The therapy documentation problem isn't writing. It's structure.

Ask a physical therapist what they like least about the job and documentation is near the top of the list. That's not new. What's newer is a clearer picture of which part of documentation actually hurts, and it isn't the part most software is trying to fix.

The measured burden

A 2024 study in the Journal of the American Medical Informatics Association followed outpatient rehabilitation therapists, physical therapists, occupational therapists, and a speech-language pathologist, through their documentation day. A few numbers stand out. The physical therapists carried a median caseload of 44 patients a week, each visit generating its own note. And the documentation didn't get done in the moment: among the adult-side therapists, an average of 74% of encounters were closed the same day, and on the pediatric side that dropped to 9%. The rest gets carried home, or carried forward.

That "carried home" part has a name in the broader clinician literature: pajama time, the after-hours record work that shows up again and again as a driver of burnout and lower professional satisfaction. One therapist in the study described the depersonalization directly: "almost like I'm watching myself treat patients."

None of that is surprising. Here's the part that is.

Where the burden actually concentrates

The same study measured how much of the objective data therapists collect, the range-of-motion numbers, the manual muscle testing grades, the standardized scale scores, actually made it into the chart as discrete, structured data. The answer: on the adult side, an average of 49%. On the pediatric side, 23%.

Roughly half of the measurable clinical data a therapist gathers during an evaluation never becomes structured data. It lives in a free-text blob, or a scanned form, or a therapist's memory of "we'll clean it up later."

This is the actual shape of the problem. An initial evaluation isn't a paragraph, it's a form. In the evaluations we've templated, a single initial PT eval runs past 200 discrete fields: bilateral range-of-motion and muscle-testing grids joint by joint, Berg, FIM, plan-of-care selections, each one a small decision and a small keystroke. Writing a narrative summary of the visit is the easy part. Filling 200 structured fields, correctly, while a patient waits, is the part that doesn't get done.

Why the gap costs money, not just minutes

Unstructured documentation isn't only a burnout problem. It's a revenue problem.

In 2024, the Massachusetts Health Policy Commission found that nearly one in five commercial claims were denied, and that incomplete claims, coding errors, and duplicates alone accounted for 4.9% of all claims submitted. Missing or invalid data sits near the top of the avoidable-denial list in revenue-cycle analysis. When the objective measures that justify medical necessity live in a text blob instead of the fields a payer's system reads, the claim is easier to deny and harder to appeal. The documentation gap and the denied claim are the same gap, seen from two ends.

The mismatch in what "documentation tools" fix

Here's why this matters for anyone shopping for a fix. The current wave of clinical AI is built to write prose: ambient tools that listen to a visit and generate a narrative note. That's genuinely useful for a primary-care visit that is mostly narrative. But a therapy evaluation is mostly structure, and a paragraph isn't a filled Berg score. A tool that hands you a beautifully written summary has solved the easy fraction and left the structured majority exactly where it was: on your plate, after hours.

The unglamorous truth is that the highest-leverage place to remove therapy documentation burden isn't the narrative. It's the grids and scales, the discrete fields that a study just told us therapists only capture about half the time. Close that gap and you get the same thing from both ends: a clinician who goes home on time, and a claim that has the structured data to stand up.