‘Critical Thinking’ is Essential Learning, Now AI Just Raised the Stakes

Our students deserve lessons that teach them to question what sounds right, not just what looks wrong.

It is that time of year when reading lists start appearing everywhere: on school websites, in library newsletters, tucked into the final pages of the end-of-year agenda. Teachers share them in faculty meetings and parents pin them to refrigerators, and in May 2025, the Chicago Sun-Times and the Philadelphia Inquirer published one in a special print section. Of the fifteen titles on that list, only five were real books. The other ten were fabrications, complete with convincing descriptions and real authors’ names attached to titles those authors never wrote. The list had been generated by AI and published without adequate verification, and readers who paid for the printed paper had no way of knowing it from looking.

What makes this story worth bringing into a post about critical thinking is not the embarrassment to the newspapers, or even the mechanics of how AI hallucinations work. It is the more uncomfortable implication sitting just underneath: if professional editors at major publications missed this, we need to have an honest conversation about what we are expecting our students to catch on their own, and whether our lessons are actually preparing them to do it.

The problem wasn’t that the AI made something up. The problem was that it made something up and nobody checked.

Critical thinking has been a priority in education for as long as most of us have been in the profession. We have rubrics for it, units built around it, and standards that name it explicitly. And yet the reading list still made it to print. That gap between the skill we say we teach and the habit that failed in a real-world moment is exactly what this post is about, because in a world where AI-generated content is polished, confident, and increasingly indistinguishable from the real thing, the version of critical thinking we have been teaching may need a meaningful update.

AI hallucinations, the tendency of language models to generate fabricated information delivered in confident prose, are a persistent feature of these systems rather than a temporary flaw (Ji et al., 2023). What makes this particularly challenging for educators is a shift researchers have begun to document: as models improve, they hallucinate less often but more convincingly. A wrong answer from an earlier chatbot was often obviously off. A wrong answer from a current model may be polished enough to fool a domain expert who does not pause to verify. The reading list was not sloppy-looking content; it looked exactly like a thoughtful, well-curated list, because the AI had learned what those look like.

That surface-level credibility is landing directly in classrooms. Recent studies have found that students tend to rely on intuition and appearance rather than systematic verification when judging AI output, in part because most students hold incomplete mental models of how language models actually work (Doe & Lee, 2026; Valeri et al., 2024). A well-formatted response with a confident, authoritative tone signals trustworthiness, even when the underlying content is fabricated. The Chicago Sun-Times situation happened to professionals with editorial training. The same dynamic plays out every time a student opens a chatbot to research a topic, draft a paragraph, or look up a fact, and accepts what comes back without stopping to question it.

The institutional backdrop makes this even more pressing. A May 2026 Education Week report found that 82% of teachers have not received formal guidance on how to use AI tools responsibly in their work (Langreo, 2026). If most educators are still navigating AI largely on their own, explicitly building AI verification skills into lesson design is not just good practice; it is something students may not be getting anywhere else.

Examining an AI-embedded Lesson with Critical Thinking

The TRACE framework’s Critical Thinking construct asks a direct question to address exactly this: Does the lesson teach learners to question, verify, and audit AI responses? A lesson satisfies this construct when it encourages students to evaluate AI output for errors or gaps, and when it provides a structured opportunity to revise or refine what the AI produced. The second part matters as much as the first, because evaluating alone is not enough and students need to do something with what they find.

The instructional research on this is fairly consistent: lessons that build genuine AI critical thinking tend to share three features. They require students to compare AI output against an authoritative source rather than treating the AI as the authority (Tlili et al., 2024). They ask students to iterate, treating the first AI response as a draft rather than a finished product (Kim & Tan, 2023). And they make the act of verification visible through a structured protocol, rather than leaving students with a vague directive to check their work (Schneider, 2025). You can build all three of these into a single classroom routine using the process of Compare, Catch, and Correct.

Start by giving students a topic specific enough to be verifiable, something in your content area where you know what an accurate answer looks like. Ask them to prompt a chatbot for information on that topic and save the response exactly as it arrives. Then ask them to work through three questions before they use anything the AI produced:

1. What in this response can you confirm with a second source?

2. What feels off, even if you cannot yet explain why?

3. What would you change, add, or remove to make this response more accurate?

The third question is the one that most lessons skip, and it is the most important of the three. Spotting an error requires enough knowledge to recognize that something is wrong, while fixing it requires understanding why it is wrong and what a better answer actually looks like, which is a significantly deeper cognitive demand. That gap between error identification and error correction is where the real learning happens, and it is the move that transforms critical thinking from a passive evaluation exercise into an active knowledge-building one.

Spotting an error is a critical thinking skill, but fixing it is a deeper one.

A practical note on using the reading list story in class: it works well as an anchor for this activity precisely because it is not abstract. Bring the actual Chicago Sun-Times situation to your students and ask them what signals, if any, might have tipped off an alert reader before the list went to print. Then give them a short AI-generated text in your subject area and ask them to apply those same signals. What would a careful editor catch? What would a less careful one miss? In my experience, that conversation tends to be illuminating, and occasionally humbling, in the best possible way, including for the teacher.

What I keep returning to when I think about the reading list story is that it did not fail because the people involved lacked critical thinking skills. It failed because of the habit of pausing to verify before accepting something that looks authoritative. In our rush to share or collect information, it has to be practiced regularly and deliberately, in conditions that make the stakes feel real enough to matter.

What generative AI has done is change the conditions. The old surface signals of credibility, such as a polished format, confident tone, or familiar author name, are now things AI can generate on demand and at scale. That means asking students to go one level deeper than we may have previously required: not does this look right, but does this hold up? Not does the source seem credible, but can I independently confirm what the source is claiming? These are not new questions in the world of information literacy. They are simply more necessary now and more frequently needed than they have ever been.

Before the final post in this series, I want to invite you to try the Compare, Catch, and Correct routine with one class, in one lesson where students are already using or consuming AI output. Try using just one of the structured verification steps with a specific question attached. See what your students actually catch when they are explicitly asked to look. Then come back and tell me in the comments: When you gave students the space to question the AI, what surprised you most about what they found? Most of all, ask them how they can make it right with novel or creative ideas.

References

Doe, J., & Lee, K. (2026). AI hallucination from students’ perspective: A thematic analysis. arXiv Preprint, 2602.17671. https://arxiv.org/abs/2602.17671

Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12), 1–38. https://doi.org/10.1145/3571730

Kim, T. W., & Tan, Q. (2023). Repurposing text-generating AI into a thought-provoking writing tutor. arXiv Preprint, 2304.10543. https://arxiv.org/abs/2304.10543

Langreo, L. (2026, May 28). Teachers say lack of AI guidance is a major problem. Education Week. https://www.edweek.org/technology/teachers-say-lack-of-ai-guidance-is-a-major-problem/2026/05

Schneider, D. (2025). LEADERS framework for AI impact and bias lessons. Code.org. https://code.org/ai

Tlili, A., Shehata, B., Adarkwah, M. A., Bozkurt, A., Hickey, D. T., Huang, R., & Agyemang, B. (2024). What if the devil is my guardian angel: ChatGPT as a case study of using chatbots in education. Smart Learning Environments, 11(1), 24. https://doi.org/10.1186/s40561-023-00237-x

Valeri, M., Karpouzis, K., & Yannakakis, G. N. (2024). Generative AI hallucinations and pedagogical risk: A review of empirical findings. Education and Information Technologies, 29(16), 21345–21372. https://doi.org/10.1007/s10639-024-12873-y