes what happened today with no access to what happened six weeks ago, no understanding of the clinical trajectory the therapist is managing, and no awareness of what this particular payer requires in order to approve the next ten visits.
The result is a draft. A useful draft, sometimes. But the therapist still carries all the longitudinal reasoning, reconstructs it at the keyboard after each session, and then reviews the output for accuracy. First-generation scribes made documentation faster. They did not change the experience of writing notes. The therapist bears the cognitive load for the quality control layer for every note, every visit, every day.
And so the presence problem remained.
What changes when the AI actually knows you
Something shifts in the clinical encounter when documentation stops competing with it.
Therapists describe being more present with patients. Not glancing at the screen. Not mentally drafting language while the patient is still talking. The session can be the session.
One therapist described a shift she could not put a metric on: when a patient goes on a long tangent, her instinct used to be to control the narrative. She had notes to write. She had other patients. Now she lets it go. She described the feeling as being more free during the eval. More present. Better at the job she trained for.
Another described catching a vascular presentation in a complex patient because the AI had captured detailed symptom language she would have condensed to a single word while typing fast. The specificity of what the AI recorded enabled a referral that changed the patient’s outcome.
These are not efficiency stories. They are presence stories. And they only become possible when the system carries the patient’s full clinical history into every session, understands the specialty-specific form structure, and has learned the individual clinician’s voice and preferences. When a correction happens, it applies going forward. When a preference is stated once, it is not repeated.
That is a fundamentally different working relationship with technology.
What the profession needs
APTA issued a formal practice advisory on AI-enabled ambient scribes in 2025, the first time the association formally addressed the technology in physical therapy practice. The guidance acknowledges both the potential and the limitations clearly, and emphasizes that the tools work best when built for the clinical specificity of the specialty, not adapted from general-purpose models.
The Physical Therapy profession is at an inflection point. Nearly 50 percent of Physical Therapists report experiencing burnout, and among all healthcare specialties studied, therapists report the highest rates of mental fatigue. APTA projects more than 26,000 PT positions will remain unfilled due to workforce shortages. Every therapist who burns out and leaves is not just a personal loss. It is an access-to-care problem for the patients who need them.
The documentation burden that drives burnout has not been treated with the seriousness it deserves. And the technology that promised relief has delivered speed without fundamentally changing what the work feels like.
The more interesting question, the one the industry has not asked seriously enough, is whether AI can change the experience of documentation itself. Not make it faster. Make it feel like less of a departure from the actual job.
The goal is not AI that gives you a better draft to edit. It is AI that is so accurate, so aware of your patient presentation, so fluent in your clinic’s form structure, and so calibrated to how you specifically document, that the note is simply there when the session ends. Ready to review. Ready to sign.
Walk into the room. Be with your patient. Walk out. The note is done.
That is the problem worth solving. And it is a different problem than the one most of the industry has been working on.