Confidence, not competence: what an AI self-efficacy study at Koç actually measured
A small Koç University nursing study asked students how comfortable they felt using AI tools. The instrument measured something real, but it was never a test of whether students could actually use them.
On 18 July 2026 a study out of Koç University School of Nursing circulated through academic press channels. It did not test whether students could use artificial intelligence. It asked them how sure they were that they could.
That distinction is the entire story.
What the researchers actually did
The study was led by Associate Professor Remziye Semerci Şahin and conducted among students at the Koç University School of Nursing. Participants completed a self-report instrument, a structured questionnaire, capturing their perceived self-efficacy in using AI technologies in academic and clinical settings. Self-efficacy, in this technical sense, is a person's belief in their own capability to perform a task. It is not the task itself, and it is not a measure of that task's actual performance.
The instrument asked students to rate, on a numerical scale, statements such as whether they felt able to apply AI tools to coursework or to evaluate AI-generated health information. The team then analysed responses across cohorts. The point of the exercise, in plain terms, was a stocktake of how nursing students at one Turkish private university think about AI: how comfortable they claim to be, and where the gaps in that confidence sit.
Why the framing matters
Coverage of education-and-AI research tends to flatten two very different questions into one. "Can students use these tools?" produces a competence measurement. "Do students believe they can use these tools?" produces a confidence measurement. The two are correlated only loosely; people routinely over- and under-rate their own ability depending on prior exposure, gendered stereotypes about technical skill, and what their peers have admitted to using.
The Koç paper sits squarely in the second column, and reading it as if it answered the first question would be a category error. A student who rates themselves a 5 out of 5 on "I can critically appraise a chatbot's clinical recommendation" may, on testing, prove unable to spot a hallucinated citation. A student who rates themselves a 2 out of 5 may, in practice, be methodical and accurate. The study measured the first population's beliefs about themselves, not the second population's skill.
Where the structural interest lies
Ninety-degree turn from the methodological caveat, the value of the study is structural. Universities worldwide are rolling out AI policies, AI literacy modules, and AI use-dilemmas onto curricula that were drafted before the current generation of large language models existed. The empirical question everyone actually needs to answer is: as the tools land, what is the incoming workforce's posture toward them?
Self-efficacy scales are the standard answer to that question in education research, because they are cheap to administer, replicable across institutions, and reasonably predictive of whether a learner will attempt a task at all. A student who does not believe they can use a tool will not use it, even if the tool is accessible and the institution endorses it. Conversely, the student who believes they are a power user will deploy the tool in settings where they shouldn't. Both failure modes are real, and the policy response to each is different.
The Koç data, taken on its own terms, tells administrators in nursing and adjacent health professions where their students stand today: how confident, where confident, where ambivalent. That is genuinely useful for module design. It is not, and should not be read as, an audit of whether those students are correct in their confidence.
What remains uncertain
The study did not, in the available reporting, publish a headline confidence number. The single source circulating on 18 July was a brief Physical (PHYS) feed item; details on sample size, demographic breakdown, and the precise wording of high- and low-scoring items remain undisclosed in the material reviewed. Replication across institutions, and triangulation against an objective skill measure, would be the obvious next moves. Neither has been reported.
The honest reading is that this is a baseline measurement from one private Turkish university's nursing programme, on one date. It is informative at that scale and not much beyond it. The temptation to treat any AI-and-education study as a verdict on "what students can do" is the perennial trap. This one measured what students believe they can do, and that answer is its own kind of evidence, narrower, and more politically loaded, than the way it will likely be cited.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://en.wikipedia.org/wiki/Self-efficacy