AI attitude surveys need to separate usefulness, self-expression, and concern
Updated Jun 25, 2026
student voicefeedbackUniversities are asking students whether they use Generative AI. That question is now too blunt to be very useful. At Student Voice AI, we see the same problem in AI-related survey comments and module feedback: one positive answer can mean practical help with learning, another can signal identity or status, and a negative one can reflect unease rather than outright rejection. That is why Angyang Li and Shuo Wang's Higher Education paper, "Functional attitudes towards artificial intelligence among university students: Development of a scale and influencing factors", matters. For UK institutions building on recent evidence about student experiences of GenAI across universities, it offers a more precise way to understand what students are actually telling you.
Context and research question
Many higher education surveys still treat student attitudes to AI as if they sit on a single line running from supportive to worried. That approach misses an important reality. A student may value AI because it helps them understand a concept more quickly, another may like what AI use says about being modern or capable, and a third may keep using it while still feeling defensive or wary about what it is doing to academic standards.
Li and Wang tackle that problem through Functional Attitude Theory, which asks what job an attitude is doing for the person who holds it. The paper develops a dedicated higher education scale, then tests it in two stages with Chinese university students. Study 1 used exploratory factor analysis with 366 students to refine the item pool. Study 2 used confirmatory factor analysis with 623 students to validate the structure, test reliability, and check measurement invariance across majors and genders. For UK higher education teams, that makes this less a generic AI-opinion piece and more a practical survey-design paper.
Key findings
The central finding is that student attitudes to AI split into three distinct dimensions rather than one overall sentiment score.
"The scale identifies three key dimensions: utility-knowledge, value-expression, and ego-defense."
That matters because each dimension points to a different institutional question. Utility-knowledge is about whether students see AI as genuinely helpful for learning and knowledge work. Value-expression suggests that AI attitudes may also reflect how students want to present themselves as learners. Ego-defense captures a more protective or wary response, where students may feel concerned about threat, dependence, or loss of control. For UK teams, the practical lesson is simple: a single "attitude to AI" score can flatten very different signals.
Usefulness and concern did not move neatly together. In the paper's statistical model, more conventional technology-acceptance measures aligned with utility-knowledge and value-expression, but not with ego-defense. A note in the article also reports that, in a follow-up sample of 266 students, utility-knowledge and value-expression predicted AI usage frequency, while ego-defense did not. In practice, that means a student can use AI regularly and still feel uneasy about it. If your surveys only measure uptake, they risk mistaking reluctant use for confident endorsement.
The subgroup patterns are also useful, though they should be treated cautiously outside the original context. The abstract reports that male students scored higher on value-expression, female students scored higher on ego-defense, and greater AI experience was linked to stronger utility-knowledge. STEM students showed lower ego-defense and value-expression, while rural students scored higher on those two dimensions than urban students. UK institutions should not import those patterns as fixed expectations, but they are a useful reminder that AI sentiment may vary by subject mix, prior experience, and student background rather than by policy alone.
The broader contribution is methodological. This paper gives universities a clearer structure for asking better questions. Much as earlier evidence showed that students use Generative AI for feedback but trust teachers more, this study suggests that AI student voice becomes more useful when it separates practical benefit from self-presentation and defensive concern. That is a stronger basis for policy than a headline figure about use or satisfaction.
Practical implications
For UK higher education teams, the first implication is to stop asking one generic AI attitude question. Instead, separate questions about practical usefulness, confidence in using the tool well, and discomfort or concern about what AI might be changing. If institutions are redesigning module evaluations, induction surveys, or AI pilot questionnaires, that shift alone will produce clearer evidence for action.
Second, universities should pair sharper closed questions with open-text prompts. A scale can tell you that a cohort scores highly on concern, but only comments will show whether that concern is about authorship, fairness, over-reliance, disciplinary norms, or unclear rules. This is where a shared vocabulary helps. Our student feedback analysis glossary is a practical starting point for naming recurring themes consistently, which makes cross-team interpretation more reliable.
Third, institutions should treat AI concern as a support signal, not only a compliance signal. If students are using AI while still feeling defensive or uncertain, the issue may sit in assessment guidance, academic skills provision, or local confidence about what counts as legitimate help. That is also why governance matters. If universities are going to analyse AI-related comments at scale, they need a traceable method for separating repeated themes from noise, especially when comparing local coding with generic LLM workflows. The benefit is stronger evidence before policy and teaching changes are rolled out more widely.
The wider lesson is that AI-related student voice is becoming a measurement problem as much as a policy problem. Better questions make it easier to distinguish curiosity from confidence, and use from trust. That gives UK universities a more defensible basis for deciding what to teach, what to regulate, and what to ask students next.
FAQ
Q: How should a university redesign its AI survey after reading this paper?
A: Start with three blocks of questions: what students find useful about AI, what using AI says about them as learners, and where they feel worried, defensive, or uncertain. Then add one open-text question asking what feels helpful, unclear, or uncomfortable in practice. That gives Student Experience and Quality teams better evidence than a single approval question.
Q: What are the methodological limits of the study?
A: This is a scale-development paper based on Chinese university students, so it is strongest on construct design rather than on cross-sector benchmarking. The two-stage validation is a real strength, but UK institutions should still test how well the framework travels into their own disciplinary mix, local policy context, and local student feedback language.
Q: What does this change about student voice on AI more broadly?
A: It suggests that AI-related student voice should not be read as simply pro-AI or anti-AI. The same cohort can use AI frequently, value it instrumentally, and still feel uneasy about what it means for standards or authorship. Universities that analyse those distinctions explicitly will be better placed to write clearer guidance and act on AI-related feedback with more confidence.
References
[Paper Source]: Angyang Li, Shuo Wang "Functional attitudes towards artificial intelligence among university students: Development of a scale and influencing factors" DOI: 10.1007/s10734-025-01497-x
Request a walkthrough
Book a free Student Voice Analytics demo
See all-comment coverage, sector benchmarks, and reporting designed for OfS quality and NSS requirements.
- All-comment coverage with HE-tuned taxonomy and sentiment.
- Versioned outputs with TEF-ready reporting.
- Benchmarks and BI-ready exports for boards and Senate.
UK-hosted · No public LLM APIs · Same-day turnaround