Why hands and judgment carry this job
Will occupational therapy be replaced by AI? The heart of the job happens between two people in a room, and software cannot stand in for that. An OT watches someone try to button a shirt, stand from a toilet seat, or hold a pen after a stroke. The useful information is in the wobble, the grimace, the shortcut the person invents without noticing. That observation then turns into a plan the person will actually follow at home.
Two tasks show the gap clearly. The first is the evaluation itself: positioning a limb, feeling tone and resistance, and adjusting the test when the patient tires. The second is fitting and training on adaptive equipment, from a wheelchair cushion to a reacher in a cramped kitchen. Both need hands, a body in the space, and a read on what this person will tolerate today.
Paperwork is the other half of the week, and that is where software has moved fastest. Progress notes, insurance justification, goal wording, and home-program handouts are all text, and text is what language models handle. That is task erosion, not a vanishing job. The same split shows up across healthcare diagnosing and treating roles.
What software does, what it assists, what stays with the therapist
The tasks our data puts in the AI-does group are documentation-shaped: drafting a progress note from entered findings, and turning a treatment plan into a printed home-exercise sheet. Those sit at 0% of task time here. Nothing in that group involves touching a patient.
The assisted group is larger in practice. Scoring standardized assessments and flagging a result that falls outside the expected range is one example; sorting a caseload so the most at-risk patients get seen first is another. Those tasks add up to 37% of task time. The therapist still decides what the number means for this patient.
Everything else belongs to the person: hands-on functional testing, caregiver and family training, and home or worksite visits where the fix depends on the actual doorway. That group covers 63% of task time. Our coverage measure, which asks how much task time AI can handle today, reads 22 out of 100; the coverage method explains how that is built.
Good to know: the documentation load is the part most likely to change first, because it is already text in an electronic record.
How strong is the evidence
Thin, and we say so. Our quality-parity grade for occupational therapists is D, which means no direct head-to-head test of an AI system against practicing occupational therapists has been recorded in our evidence set. Because of that, we publish no parity number for this job. A grade with no test behind it would be a guess dressed up as data.
What would settle it: a prospective study comparing AI-generated treatment plans with therapist plans on real patient outcomes, a blinded trial of automated functional assessment against in-person OT evaluation, and an audit of AI-drafted documentation checked by clinicians for accuracy and payer acceptance. Until work like that exists, the honest reading is that AI handles parts of the paperwork and gives no measured evidence on the clinical core. The quality-parity method sets out what each grade requires.
When the picture could shift
Most likely after 2042 (8 in 10 of our scenarios). For what that window measures and how it is built, see the replacement-year method.
Two things could pull it earlier. Electronic record vendors are shipping note-drafting into the tools therapists already open each shift, so adoption needs no new purchase. And running a language model on documentation is cheap next to a clinical salary, which the cost panel on this page shows.
Two things hold it back. The physical share of this job needs a machine that can support a body, adjust grip pressure, and work in a cluttered home, and the robot tier our data assigns here is a dexterous humanoid, which is not a bought-and-installed product. State licensure and payer rules also require a credentialed therapist to perform and sign off on evaluations, so a tool that drafts is not a tool that practices.
The job market context is not shrinking either. The Bureau of Labor Statistics counts about 162,450 occupational therapists in the United States, with employment projected to grow 14.8% between 2025 and 2035 and median pay of $100,330 (BLS, 2025). Aging demographics and rehab demand drive that, and software does not change the number of people recovering from strokes and falls.
How to stay needed as an OT
Lean into the three things on this page that stay with people. Complex functional evaluation, especially with patients whose presentation does not match the textbook. Caregiver and family training, where success depends on persuading a tired spouse to change a routine. And environmental assessment in homes, schools, and workplaces, where the recommendation has to survive a real budget and a real landlord.
Two skills raise your floor. First, learn to supervise AI documentation: check a drafted note against what you observed, and catch wording that would misstate medical necessity. Second, get fluent in outcome measurement, so you can show payers and employers what your treatment changed. Both make you the person who signs, not the person who types.
It also helps to see where nearby roles sit. Compare this job with physical therapists, with speech-language pathologists, or with occupational therapy assistants, whose task mix leans harder on hands-on delivery. You can put any two side by side on the job comparison tool, see how the wider healthcare sector scores, or browse the jobs that most need a person. Our full scoring approach is set out in the methodology.