US–UK survey finds clinical AI use extends beyond employer-provided tools

Newly published research maps how 524 healthcare professionals use AI for documentation, literature work and clinical support.

A survey published in JMIR Formative Research on 19 October offers a detailed view of AI use among healthcare professionals in the United States and United Kingdom. Of 524 respondents, 75.8% reported at least one workplace AI use in the preceding 30 days.

Documentation, literature search and summarisation, and clinical decision support were the most common categories. The study also examined how people accessed tools, finding that personal, non-employer-provided systems were an important route.

The paper describes differences by professional role as well as country. These are findings from the surveyed sample, not a census of every clinician or an evaluation of patient outcomes.

Different tasks create different adoption paths

The results are useful because they separate several kinds of work that are often grouped under clinical AI. Drafting documentation, finding a paper and interpreting a clinical question involve different inputs and different next steps.

A documentation tool has to fit the record-writing process. A literature assistant has to help a professional find and understand relevant evidence. A decision-support tool has to connect its output to a clinical assessment. The same conversational interface can conceal those distinct operating requirements.

For a health organisation, a task-level view can make implementation more practical. Teams can identify where staff already use AI, understand the work they are trying to complete and decide which supported tools and training would help.

Access is part of the workflow

The distinction between personal and employer-provided tools also changes how adoption should be understood. A member of staff who finds an application useful may begin using it before a formal organisational programme exists.

That creates an opportunity for service design. Approved access needs to be easy to locate and appropriate to the job, while staff need clear ways to ask questions about use. A tool that is available in theory but awkward in the working day may not become the preferred option.

The practical conversation can begin with the actual task: what information is being supplied, what output is needed and where that output goes next. This makes it easier to choose a suitable workflow than starting with a broad list of AI features.

Training can follow professional roles

The survey’s role differences suggest that one course may not serve every group equally. A nurse preparing documentation, a pharmacist reviewing information and a physician researching a clinical question may need different exercises.

Training can use examples from the relevant work and show how to complete a task with the institution’s chosen tools. It can also explain how to retain the underlying information and record a useful handoff to the next professional.

That approach makes adoption something an organisation can improve through ordinary operational work: accessible systems, clear procedures, targeted instruction and feedback from staff.

The newly published study supplies a more granular starting point for those decisions. It shows that workplace use already spans several tasks and access routes. The next step for a health service is to understand which of those patterns exists locally and make the useful workflows easier for its own teams to complete.

Sources

- [Artificial Intelligence Use Among Health Care Professionals in the United States and United Kingdom: Cross-National Survey](https://formative.jmir.org/2026/1/e100681) — JMIR Formative Research · 2026-10-19