GenAI can increase student activity while weakening critical learning agency
Updated Jul 31, 2026
GenAI can make students look more active without necessarily making them more agentic. At Student Voice AI, we are interested in that distinction because universities now collect large amounts of AI-related student voice through module evaluations, pilot surveys, and open comments, yet still struggle to tell the difference between reflective use and quiet dependency. That is why Liangliang Xia, Liyin Zhang, Kexin Shen and Yan Dong's Teaching in Higher Education paper, "The impact of generative artificial intelligence (GenAI) on university students' learning agency: A mixed-methods study", matters. It suggests that more AI-supported activity is not the same as stronger judgement.
Context and research question
Many universities have moved quickly from asking whether students use GenAI to asking how far that use should be encouraged, guided, or constrained. The problem is that broad questions about usefulness or frequency flatten important differences. A student who uses AI to test an idea, challenge a line of reasoning, or regulate their own progress is doing something very different from a student who leans on it to generate plausible content without much scrutiny.
This paper tackles that problem through a mixed-methods design grounded in a three-dimensional framework of learning agency. The authors assigned 78 participants to either a GenAI group or a control group, then compared quantitative and qualitative evidence rather than relying on one method alone. That design makes the study useful for UK higher education teams because it speaks directly to a familiar local problem: AI-related feedback often looks positive at first glance, but institutions still need to know whether students are thinking more carefully or simply producing more.
Key findings
The quantitative result is the encouraging part. According to the abstract, GenAI supported learning agency by strengthening students' self-regulated learning propensities. In other words, the tool appears to have helped students organise and manage their learning process more actively. That matters because many institutions hope GenAI can widen access to low-friction academic support, especially where students want immediate help outside staff hours.
The qualitative result is the warning. The paper reports that the GenAI group produced more arguments than the control group, but also showed weaker critical distance from the material it generated. That is a more useful finding than a simple pro-AI or anti-AI conclusion because it shows how visible academic activity can rise while evaluative judgement stays fragile.
The tension is captured neatly in the abstract:
"they exhibited excessive reliance and uncritical reproduction of AI-generated content."
That point aligns with a wider pattern in recent evidence that students use Generative AI for feedback but trust teachers more when academic stakes rise. Students may use AI readily, but they do not automatically treat that use as thoughtful learning. For universities, the practical issue is not only whether GenAI increases output. It is whether it helps students question, adapt, and take responsibility for that output.
The paper's broader contribution is methodological. By combining quantitative and qualitative evidence, it shows why single-source AI surveys can be misleading. A student can report feeling more productive or more supported while still becoming more dependent on AI-generated framing. That matters for module evaluation design because a headline result such as "AI helped me learn" can conceal very different underlying behaviours.
Practical implications
For UK universities, the first implication is to separate AI use from AI judgement in student feedback collection. A broad question such as "Was GenAI helpful?" is not precise enough. Teams should ask whether students used AI to start work, test an interpretation, challenge their own reasoning, improve structure, or simply generate material faster. That produces better institutional evidence and avoids treating all AI-supported learning as one behaviour.
Second, institutions should look for signs of over-reliance as well as signs of confidence. That means collecting open comments that ask what GenAI helped students do for themselves, and what it may have tempted them to stop doing. This is where student voice in assessment and feedback becomes practically useful: comments can reveal whether students are using AI to deepen reflection, bypass uncertainty, or replace the difficult parts of thinking. Student Voice Analytics fits naturally here because it helps universities categorise repeated themes such as initiative, dependence, trust, and authorship across large comment sets rather than relying on isolated anecdotes.
Third, universities should treat AI-related student feedback as a governance issue as well as a pedagogic one. If institutions want local AI pilots, module evaluations, or pulse surveys to shape policy, they need a clear method for reviewing and interpreting that evidence. A student comment analysis governance checklist is a sensible starting point because it helps teams define how sensitive AI-related comments are grouped, checked, and turned into decisions. The benefit is clearer accountability when universities revise assessment guidance or student support in response to what students say.
The deeper lesson is that GenAI should not be judged only by whether students appear busier or more efficient. Universities need evidence about whether students are becoming more agentic, more reflective, and more capable of challenging the tool rather than just collaborating with it passively.
FAQ
Q: How should a university ask students about GenAI and learning agency in a useful way?
A: Break the issue into smaller questions. Ask whether students used GenAI to clarify concepts, generate ideas, test arguments, refine wording, or manage workload. Then ask where they still felt responsible for judging quality and where they felt the tool started to take over. One open-text prompt about what felt helpful and what felt risky is often more useful than a headline question about satisfaction.
Q: What are the main methodological limits of this study?
A: The study used 78 participants in a controlled mixed-methods design, which makes it useful for showing mechanisms rather than proving how every student group will behave in every institutional setting. It also focuses on learning agency in one specific GenAI-supported context, so universities should treat it as a strong directional signal and test similar questions in their own modules, surveys, and feedback exercises.
Q: What does this change about student voice practice more broadly?
A: It shifts the focus from "Do students use AI?" to "What kind of learner does AI use seem to be producing?" That is a much better question for Student Experience teams, PVCs, and Market Insights professionals. Student voice becomes more useful when it distinguishes confidence from dependency, and productivity from judgement.
References
[Paper Source]: Liangliang Xia, Liyin Zhang, Kexin Shen and Yan Dong "The impact of generative artificial intelligence (GenAI) on university students' learning agency: A mixed-methods study" DOI: 10.1080/13562517.2026.2649818
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