Critical Thinking in the AI Age: The Skill Under Siege

Quick answer

Yes, critical thinking is genuinely under pressure from casual AI use, and the evidence is specific about why. RAND's American Youth Panel found 67% of students say using AI for schoolwork harms critical thinking, and a 2025 study in Societies linked heavy AI use to "cognitive offloading" - letting the tool do the thinking instead of you - with the weakening effect strongest in 17- to 25-year-olds. The reassuring half of the story is that critical thinking is not a fixed trait a teenager either has or lacks. It behaves like a trainable skill, and the habit that damages it (accepting AI's first answer) has a mirror-image habit that builds it (treating AI as something to argue with, then check). This article sets out both directions, and what a parent can put in place this term.

Why critical thinking is the skill under the most pressure

Every wave of technology removes effort from somewhere. What makes AI different is that it removes effort at exactly the point where critical thinking used to happen: the moment of forming your own view before you see anyone else's, including the machine's.

RAND's data captures the trend and the worry together. AI use for schoolwork rose from 48% to 62% of students across 2025, and in that same survey, 67% of students said the use harms their critical thinking, sharper among girls (75%) than boys (59%). That is a teenager telling researchers, unprompted, that the tool making homework easier might be making their thinking worse.

The peer-reviewed evidence backs the instinct. Gerlich's 2025 study in Societies, of 666 participants, found heavy AI use strongly associated with cognitive offloading, and offloading associated with weaker critical thinking, most pronounced in 17- to 25-year-olds - uncomfortably close to the age range most Australian secondary students sit in.

None of this makes AI the enemy of thinking. It makes unsupervised, first-answer-accepting AI use the enemy of thinking - a narrower and much more fixable problem than the headline suggests, and one closely related to the wider set of durable skills AI cannot replace.

The good news: critical thinking is trained, not fixed

The most useful reframe for a parent is that critical thinking behaves like fitness, not eye colour. Evidence for Learning, the Australian home of the Education Endowment Foundation's evidence base, rates metacognition and self-regulated learning - noticing your own thinking, planning, checking - as worth around seven months of additional progress, among the highest-impact, lowest-cost things a young learner can do.

Good critical thinking practice is metacognition with a specific target: notice a claim, question it, verify it, decide. That means the fix is not less AI. A teenager who has practised disagreeing with AI often enough that scepticism becomes the default reflex, not an effortful override, gets there through repeated structured contact with the tool - not by avoiding it. This links closely to the broader case for building that habit, covered in metacognition - the AI-age superpower.

What weakens it, and what builds it

The difference between an AI habit that erodes thinking and one that trains it usually comes down to a single decision point: what happens the moment the answer appears on screen.

How to guide this at home

You do not need to supervise every AI conversation your teenager has. You need three habits, repeated until they are automatic rather than effortful.

1. Argue back before accepting. Make "why might that be wrong?" a standing instruction for any AI answer that matters - an essay claim, a maths method, a factual statement. The point is not distrust for its own sake; it is that the act of generating a counter-argument is itself the critical-thinking rep.
2. Apply a two-source rule. Any AI claim that will end up in submitted work, a decision, or an argument needs a second, independent check - a textbook, a teacher, a primary source. This single rule catches the confident-but-wrong answers that AI produces routinely, and it's the same discipline covered in teaching teenagers to fact-check AI.
3. Run the explain-it-yourself test. After using AI to work through something, close the screen and have them explain it in their own words, unaided. If they can't, the understanding was never built - only borrowed. If they can, the AI helped them learn rather than replaced the learning.
4. Praise the disagreement, not just the output. When your teenager pushes back on something AI told them, notice it out loud. That is the exact behaviour RAND and Gerlich's research suggests is being lost, and it responds to reinforcement like any other habit.

Common mistakes parents make

• Banning AI to protect thinking. This removes the practice ground, not the pressure - the teenager still uses AI, just without guidance.
• Treating fluent output as understanding. Polished AI-assisted writing can mask a student who could not reproduce the argument unaided. The explain-it-yourself test catches this quickly.
• Focusing on whether AI was used, not on whether a view was formed first. The RAND and Gerlich findings both point at the same root cause: skipping the moment of independent thought, not the tool itself.
• Treating critical thinking as innate. A "sharp" teenager who never questions AI's answers is not exercising the skill any more than an athletic teenager who never trains is exercising fitness.

The recommendation: treat critical thinking as a muscle under a new kind of strain, not a lost cause. Put the two-source rule and the explain-it-yourself test in place this week, praise the disagreements more than the polished answers, and keep AI in the room rather than banned from it. Done this way, the same technology that RAND and Gerlich found eroding critical thinking in unsupervised use becomes the sparring partner that builds it - and a teenager who leaves school able to question a confident, fluent, wrong answer has an advantage that compounds well beyond the broader case for AI education in Australia.