When the computer makes things up
Ask any AI chatbot for the source of a famous quote and a meaningful share of the time you'll get a confident, complete, invented citation. A book that doesn't exist. A real historian on a topic they've never written about. The verification habit isn't new, just newly urgent.
Try this experiment yourself. Ask any AI chatbot for the source of a famous quote, one you can check independently. A meaningful share of the time you'll get a confident, complete, invented citation. A book that doesn't exist. A real book with a fake page number. A real historian on a topic they've never written about.
This is called a hallucination, and it isn't really a bug. The technology works by predicting what a plausible answer looks like, and plausibility and truth happen to overlap most of the time, but not always.
The skill that matters most: verification
Verification is the habit of checking a claim against a second source before treating it as true. Encyclopaedias, librarians, and any decent journalism teacher have been pushing it for a century, and AI hasn't really changed the habit so much as raised the cost of skipping it.
That's most of what critical thinking is in an AI context: the small reflex to check before believing. Verification builds on the more basic trust-calibration habit, the one that asks "are you sure?" before accepting an answer in the first place.
What confident phrasing actually buys
A January 2025 study in Societies surveyed 666 people across age groups on AI use, cognitive offloading, and critical thinking. Younger participants leaned on AI most heavily and scored lowest on critical thinking measures, and the mechanism the researchers identified was cognitive offloading: when an external tool does the thinking for you, your own thinking thins out.
A 2024 paper in the Journal of Pediatric Psychology found that parents rated ChatGPT-generated child health information as more trustworthy and accurate than the equivalent written by actual paediatricians, even though the AI text hadn't been fact-checked. The trust came from how it sounded.
Confident phrasing buys trust that the content hasn't earned, and the pattern shows up everywhere once you start looking. The flipside also matters: Oxford found measurable brain changes in teenagers who'd stopped studying maths, and verification is one of the active cognitive habits at risk when AI removes the friction.
Why information literacy needs a redesign
Most curricula weren't built for any of this. Information literacy in schools tends to be a single module, taught once and largely dropped after a half-term, and AI is now forcing the issue.
UNESCO's AI Competency Framework for Students names verification of AI output as a foundational skill, and the MIT Media Lab's research on cognitive debt, published in 2025, makes a stronger version of the case: that the verification habit may be the single most important thing schools can teach right now.
Three questions for any AI answer
How would I check this? If your child can't name a way to verify the answer, the answer isn't yet safe to use.
What is the AI guessing about? Some questions are factual and some are inventive, and children spot the difference faster than adults tend to expect, once they know to look.
What would I do if a friend told me this at lunch? This one is often the most useful of the three. A child wouldn't repeat a cousin's lunchtime claim in a school project without checking, and the bar for AI should be the same.
A real example to try
Ask the AI for a children's book recommendation on a topic your child likes, then look up whether the book actually exists. Sometimes the title sounds plausible, the author is real, and the book turns out to be invented; the lesson is built into the searching itself.
The verification habit cuts both ways. Children who learn to check sometimes go through a phase of refusing to believe anything anyone tells them, but the phase passes. What you're aiming for isn't permanent suspicion but a reflex that switches on when it should and stays quiet otherwise.
The technology will keep getting better, but the reflex won't get any less useful. (Cognitive debt is the next piece in this series, since the implication for children is uncomfortable enough to deserve one.)
