AI can help students finish academic tasks more quickly. It can translate a difficult passage, reorganize a paragraph, suggest clearer wording, and produce a polished response within seconds.
But does this efficiency lead to better thinking?
This question is at the centre of a study by Yuanyan Hu, Xiao Lan Curdt-Christiansen, and Jufang Wang. The researchers followed 102 Chinese undergraduates during a 16-week English for Academic Purposes course. Students completed three AI-assisted writing tasks involving translation, research writing, literature review, and data commentary.
In one activity, students compared their own translations of research abstracts with AI-generated versions. In another, they used AI while revising a research introduction and reviewing the literature. They also interpreted a diagram, evaluated an existing data commentary, and rewrote it.
The students reported improvements in several areas of critical thinking. They became better at identifying information, comparing ideas, questioning assumptions, forming hypotheses, and evaluating arguments. They also paid more attention to relevance and logical coherence.
The gains were much smaller in depth, flexibility, precision, and the ability to monitor their own thinking.
The missing layer of AI and critical thinking
The researchers call this ability meta-critical thinking. It involves stepping back from a task and examining how you are reasoning. Are you approaching the problem carefully? Have you considered other explanations? Is the evidence strong enough? Does your conclusion actually follow from the information available?
This kind of reflection was almost entirely absent from the students’ interactions with AI.
Most of their prompts focused on practical work: translating text, improving language, organizing a response, and saving time. Very few students asked AI to help them examine their logic or reflect on how they had reached a conclusion.
This finding connects with concerns about metacognitive laziness when students use generative AI. When a tool can quickly complete part of a task, students may feel less need to plan, monitor, and evaluate their own learning.
Peer interaction produced a different pattern. Students turned to one another when they needed to unpack a difficult task, question a causal connection, defend an interpretation, evaluate an argument, or negotiate meaning. AI handled much of the operational work. The deeper reasoning often emerged through conversation with other people.
This supports the thinking behind combining AI review, peer review, and student reflection. Students can examine an AI response, but they still need opportunities to challenge ideas, explain their choices, and hear how someone else understands the same problem.
Task design makes the difference
The strongest evidence of critical thinking appeared in the most cognitively demanding activity. Students had to critique and reconstruct a research introduction. They needed to bring together different sources, assess coherence, identify gaps, and build an argument.
A more procedural data-commentary task made students efficient, but AI sometimes completed parts of the cognitive work for them.
Hu and colleagues describe their proposed approach as “AI-triggered, peer-constructed.” AI can produce an initial response, reveal a problem, introduce another perspective, or give students something to question. Peers then discuss, challenge, justify, and build understanding together.
For me, this is the most useful message from the study. Placing AI inside an assignment does not automatically create critical thinking. The task must give students something worth questioning, enough openness to make meaningful decisions, and opportunities to test their reasoning with others. Research on how critical thinking can be taught also reminds us that it requires deliberate instruction and practice.
The study relied partly on student self-reports and task observations, so it cannot establish objective gains in critical-thinking ability. Still, it leaves us with a practical question worth asking whenever we design an AI-assisted activity:
What kind of thinking does this task still require from the student?
Reference
Hu, Y., Curdt-Christiansen, X. L., & Wang, J. (2025). Smart tools, smarter minds? Learner-AI interaction and AI assistance on critical thinking in EAP contexts. Journal of English for Academic Purposes, 78, Article 101586. https://doi.org/10.1016/j.jeap.2025.101586
