# How do you teach critical thinking when answers are instant?

Date: 2026-02-17 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/teach-critical-thinking-answers-instant/

**Answer:** You stop grading the answer and start grading the judgment around it: how students question, check, compare and improve what an AI gives them. The AI becomes the material to think about, not a shortcut past thinking.

Critical thinking is the habit of asking whether something is true, how we know, and what else might explain it. For most of the history of schooling, finding an answer took effort, and some of that thinking happened along the way. A chatbot removes the effort of finding. It does not remove the need to judge. So the teaching moves to the point where an answer arrives: what does a careful person do next?

## Why instant answers make it harder

Research on adults at work shows why this needs deliberate practice. A Microsoft Research team surveyed 319 knowledge workers about 936 real tasks done with generative AI, and published the results at the CHI 2025 conference. People who trusted the AI more reported doing less critical thinking; people more confident in their own skill reported doing more. The researchers also found that the thinking did not disappear but changed shape, toward checking information, fitting the AI's output into the work, and overseeing the task as a whole. Those are skills a classroom can train directly.

## What changes in the classroom

Three shifts make the difference.

**Assess the process, not just the product.** If a polished paragraph can be produced in seconds, the paragraph alone shows little. Ask for the prompt, the AI's reply, and the student's notes on what they kept, changed or rejected, and why.

**Treat AI output as a source to evaluate.** Students already learn to question a website or an advertisement. An AI reply belongs in the same category: useful, often right, sometimes wrong in confident language.

**Build confidence in the student's own knowledge.** The survey's finding cuts both ways. Students who know a topic well are the ones most likely to notice when the AI is wrong, so content knowledge is not old-fashioned; it is the tool for checking.

OpenAI's own guide for teachers, published in 2023, points the same way: it collects examples such as a teacher who tells students that chatbot answers may not be accurate and must be confirmed against primary sources.

## Which exercises work with AI in the room?

Here are three that work at almost any level from upper elementary to university. Each takes one class period.

**1. Verify the AI.** Give pairs of students one AI answer on a topic you have just taught. Their task is to find every factual claim, mark each as confirmed, wrong or unverifiable, and show the evidence. A worked setup for a history class:

| Step | What students do | Time |
|---|---|---|
| 1 | Read the AI's 200-word answer on the causes of a historical event | 5 min |
| 2 | Underline each factual claim (names, dates, numbers, causes) | 10 min |
| 3 | Check each claim in the textbook or a primary document | 20 min |
| 4 | Rewrite the answer with corrections and one sentence on what the AI got wrong | 15 min |

Prepare the answer yourself in advance, so you know where it is weak. If it happens to be flawless, that is a lesson too: students learn that checking is how you earn trust, not a hunt for errors.

**2. Argue the other side.** Students write a short position on a debatable question, then ask the AI to argue against it as strongly as it can. They must answer the best objection in their own words. A prompt they can use:

```prompt
I am a grade 10 student. My position is: [one sentence]. Act as a fair but tough debate opponent. Give me the two strongest arguments against my position, each in three sentences, with the kind of evidence someone would use. Do not tell me which side is right. Then ask me one question I would need to answer to defend my view.
```

This uses the AI for what it does well, producing counterarguments quickly, while the thinking stays with the student, who has to weigh them. OpenAI's teacher guide describes a similar role-play use, with the AI acting as a debate partner.

**3. Source hunt.** Ask the AI a factual question and request its sources. Students then try to find each source and check whether it exists and says what the AI claims. Chatbots without web access sometimes produce references that look real and are not, and tools with web search can still summarize a page inaccurately. Students who have tracked down one invented citation remember the lesson.

> **Try it:** Before your next class, ask a chatbot a question from your current unit and save the reply. Mark every factual claim yourself, then decide which of the three exercises fits it best. If you find an error, keep the answer: it is now a ready-made lesson.

For all three, the student's written reasoning is what you assess. A simple rubric with three lines works: did they identify the claims, did they check them against a real source, and did they explain their decision clearly? Students who never touch the AI in class still benefit, because the exercises teach a posture they will need outside it: treat a fast answer as a first draft from a confident stranger, and decide for yourself what to keep.

## Sources

- The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers: https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/
- Teaching with AI: https://openai.com/blog/teaching-with-ai