Generative AI tools like ChatGPT are now part of everyday learning—used to brainstorm, summarize, draft, translate, and even “explain” difficult concepts on demand. For schools, that raises a high-stakes question: are students using AI as a learning support, or as a substitute for thinking?
At TinyEYE, we work with schools that are balancing innovation with student well-being and skill development. Whether we’re talking about communication goals, executive functioning, or self-advocacy, students benefit most when adults can see what’s really happening beneath the surface: what the student believes, what the student relies on, and what the student avoids. The same is true with AI use. Students’ beliefs about AI shape how they engage with it—and whether they challenge it.
A 2025 study by Oliveira and colleagues, “Prompting Minds: Evaluating how Students Perceive Generative AI’s Critical Thinking Dispositions,” offers a timely tool for that exact issue. The researchers developed and validated a scale designed to measure how students perceive the “critical thinking dispositions” of generative AI. In other words: do students think the AI reasons well, seeks evidence, tells the truth, stays open-minded, and works systematically?
Why student perception matters more than we think
In education, we often focus on what students can do—skills like analyzing a text, evaluating a source, or building an argument. But critical thinking also depends on dispositions: the habits and attitudes that make someone choose to use those skills. A student might have the ability to verify information, but if they assume AI is always correct, they may not bother.
This is why the study’s focus is so important. Instead of measuring students’ own critical thinking dispositions, the new instrument measures students’ perceptions of the AI’s dispositions. That difference matters because:
- If students believe AI is highly capable, they may accept outputs too quickly and reduce independent effort.
- If students believe AI is unreliable, they may ignore a potentially helpful support tool—or use it in a shallow, unproductive way.
- If perceptions vary widely across students, teachers may see inconsistent outcomes even when the same AI tool is available to everyone.
From a special education lens, perception is not a “soft” variable—it’s a driver of behavior. Students who struggle with anxiety, perfectionism, attention, or task initiation may be especially vulnerable to overreliance if AI feels like a fast, confident answer machine.
The PCTD-GAI scale: a new way to measure perceived AI thinking
The researchers created the Perceived Critical Thinking Disposition of Generative Artificial Intelligence (PCTD-GAI) scale by adapting an existing critical thinking disposition measure (the Marmara Critical Thinking Dispositions Scale). They reframed items so students weren’t rating themselves; they were rating what they believe ChatGPT can do.
The final scale includes 27 items rated on a 5-point Likert scale (from Strongly Disagree to Strongly Agree) across six dimensions:
- Reasoning
- Access to justice (in this scale, focused on fair categorization, drawing conclusions, identifying risks, and asking appropriate questions)
- Search for evidence
- Search for truth
- Open-mindedness
- Systematicity
The study surveyed 931 university students in Portugal and Poland and used strong statistical methods (exploratory and confirmatory factor analysis) to validate that the scale reliably measures these perceptions.
What students believed ChatGPT does well—and where they were unsure
Across both countries, perceptions were moderately positive overall, with Portuguese students rating ChatGPT slightly higher across domains and showing a bit more agreement (less variability).
Highest-rated: Systematicity
Students tended to see ChatGPT as systematic—good at organizing steps, planning strategies, and drawing conclusions from provided information. This makes intuitive sense: AI often sounds structured, confident, and orderly, especially when it formats answers into lists and sequences.
In classrooms, that “systematic” feel can be a double-edged sword. Students may confuse neat structure with true understanding. A well-formatted answer can still be inaccurate, biased, or missing context.
Moderately positive: Reasoning, access to justice, open-mindedness
Students rated ChatGPT as reasonably capable of explaining problems, comparing information, drawing conclusions, and considering different perspectives. These are the very features that make AI appealing for school tasks: it can generate plausible explanations quickly and “sound” balanced.
For educators, this suggests a practical teaching opportunity: if students already see AI as a decent reasoner, we can explicitly teach them to treat AI as a draft thinking partner, not a final authority.
More neutral: Search for evidence
Students were less confident that ChatGPT reliably seeks strong evidence or draws from credible, diverse sources. This is a healthy skepticism, and it aligns with a key reality: unless students request citations and then verify them, AI may provide information that is incomplete, outdated, or unsupported.
Most uncertain: Search for truth
The most neutral (and in some cases slightly negative) ratings were in truth-seeking. That domain included items about being thoughtful when evaluating information and coping with problems realistically. Students were not fully convinced the AI “seeks truth” in a human sense—and that’s an important insight.
In plain terms: students may experience AI as organized and helpful, but not necessarily as a tool that reliably “cares” about truth. This is a strong foundation for AI literacy instruction, because it opens the door to teaching students how to verify, triangulate, and reflect.
What this means for schools: prevent passive reliance and build active AI literacy
The study’s discussion highlights a central risk: when students attribute high critical thinking dispositions to AI, they may invest less cognitive effort themselves. Over time, that can contribute to superficial learning habits—especially under workload pressure, time constraints, or low confidence.
Schools don’t need to respond with fear or blanket bans. They need a plan that teaches students to engage actively. The research points to metacognitive training as a powerful approach—teaching students to monitor their own thinking and regulate how they use AI.
Here are school-friendly ways to turn that idea into practice:
- Teach “verify before you trust” routines. Require students to identify which claims need checking and to confirm them using credible sources.
- Use AI as a contrast tool. Provide an AI-generated response and ask students to find missing context, weak reasoning, or unsupported claims.
- Make prompting a thinking skill. Have students write prompts that request evidence, counterarguments, assumptions, and limitations.
- Normalize uncertainty. Teach students that “sounds right” is not the same as “is right,” especially when answers are polished.
- Support executive functioning. For students who use AI to escape task initiation or reduce anxiety, pair AI use with structured planning steps and reflection (what I used, why I used it, what I checked, what I changed).
From a special education perspective, these supports are not just “AI rules.” They are skill-building routines that strengthen self-regulation, persistence, and independent reasoning—skills that generalize far beyond any single tool.
How a perception scale can help educators and school leaders
One of the most practical contributions of the PCTD-GAI scale is that it gives educators a way to measure something that is usually invisible: student trust in AI’s thinking.
That matters because interventions work best when they match the actual need:
- If a group rates AI very high on evidence-seeking, they may need explicit instruction in source checking and citation verification.
- If students rate AI high on reasoning and systematicity, they may need tasks that require them to generate their own argument first, then compare it to AI.
- If students rate AI low across the board, they may need guidance on productive, ethical uses—so the tool supports learning rather than becoming a hidden shortcut.
Even though this study was conducted with university students, the underlying issue—how learners judge AI’s “thinking”—is already present in K–12 settings. The earlier we teach students to evaluate AI thoughtfully, the more we protect critical thinking development as AI becomes routine.
For more information, please follow this link.
