Students reflect more critically when they stay in charge of AI
Updated Jun 13, 2026
student voicefeedbackAI guidance is easy to reduce to a misconduct rule or a usage rate. The harder question is whether students still experience themselves as the person doing the judging. That is why Aniekan Essien, Xue Zhou, Marios Kremantzis and Da Teng's Studies in Higher Education paper, "The agency gap: perceived human AI agency, reflection and generative AI learning across UK and China based higher education contexts", matters for universities using student voice to understand AI-supported learning. The paper suggests that when students report keeping more initiative, monitoring, and final decision-making power, they also report stronger reflection and critical thinking.
Context and research question
Much of the higher education conversation about Generative AI still falls into a blunt binary: cheating or assistance. That framing is too thin for UK universities trying to write sensible policy, review assessment design, or interpret AI-related comments in module evaluations and pulse surveys. Students do not only need rules. They need to know how to use AI without handing over too much of the intellectual work.
Essien and colleagues address that problem by focusing on perceived human-AI agency. In the study, that means students' reported initiative, their monitoring of AI outputs, and who they believe makes the final decision in AI-supported learning. The authors then test how that perception relates to reflective engagement and self-reported critical thinking across higher education contexts in the UK and China. That makes the paper especially useful alongside Advance HE's recent UK evidence on student experiences of GenAI, because it moves the discussion from simple adoption towards the quality of students' engagement with the tool.
Methodologically, the study used a comparative quantitative design and Partial Least Squares Structural Equation Modeling. The dataset covered university students in the UK (n = 145) and China (n = 164). That is a helpful design for Student Experience and Market Insights teams because it tests relationships between agency, reflection, and reported outcomes across two different higher education contexts rather than treating student AI use as a single, universal behaviour.
Key findings
The core finding is that keeping human decision-making power matters. Across both samples, perceived human-AI agency was positively associated with reflective engagement. In other words, students who said they stayed more actively involved in directing, checking, and deciding also reported reflecting more on what they were doing. That is a more useful finding than a simple "AI helps" or "AI harms" conclusion, because it suggests the educational issue is not just whether AI is present, but how much agency students retain when they use it.
The abstract captures that point clearly:
"students who report retaining decision making power when using AI also tend to report stronger reflective engagement"
Reflective engagement then mattered because it was associated with self-reported critical thinking. That is the second important step in the model. The paper does not present AI as automatically improving critical thought. Instead, it suggests that reflective use, rather than passive dependence, is what links AI-supported learning to stronger intellectual engagement. That fits neatly with other recent evidence that students use Generative AI for feedback, but trust teachers more when judgement becomes more consequential.
The study also found meaningful contextual differences between the two samples. For UK students, reflective engagement was associated with stronger academic self-concept. For Chinese students, that relationship was non-significant, while AI literacy was modelled as a moderator in the relationship between agency and reflection. The practical point is not that one national context is "better" than the other. It is that AI-supported learning is shaped by context, and universities should be careful about assuming one student response pattern will generalise neatly across cohorts, disciplines, or institutional settings.
That is why the idea of an "agency gap" is useful for student voice work. Two students may both report using AI, but one may be using it as a prompt, critic, or drafting aid while the other is letting it drive the process. A headline survey item on AI use will flatten those differences. Open-text responses, by contrast, can show whether students describe AI as something they steer, something they negotiate with, or something that is quietly taking over tasks they no longer feel confident doing themselves.
Practical implications
For UK higher education teams, the first implication is to stop asking only whether students use AI. Survey and module-evaluation questions should separate initiative, checking behaviour, final decision-making, and perceived learning value. A university that only measures uptake will miss whether students feel more capable, more dependent, or more uncertain after using the tool. The benefit is a more diagnostic evidence base for AI policy.
Second, institutions should treat reflection as a design goal, not just a by-product. If reflective engagement is the bridge between human-AI agency and self-reported critical thinking, then course teams should build in activities that ask students to justify prompts, evaluate outputs, compare alternatives, or explain what they changed and why. That is a more defensible educational response than simply banning or permitting tools at module level. The benefit is stronger critical practice rather than shallower compliance.
Third, universities should segment AI feedback more carefully and analyse the open comments properly. The abstract itself flags that cultural values, nationality, ethnicity, institutional policy, tool availability, and educational socialisation were not directly measured, yet all could shape how students experience agency. That means local evidence matters. If institutions collect AI-related comments, they need a governed way to compare themes such as confidence, dependence, trust, and judgement across faculties or cohorts. Our student comment analysis governance checklist is a practical starting point. The benefit is evidence that stands up better when policy decisions follow.
The broader lesson is that student voice on AI should focus on who is still doing the thinking. That is where Student Voice Analytics fits naturally. It helps universities categorise AI-related free-text comments consistently, so teams can distinguish useful assistance from emerging over-reliance and avoid building policy around anecdotes or one-off anxieties. The benefit is a clearer line from student comment to institutional action.
FAQ
Q: How should a university measure student agency around AI in its own feedback processes?
A: Start with more precise prompts. Ask students separately whether AI helps them generate ideas, test understanding, evaluate outputs, or make final decisions. Then add one open-text question such as "When using AI for study, where do you still feel in control, and where do you feel the tool is taking over?" That gives institutions evidence about agency rather than a blunt usage percentage.
Q: What are the main methodological cautions in this study?
A: The paper uses self-reported data and a comparative quantitative model, so it is strongest as evidence about relationships in reported experience rather than direct proof of learning gains. The abstract also notes that important contextual factors, including institutional policy and tool availability, were not directly measured. UK universities should therefore use the findings as a strong interpretive framework, then test the same issues in their own surveys and comments.
Q: What does this change about student voice in AI policy more broadly?
A: It shifts the question from "Are students using AI?" to "How are students experiencing authorship, judgement, and responsibility when they use AI?" That is a more useful frame for student voice because it connects AI feedback to academic confidence, assessment design, and the quality of support students receive, not only to compliance.
References
[Paper Source]: Aniekan Essien, Xue Zhou, Marios Kremantzis, Da Teng "The agency gap: perceived human AI agency, reflection and generative AI learning across UK and China based higher education contexts" DOI: 10.1080/03075079.2026.2686986
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