Why this job stays with people
Recreational therapy runs on presence. A therapist watches how a stroke patient reaches for a paintbrush, notices the frustration before it turns into a refusal, and changes the activity on the spot. That loop of reading a person and adjusting the plan in the same minute is the core of the work, and it is the part software does not carry. It is also why the question of whether AI will replace recreational therapists looks different from the same question asked about desk work.
Two tasks show the gap clearly. Assessing a patient’s physical, social and emotional function means sitting with them, with family, and often with a nurse who has seen the last three bad nights. Leading a session means moving chairs, steadying a walker, pacing a group so the quietest person still gets a turn. Our coverage score, which estimates the share of task time AI can handle today, is 17 out of 100 for this occupation.
The physical share matters too. About 40% of the task load here involves bodies in rooms, and the hardware class that could plausibly do it is a dexterous humanoid, which is not in nursing homes or rehab units in any serious number. Meanwhile the field is small and steady: 14,930 people hold the job in the US at a median wage of $61,960, with employment projected to grow 4.8% from 2025 to 2035 (BLS, 2025).
What AI does, what it assists, and what needs a therapist
The writing and planning layer is where AI already contributes. Drafting progress notes from a therapist’s own observations, turning an assessment into a readable treatment-plan document, and generating activity ideas for a given diagnosis are all jobs a language model can take a first pass at. By our split, 0% of task time falls in the group where AI can handle the work with light review.
A second slice is assistance rather than handover. Tracking how a patient’s participation changes week to week, flagging goals that have gone stale, and prepping materials or schedules for group sessions go faster with software, but a therapist still decides what the pattern means. That assisted group covers 23% of task time.
The rest stays with the person. Running the session itself, adapting an activity mid-task for someone with limited grip or short attention, and talking a reluctant patient into trying again are not document problems. Teaching families how to keep an activity going at home sits here as well. 77% of task time needs a human, and that is the number driving the headline answer above.
What the evidence shows so far
There is no head-to-head test of AI against a qualified recreational therapist on this job’s own tasks. Published work on generative AI in therapeutic recreation discusses how tools might fit into assessment, planning, implementation and evaluation, but discussion is not measurement. Our evidence grade for this occupation is D, and where nothing has been measured we publish no quality-parity number at all rather than guess one. You can read how that grading works on the quality-parity method page.
What would settle it is narrow and doable. First, a blind comparison of AI-generated treatment plans and therapist-written plans, scored by experienced clinicians who do not know the source. Second, a trial that follows patient outcomes — mobility, mood, social participation, discharge readiness — when the plan comes from a model with therapist review versus a therapist working alone. Until something like that exists, claims in either direction are opinion.
When the picture could change
Most likely after 2042 (8 in 10 of our scenarios). The reasoning behind that span is set out on the replacement-year method page.
Two things could pull it earlier. Documentation tools are cheap next to a staffing line, so facilities under budget pressure adopt them quickly, and once notes and plans are drafted by software the role tilts further toward delivery. Electronic health records that feed structured outcome data back into a model would also widen what automated planning can see.
Two things hold it back. The hands-on share of the work needs dexterous robotics that are not deployed in long-term care, and the cost of that hardware is nowhere near a therapist’s wage. Licensure and certification are the second brake: state rules and payer requirements name a credentialed person as responsible for assessment and plan of care, so a tool can draft but not sign.
How to stay needed
Lean into the parts that do not transfer. Keep ownership of the live session, including the judgment calls when a patient shuts down or a group dynamic sours. Keep direct family and caregiver teaching, which depends on trust built in a room. Keep the interdisciplinary work — arguing a patient’s case in team rounds, where your read on function carries weight that a generated note does not.
Two skills pay off. Learn to use documentation and planning tools well enough to review their output fast and catch what they get wrong about a specific patient. And get sharper at outcome measurement: therapists who can show, with numbers, that a program improved function or reduced falls are the ones whose programs survive budget reviews.
What to do: Pick one standardized outcome measure you already collect and start reporting it by program, not just by patient.
If you are weighing adjacent paths, the closest work sits with music therapists and art therapists, who share the activity-based model, and with occupational therapists, who go deeper into function and daily living. You can put any two of them side by side on the compare tool, see the wider group on the diagnosing and treating practitioners family page, or look at how the whole healthcare sector scores. Jobs built on similar in-person demands show up on our list of jobs that mostly need a person, and the full scoring method explains every figure on this page.