No. Physical therapy is one of the strongest growth occupations in healthcare, and the reason it resists automation is that the core of the work is hands-on assessment and treatment of a body that responds differently every session. Federal projections put physical therapy at 12 percent growth from 2025 to 2035, much faster than average, taking the profession from 283,700 positions to roughly 317,500. Median pay was $102,760 in 2025, and entry requires a doctoral or professional degree.
Where AI is genuinely arriving is in documentation, exercise prescription and remote monitoring between sessions, none of which involves removing the therapist from the room.
Key points
- The assessment is manual. Palpation, range of motion testing, watching how someone compensates when they move. No sensor replicates a trained pair of hands.
- Growth is 12 percent to 2035, adding roughly 33,800 positions, far above the whole-economy rate near 3.5 percent.
- Treatment is physical. Manual therapy, guided movement, hands-on correction and progressive loading all require a therapist present.
- Adherence is the real clinical problem, and it is a relationship problem more than an information one.
- Entry is expensive and long. A doctoral qualification is the barrier, and it is also what protects the profession.
What physical therapists actually do
Describing the job as prescribing exercises understates almost everything that makes it work.
A therapist meeting a patient after a knee replacement, a stroke or a back injury has to establish what the person can currently do, what is limiting them, whether the limit is pain, weakness, stiffness, fear or a compensation pattern learned since the injury, and what a realistic trajectory looks like. Much of that comes from watching someone move and from feeling how a joint behaves under load. The patient’s own report is useful but incomplete, because people are unreliable narrators of their own bodies.
Treatment then combines manual work, guided movement, progressive loading and a great deal of coaching. The exercises themselves are frequently simple. Getting a frightened patient to load a joint they believe is fragile, or an impatient one to stop doing too much, is the difficult part. That part is entirely interpersonal.
This is why the profession is often misread from outside. A list of prescribed exercises looks like information that could be delivered by any channel. The exercises are not the intervention. The intervention is a person deciding, session by session, what this particular patient is ready for and then persuading them to do it.
Four properties that resist automation
Assessment is tactile and visual together. A therapist feels resistance, end-feel and muscle guarding while watching the whole body compensate. Motion capture measures the second and nothing of the first.
Every session differs. The same patient presents differently depending on sleep, pain, mood and what they did yesterday. Treatment is adjusted continuously within a session rather than followed from a plan.
Progression is judgement under uncertainty. Loading a healing tissue too fast causes setback; too slow wastes recovery time. That call depends on tissue type, surgical technique, patient history and how the last three sessions went.
Adherence depends on trust. Most rehabilitation outcomes are determined by what the patient does between appointments. That is driven by whether they believe the person telling them. This is the part technology has repeatedly failed to solve.
The last point deserves more weight than it usually gets. A rehabilitation programme is only as good as the proportion of it that actually happens. Studies of home exercise adherence consistently find that a large share of prescribed work is never done, and the strongest predictor of whether it is done is the relationship with the clinician rather than the quality of the instructions. That is an uncomfortable finding for anyone building a purely digital product, and it explains why so many have struggled commercially despite good content.
Where AI is genuinely being used
The technology is real and expanding, and every application supports the therapist.
Documentation tools transcribe and structure clinical notes, which matters because record keeping consumes a large share of a therapist’s day and is a leading contributor to burnout. Exercise prescription platforms generate programmes, deliver video instruction and track completion between visits. Computer vision applications check exercise form through a phone camera and flag when a movement is being done wrong at home. Wearables and sensors record load, step count and range of motion between appointments, which turns a patient’s vague answer into data. Outcome prediction models estimate likely recovery trajectories from similar cases, which helps set expectations.
The consistent pattern is that these tools extend the therapist’s reach between sessions rather than replacing the session. A programme delivered by an app with no clinician attached has consistently worse adherence than the same programme delivered by a person who follows up, and that finding has held across many attempts to build a purely digital product.
There is one genuine caution. Form-checking by camera is useful for gross errors and unreliable for subtle ones. A patient who trusts it may reinforce a compensation the system cannot see. Clinics that use these tools well treat them as a supplement the therapist reviews, not as unsupervised instruction.
There is also a workload effect that cuts against therapists rather than for them. If remote monitoring lets one clinician oversee more patients, employers will expect exactly that. The technology does not remove the job, but it can quietly expand the caseload attached to it, and that is the change most likely to affect working conditions over the next decade.
What to know before deciding
| Measure | Physical therapists, 2025 |
|---|---|
| Median annual pay | $102,760 |
| Number of jobs | 283,700 |
| Projected growth, 2025 to 2035 | 12 percent (Much faster than average) |
| Projected employment change | 33,800 |
| Typical entry-level education | Doctoral or professional degree |
Three things about that table matter for a career decision.
The growth is demographic rather than technological. An ageing population produces more joint replacements, more stroke rehabilitation and more chronic musculoskeletal conditions. None of that reverses in the next decade.
There is a second driver that gets less attention. Health systems under cost pressure increasingly prefer conservative treatment before surgery, because rehabilitation is cheaper than an operation and often works. Every referral diverted that way lands in a physical therapy clinic. That is a policy trend rather than a technological one, and it points the same direction as the demographics.
The pay is strong for healthcare outside medicine, but the entry cost is real. A doctoral qualification takes years and frequently leaves substantial debt. That calculation deserves running properly before committing rather than after, and it should use starting salaries in the setting you actually want to work in, not the national median across all settings and all levels of experience.
The comparison across healthcare is instructive. EMTs and paramedics are forecast at 6 percent growth with median pay of $48,150. Hands-on clinical roles are growing across the board; the wage differences track training length and licensure rather than exposure to automation.
For context on how exposure is measured, the Bureau publishes AI exposure categories for 831 occupations and states plainly that exposure “does not imply job loss, productivity gains, automation probability, or wage effects.”
What actually changes over the next decade
- Documentation stops consuming evenings. Automated note generation is the change practising therapists rate most highly.
- Between-session care becomes measurable. Wearables and app-tracked programmes turn adherence from a guess into data.
- Telerehabilitation holds a permanent share. Not a full substitute, but effective for education, progression checks and rural access.
- Caseloads get larger. If a therapist can monitor more patients remotely, employers will expect them to, which is a workload risk rather than a job risk.
- Prediction shapes expectations. Outcome models help set realistic timelines with patients, which improves adherence.
- Specialisation deepens. Neurological, paediatric, pelvic health and sports rehabilitation continue to diverge into distinct careers.
- Direct access widens. More jurisdictions let patients see a therapist without a physician referral, which increases both autonomy and diagnostic responsibility.
Decision framework
Five questions before committing to this profession.
- Can you afford the entry route? A doctoral programme is a significant investment of time and money. Run the numbers against realistic starting salaries in your region rather than national medians.
- Do you want the interpersonal side? A large share of this job is persuasion, reassurance and behaviour change. People who want a purely technical clinical role often find that surprising.
- Which setting suits you? Hospital, outpatient clinic, sports, home health and neurological rehabilitation are very different working lives with different pay and hours.
- Can you sustain the physical load? Manual therapy is demanding on hands, back and shoulders over decades, and planning for that early is sensible.
- Will you use the tooling properly? Therapists who understand what a form-checking system can and cannot see, and who use remote monitoring to target their in-person time, get more from a caseload than those who ignore it or over-trust it. The skill is knowing which patients need the room and which are fine with a check-in.
That last skill generalises well beyond clinical work. Knowing how these systems reason, and where their confident output needs checking, is transferable. If you want a structured route in, explore Coursiv AI lessons and check current plan details on the official site.
Your next step
If you are considering this profession, shadow therapists in two different settings before applying. Outpatient orthopaedics and inpatient neurological rehabilitation are so different that liking one tells you almost nothing about the other, and the qualification is too expensive to discover that afterwards.
One practical note on choosing where to start. Newly qualified therapists often take the first post available and discover two years later that the setting does not suit them. Moving between settings is possible but slower than it looks, because each has its own competencies and referral patterns. Spending time deciding deliberately at the outset is worth more than it feels like at the time.
If you already practise, the highest-return change is treating documentation automation seriously. It is the single largest recoverable block of time in the working week, and the therapists who reclaim it report it as the difference between a sustainable caseload and an unsustainable one.