TL;DR
Polling of 1,000 NHS staff reports that 90% now use AI in clinical work, and that 70% believe patients view it positively. The more revealing number is 65%: the share who say they worked out how to use these tools before their employer issued any guidance. The research was commissioned by Heidi, which sells AI clinical documentation software, and carried out by Censuswide between 17 and 25 June.
Take the provenance seriously before taking the headline. A supplier of clinical AI tools commissioning a poll that finds near-universal clinical AI use has an obvious interest in the answer. The fieldwork agency and dates are disclosed, which makes the work auditable, but the framing is the vendor’s.
Read past that and the governance finding is the one that should worry trusts. Two-thirds of respondents describe navigating AI use ahead of formal policy. Among those who avoid these tools entirely, the most cited reason — given by roughly one in five — is the absence of clear guidance on using them in clinical practice. Staff are not waiting for permission; they are waiting for rules that have not arrived.
The pressure driving this is unglamorous. Four in five report administrative load rising in recent years, averaging 1.3 hours of paperwork stacked on top of a full day of patient contact. Two-thirds find AI useful for patient follow-ups and 63% think it could extend how long they can sustain the job. Adoption is not enthusiasm for the technology so much as a response to workload.
Reservations run in the other direction too: 28% doubt accuracy, 30% fear leaning on the tools too heavily, and 35% — the largest group — worry about what happens to patient data. Asked what they want policymakers to grasp, the top answer was clear accountability when AI shapes a clinical decision, again named by 35%.
That accountability gap has a supply-side mirror. A study published this week in npj Digital Medicine found two-thirds of medical AI developers work for organisations that have adopted no regulatory framework at all. Clinicians want to know who carries the liability; a sizeable share of the people building the tools are not working to any published standard. The MHRA’s reform consultation, which reported strong support last week, is the mechanism meant to close that — and it has not closed it yet.
Looking Forward
The number to watch is not adoption, which has plainly happened, but whether trust-level guidance catches up with practice. Royal College of Radiologists census data last month suggested the NHS still underuses AI for admin — the gap between official rollout and unofficial use is where the risk sits.