Analysis based on the August 12 peer-reviewed worldwide survey led by Jana Cason and colleagues, AOTA's current AI policy and 2025 Code of Ethics, and HHS privacy and security guidance for health information.
Occupational therapy's artificial-intelligence debate now has a head count. In a worldwide survey published August 12, 56.3% of 884 respondents said they were already using AI at work. The tools were helping with documentation, administration, education, research, intervention planning, and communication. That list places AI squarely inside the ordinary workday, where a convenient draft can become part of a clinical record and a casual prompt can carry information farther than anyone in the room intended.
In brief
- The study analyzed 884 responses from 81 countries; 87.8% of respondents were occupational therapists, 68.4% viewed AI integration positively, and 56.3% reported using AI at work.
- The voluntary online survey describes its respondents rather than the entire profession, yet it shows that AI use is established enough for employers and educators to need operational rules now.
- AOTA's current policy keeps professional judgment with the practitioner, and its 2025 Code of Ethics requires acknowledgment of AI resources in documentation plus transparency and applicable consent when AI is used in recorded sessions.
The future arrived through the side door
The survey was led by Jana Cason, Liat Gafni-Lachter, Tammy Richmond, and Karen Jacobs and published in WORK: A Journal of Prevention, Assessment & Rehabilitation. A World Federation of Occupational Therapists panel reviewed the 28-item questionnaire, which was distributed online in English, French, German, and Spanish from September through November 2025.
Most of the reported uses have little of the robot clinic about them. Respondents described turning to AI for the tasks that already crowd a schedule: supporting documentation, organizing administrative work, teaching and research, developing intervention ideas, and shaping communication. Those are precisely the tasks where speed feels helpful and the line between assistance and authorship can blur.
Two-thirds of respondents, 68.4%, expressed a positive view of AI integration. The study's qualitative findings also carried three cautions into the room: equity, the conditions needed for responsible implementation, and ethical obligations. Interest and concern were traveling together.
Keep the denominator attached to 56.3%
The headline number deserves its full label: 56.3% of this survey's respondents reported workplace AI use. The sample reached 81 countries, 87.8% of respondents were occupational therapists, and about one-third had more than 20 years of experience. That breadth makes the findings useful as an early map of adoption across the profession.
It does not establish that 56.3% of all occupational therapy practitioners worldwide use AI. Participation was voluntary, recruitment ran through online communications and social media, and people already interested in the topic may have been more likely to respond. Four survey languages also leave much of the world's linguistic diversity outside the questionnaire.
Even with those limits, the governance questions are current. A profession does not need a perfect census before deciding how client information may enter a tool, who must review an output, or how an AI-assisted note should be identified.
The clinical note is where convenience meets accountability
Documentation was among the most common reported uses, and it carries the clearest professional instructions. AOTA's Policy E.19 supports ethical AI integration that improves outcomes and reduces administrative burden while prioritizing accuracy, safety, transparency, evidence, and sustainability. The policy leaves clinical judgment with the practitioner and directs practitioners to evaluate AI output against the evidence and AOTA policy.
The 2025 Occupational Therapy Code of Ethics goes further. Standard 3E says AI resources used in clinical, professional, or academic documentation must be acknowledged in the document. When AI tools are used during recorded sessions for transcription, documentation, or data analysis, the Code calls for advance transparency about the recording, intended AI use, data collection, storage, and dissemination, with informed consent obtained when applicable.
A generated paragraph still needs a clinician who can defend every sentence. The review should confirm the objective findings, the client's own priorities, the skilled reasoning, the service delivered, the response to intervention, and the plan. Fluent language is no substitute for a record that belongs to this person and this encounter.
Privacy begins before the first prompt
The fastest way to create an AI problem is to treat every chat box as an approved clinical system. HHS requires covered entities and business associates to protect electronic protected health information with administrative, physical, and technical safeguards. When a business associate handles protected information for a covered entity, a written business-associate contract or other compliant arrangement is required.
That means a vendor's general assurance about security cannot settle the question for a clinic. Before identifiable client information enters an AI service, the organization should know which product and account are approved, whether the vendor will receive or retain protected information, what agreement governs the relationship, where data are stored, whether prompts or outputs train a model, who can access them, and how records are deleted or audited.
Practitioners should route those questions through the employer's privacy, security, legal, or compliance process. A consumer account and an employer-approved health-care deployment may display similar screens while carrying very different contractual and data controls.
Intervention planning needs a higher bar
The survey also found AI use in intervention planning. This is where a plausible answer can travel furthest from the evidence. A system may supply a tidy activity list without knowing the person's precautions, cognition, environment, culture, equipment, health literacy, caregiver capacity, payer limits, or reasons for valuing an occupation.
Useful review begins with the occupational profile and the evidence already gathered. Ask which claim in the output depends on a fact the tool does not possess. Check citations at their original source. Test whether the suggestion fits the plan of care, the practitioner's competence, the service setting, and the client's informed choices. Remove ideas whose benefit, dosage, safety, or relevance cannot be supported.
Equity belongs in the same review. The survey's respondents raised it as a central theme, and occupational therapy has particular reason to notice whose routines, language, devices, homes, and support systems an AI response assumes. A polished recommendation can still be unusable in the life where it is meant to land.
A policy small enough to use
An organization does not need a fifty-page manual before it can draw the first safe boundary. Start with an inventory: which AI tools staff and students are using, for which tasks, through which accounts, and with what information. Include features embedded in electronic records, dictation systems, browsers, office software, scheduling tools, and learning platforms; many users will not describe every embedded feature as AI.
Next, publish a short permitted-use table. Name the approved tools, allowed data, prohibited data, tasks requiring human verification, documentation-disclosure language, and the person who owns the final work. Separate low-risk uses such as reorganizing public information from clinical uses that handle protected information or influence care.
Finally, create a way to report a wrong answer, privacy concern, biased output, near miss, or workflow failure without hiding it. Review those events, revise the boundary, and train to the failures people actually encounter. Governance becomes credible when the policy changes the next shift.
What OT professionals should do now
Practitioners can begin with one blunt question: Is the tool I used approved for the information I gave it? Then ask whether the output is accurate, evidence-based, individualized, and acknowledged as required. If a session is recorded or analyzed, confirm that the person was informed and that applicable consent and organizational requirements were met.
Managers should bring privacy, compliance, information security, educators, and frontline clinicians to the same table. Students and faculty need matching rules across assignments, simulated documentation, research, fieldwork, and clinical systems; otherwise the least restrictive screen becomes the de facto policy.
The verbs in this survey are present tense. Occupational therapy respondents are already using AI to write, plan, teach, research, and communicate. Professional organizations are building guidance, and AOTA has placed disclosure, judgment, safety, and privacy into the governing documents. Clinics and classrooms now need those obligations visible at the exact moment someone reaches for the tool.