Incredible Health’s newly published State of Nursing report finds clinicians adopting AI well ahead of the strategies, training, and vetting meant to govern it. For health IT, that is both a genuine exposure and an unusual opening.
For two years, health IT leaders have been building plans for artificial intelligence. Their clinical staff, according to new survey data, decided not to wait for them. For its seventh annual State of Nursing report, Incredible Health asked 2,240 nurses and technicians how they are actually putting AI to use, and the answers should force informatics and IT teams to rethink their timelines. Compared against the employer-side findings in the company’s 2026 executive report, the dynamic is one every technology leader has seen before: adoption is outpacing governance, and now there are numbers to prove it.
The adoption itself is already well underway. Today 44% of nurses use AI in the course of their work, yet only 8% say any clear plan for applying it has ever made its way down to them. The picture grows murkier from there: just 14% call the strategy clear even where AI tools already run, 17% have no idea whether their organization has an AI strategy at all, and for one in five nurses the tools simply appear one day, with no plan and no explanation attached.
This is shadow IT rebuilt for the age of large models. When sanctioned tools and clear direction are missing, clinicians turn to whatever gets the job done, which in practice means consumer-grade assistants running well outside any institutional review. The consequences for output quality are immediate. Eighty-three percent of nurses say AI results are only sometimes or rarely reliable enough to act on without checking them first, which shifts the entire verification burden onto individual clinicians rather than a governed process. Half of the nurses who use AI say their most recent attempt saved them little or no time, a reminder that adoption on its own has never been the same thing as productivity.
The report is refreshingly specific about what would change that picture, and most of it sits squarely with IT and informatics. Training is the most obvious piece. Nurses who received training were half again as likely to gain real time from AI, with 24% saving more than an hour a day against just 16% of those left to fend for themselves. Frequency deepens the effect: 68% of nurses who use AI regularly report saving more than 15 minutes a day, a benefit that reached only 46% of occasional users. Yet 46% of nurses got no training whatsoever in the past year, and the formal instruction that genuinely prepared people reached a mere 5% of them. The tools are already inside the building; the knowledge to use them well has not followed.
Involving the people who will use the tools is the other piece leaders tend to leave idle, and it is entirely within IT’s gift. Nurses who had a say in choosing AI tools were far likelier to actually use them, 81% against 62% of those never consulted, and far likelier to trust them, 74% against 38%. That involvement also blunts the resistance governance teams so often expect. Among nurses who use AI regularly, 67% believe it will help the workforce, compared with 35% of those who avoid it, and the skepticism inverts in the same way, with only 16% of regular users fearing harm against 38% of non-users. In this data, familiarity is what converts apprehension into genuine buy-in.
One corner of the workflow stands out for how little AI has touched it. Among nurses who use AI, only 4% have put it to use during the hiring, screening, or interview process, even as job candidates increasingly use the same tools to apply. For any organization deciding where its AI governance should reach next, recruitment is a conspicuous blank space.
The strategic conclusion for health IT writes itself, and it is not the comfortable one. Clinicians are going to use AI, since a large and growing share already do, so the question that matters is whether the institution will give that use a shape. Governance built purely as a set of prohibitions gets the moment badly wrong, because what the data actually shows is a staff that has already opted in and is asking, through its own behavior, for clearer direction and the training to make it work. The institutions that give this behavior a structure will turn ungoverned sprawl into genuine advantage; those still treating AI as next quarter’s problem will keep learning about their own AI footprint secondhand, from the clinicians who built it without them.
