Nursing students tell researchers they trust themselves with AI, more than the tools deserve

A small questionnaire study at Koç University School of Nursing in Istanbul has reached a conclusion that will sound familiar to anyone who has watched a workplace adopt a new piece of software: the people who say they are most comfortable using it are not always the people who use it best. The study, led by Associate Professor Remziye Semerci Şahin and colleagues, asked nursing students how confident they felt applying artificial intelligence tools in clinical settings, and the answers were uniformly high.

The result is less reassuring than it appears. Self-efficacy, the psychological term for what the researchers actually measured, is not competence. It is the belief that one can perform a task, untested against any objective standard. A field that recruits on confidence rather than demonstrated skill is a field that will, at some point, ship the wrong dose, misread a model, or accept an algorithmic suggestion it should have overruled.

What the students said

The instrument used in the study was a self-report scale: students rated their own perceived ability to work with AI technologies in nursing practice. Per the published abstract, the work was framed as an examination of "university students' perceived self-efficacy in using artificial intelligence technologies." The phrasing matters. The paper asks how capable students feel, not whether they are capable.

This is a recurring shape in adoption surveys across industries. New tools enter curricula; students pick them up; students rate themselves as fluent; the institutions that employ them later discover a gap between reported fluency and chart-side performance. The Koç study sits inside that pattern, with the additional weight that nursing is a clinical profession where the cost of a confident mistake is measured in patient harm.

The counter-narrative

There is a defensible read of this finding that runs in the opposite direction: students who rate themselves highly may simply be early adopters, and the cohorts that self-report lower confidence may be the ones who have not yet encountered the tools. In that framing, the survey is a snapshot of exposure rather than a verdict on competence, and the high scores are a leading indicator of where the curriculum has already landed.

Both readings can be true at once, and that is the problem. A scale that does not distinguish between "I have used an AI tool three times" and "I can defend a clinical decision against an AI suggestion" will return high numbers from anyone who has opened the app. Without an instrument that pairs self-report with a task test, the study tells readers something useful about student psychology and almost nothing about clinical readiness.

The structural frame

Healthcare systems across Europe and the Middle East are pressing AI tools into frontline use faster than the workforce is being trained to interrogate them. The pattern is not unique to nursing. Radiology has been living with this tension for a decade, where junior clinicians can be tempted to defer to a model's output because the model's confidence is itself a confidence-display, a number dressed up as certainty. When a tool presents its answer in the visual language of authority, the user's self-efficacy and the model's claimed accuracy blend together. A student who has internalised that the tool "is reliable" inherits the model's confidence as their own.

Universities are responding, but unevenly. AI literacy modules tend to teach prompting and tool selection, the equivalent of teaching someone to drive by explaining the dashboard. What is missing in most curricula, including, on this evidence, at Koç, is the harder skill: knowing when the tool is wrong. The Koç study's high self-efficacy scores may simply be a measurement of the curriculum's success in delivering the first layer, without delivering the second.

What is at stake

The patients are the obvious answer, and the right one. Nursing is the largest clinical workforce in most health systems, and nursing students entering practice now will spend the bulk of their careers working alongside algorithmic decision-support. If their self-reported fluency outpaces their ability to challenge those systems, the errors will not be catastrophic ones. They will be slow ones: a risk score accepted without cross-check, a triage suggestion followed without question, a discharge summary generated and not read closely.

The less obvious stake is institutional. Hospitals that hire on the assumption that a nursing graduate is "AI-ready," because every recent survey tells them incoming cohorts are confident with the tools, will design workflows that depend on that readiness. When the assumption proves soft, the cost lands on staffing and supervision, not on the survey that produced the false comfort.

What remains uncertain

The published material on the Koç study does not, in what is currently available, specify the sample size, the year of data collection, the response rate, or which specific AI tools the students were asked about. It is also unclear whether the instrument distinguished between generic AI tools (chatbots, search assistants) and clinically validated decision-support systems. Without those details, the headline finding, students feel capable, cannot be weighted against any external benchmark. A second, larger study that pairs a self-efficacy scale with a clinical task test would settle the question this one opens.

Desk note: Monexus framed this study as a measurement problem rather than a competence verdict, the news is not that nursing students are confident with AI, but that the instrument used to ask them rewards confidence without checking it. Wire coverage tends to report the headline number; the more durable story is the gap between feeling capable and being capable.