AI Clinical Assistants: Supporting Therapists Without Replacing Human Expertise
AI assistants can draft, summarize and organize. They cannot examine, empathize or decide. A practical framework for where AI belongs in a therapist's day and where it does not.
Few topics in healthcare generate as much excitement and anxiety as AI. For therapists, the anxiety is understandable. If software can write a note, what else can it do? The answer, grounded in the evidence, is reassuring: AI is good at supporting clinical work and poor at replacing it.
What AI assistants do well
AI clinical assistants are best at tasks that are repetitive, text-heavy and structured:
- Drafting documentation from a visit, ready for review.
- Summarizing long histories, messages or prior notes.
- Organizing information into consistent formats.
- Surfacing patterns such as declining adherence or missed check-ins.
- Handling routine communication like reminders and appointment confirmations.
These are real burdens. Nearly half of U.S. physical therapists reported burnout in a recent national survey (Evidence in Motion), and administrative work is a recurring cause. When AI takes on the typing and sorting, therapists get time back for patients.
What the evidence shows
In a randomized trial at UCLA Health, ambient AI scribes were tested with 238 physicians across about 72,000 encounters. One tool reduced time in each note by about 41 seconds compared with a smaller change in the control group, and both tools produced modest improvements in burnout and cognitive workload (UCLA Health). A Stanford pilot found significant reductions in task load and burnout among 48 physicians over three months (JAMIA).
The same UCLA research also found that AI notes occasionally contained clinically significant inaccuracies, mostly omissions, and concluded that the technology "requires active physician oversight, not passive acceptance."
AI is a capable assistant and a poor clinician. Design every workflow around that fact.
What AI cannot do
Examine
Palpation, observing compensation patterns, testing strength and range: these require a trained clinician in the room or on a live video visit.
Judge
Deciding when to progress, regress or refer depends on clinical reasoning across many inputs, including the ones that never make it into the chart.
Connect
Patients stay in therapy because they trust their therapist. Encouragement, reassurance and accountability are human skills.
Take responsibility
Licensed clinicians are accountable for the care they provide. AI tools are not, and should never be positioned as if they were.
A framework for responsible use
Think of AI tasks on a spectrum from low risk to high risk:
- Administrative: reminders, scheduling and confirmations. Good candidates for automation.
- Drafting: notes and summaries. Good candidates for AI, with clinician review before signing.
- Flagging: adherence drops or missed visits. Useful prompts for human follow-up.
- Recommending: program changes or diagnoses. Use with caution, with clear clinician control and attention to fairness. Section 1557 requires covered providers to identify and mitigate discrimination risks from patient care decision support tools (HHS).
- Deciding: treatment decisions. Always human.
Questions to ask before adopting an AI assistant
- Is every clinical output reviewed by a licensed clinician before it enters the record?
- Can the clinician easily edit or reject the output?
- Is it clear to staff which content was AI-assisted?
- Is patient data protected under a business associate agreement?
- How does the vendor measure accuracy, and can you see the results?
How TheraPro uses AI
TheraPro Clinical applies AI where it supports therapists and keeps them in control. Smart Notes™ prepares a SOAP draft shortly after a telehealth visit, and the therapist always reviews, edits and signs it. Nothing is filed automatically. Analytics and follow-up queues surface patients who need attention, while the decision about what to do remains with the therapist.
The future is collaborative
The most realistic future for AI in therapy is not autonomous care. It is a therapist with a capable assistant: one that handles the paperwork, highlights what needs attention and gets out of the way. The expertise, the hands and the relationship stay human.
