We’ve all asked AI questions that we wouldn’t dare share with another living soul.
There’s a particular half-second worth noticing here. You’ve asked the chatbot a question, the answer arrives, and for a split second (almost too fast to notice) we consider further verification or simply accept the answer and move on.
Most of us know which way that decision tends to go, more often than we’d probably admit.
That small moment is really what this blog is about — not whether AI is good or bad for the mind, but what tends to happen in that instant, and what the research can tell us about it.
Six things worth knowing follow, some reassuring, some less so, all grounded in what psychologists, educators, and workplace researchers have actually found.
Most of the studies referenced here are from the last two years, which means the field is still young by scientific standards, and the tools themselves keep changing faster than the research can fully keep pace with.
1. The Evidence Pulls in Two Directions
A large UK survey of 666 adults found a strong link between frequent AI use and weaker critical-thinking scores, with the pattern most pronounced among younger, heavier users (Gerlich, 2025).
Around the same time, other research found AI-assisted learning improving people’s higher-order thinking in controlled classroom settings (Deng et al., 2025).
Both findings are real. Both come from careful work. They simply point in different directions, and that tension isn’t a flaw in the research — it points to the fact that different conditions may yield different results.
2. The Quiet Handover
There’s a name for what happens when that half-second pause stops happening: cognitive offloading, or letting an external tool carry a piece of mental work we’d otherwise do ourselves (Risko & Gilbert, 2016).
It isn’t new — we’ve offloaded memory to notebooks and arithmetic to calculators for years without much hand-wringing.
What’s different with AI is how seamless the handover feels, and how rarely it registers as a decision at all. In Gerlich’s (2025) survey, this was one of the clearest threads linking heavier AI use to lower critical-thinking scores.
3. Convenience is the Real Culprit
The very thing that makes AI appealing — speed, low effort, a good-enough answer without the work — is often what erodes the instinct to hold information to higher standards.
A review of AI dialogue systems found people tend to favour quick AI answers over their own slower reasoning, sometimes accepting biased or inaccurate output simply because questioning it felt like more effort than it seemed worth (Zhai et al., 2024). This is called automation bias.
Automation-bias research points to something similar: over-reliance shows up most where an answer is genuinely hard to verify, not just when someone is distracted or juggling several things at once (Lyell & Coiera, 2017).
The harder something is to check, the more tempting it becomes not to.
None of this means AI is quietly working against us. It means ease and vigilance tend to trade off against each other, and it’s worth knowing that before you’re in the moment where it matters.
4. The Relationship You Build With It
A review of 67 studies on university students found the same AI tool producing very different outcomes depending on how students used it. Asked to challenge an argument, question a conclusion, or generate counter-examples, it sharpened their thinking. Handed over as a straightforward answer machine, it dulled it (Li et al., 2026). The tool itself doesn’t decide which way that goes. What you ask of it does.
There’s a useful distinction behind that finding. Researchers separate convergent thinking — narrowing toward a single sound judgement, which is largely what critical thinking is — from divergent thinking, or generating many possible ideas, which sits closer to creativity.
Left to unstructured use, AI tends to boost divergent thinking while quietly eroding convergent thinking (Li et al., 2026). That can feel deceptive from the inside: you may well feel sharper and more prolific, full of ideas, even as the specific skill of weighing and judging between them gets a little less exercise.
You can see a smaller version of the same pattern at work. In a study of knowledge workers using AI on real tasks, one common thread was people having to actively check whether the AI had correctly grasped something specific to their field — a regulation, a technical term, a local context the AI had no way of knowing on its own (Lee et al., 2025). That checking is a form of thinking in its own right. It just doesn’t always feel like it, because it looks like editing rather than reasoning from scratch.
5. When Trust Outruns Your Own Judgement
That same study of 319 knowledge workers found something worth carrying into your own habits: the more people trusted the AI’s output, the less they scrutinised it — and the more they trusted their own judgement, the more critical thinking they reported doing (Lee et al., 2025).
Put simply, the risk isn’t really about how often you use AI. It’s about the moment your confidence in the machine quietly overtakes your confidence in yourself, since that’s usually when the checking stops.
Trusting yourself is only half of it, though. Sometimes the checking doesn’t happen not because someone couldn’t be bothered, but because they genuinely didn’t have the grounding to do it — in the same worker study, a shortage of domain expertise was one of the most common reasons people gave for not properly scrutinising AI output (Lee et al., 2025).
Relatedly, across the wider population, higher education levels softened the AI–critical-thinking link considerably (Gerlich, 2025).
None of that is a reason for self-blame. It’s a reason to be honest about which of your own judgements are genuinely well-informed, and which are really just based on AI-approved confidence.
6. A few honest questions worth asking yourself
None of this calls for suspicion of every AI answer, or for abandoning the tool altogether — that’s its own kind of exhaustion, and not really what the evidence is asking of us. What seems to matter more is a handful of small, honest checks.
Consider asking yourself these questions:
- Could you still reason through this if the tool weren’t here?
- Is there something you’ve quietly stopped double-checking, that you used to question without a second thought?
- Are you asking the AI to help you think this through, or asking it to think it through instead of you?
None of these are rules, and none demand a particular answer. They’re simply worth noticing, the way you’d notice any habit worth keeping an eye on.
Is AI dulling how we think?
The evidence doesn’t offer a clean yes or no, and it probably shouldn’t.
What research indicates instead is something closer to a habit worth tending: keep using the tool, but keep noticing that instant before you decide to trust it, and every so often, choose to stay in it a little longer.
For all its brilliance, AI will never be able to replace therapists. So, if you need help on your journey, we’re always just a call away!
References
- Deng, R., Jiang, M., Yu, X., Lu, Y., & Liu, S. (2025). Does ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies. Computers & Education, 227, Article 105224. https://doi.org/10.1016/j.compedu.2024.105224
- Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), Article 6. https://doi.org/10.3390/soc15010006
- Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI ’25), Article 1121. Association for Computing Machinery. https://doi.org/10.1145/3706598.3713778
- Li, C., et al. (2026). The cognitive impact of ChatGPT in higher education: A systematic review of critical and creative thinking outcomes. Computers and Education: Artificial Intelligence. https://doi.org/10.1016/j.caeai.2026.100571
- Lyell, D., & Coiera, E. (2017). Automation bias and verification complexity: A systematic review. Journal of the American Medical Informatics Association, 24(2), 423–431. https://doi.org/10.1093/jamia/ocw105
- Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
- Zhai, C., Wibowo, S., & Li, L. D. (2024). The effects of over-reliance on AI dialogue systems on students’ cognitive abilities: A systematic review. Smart Learning Environments, 11(1), Article 28. https://doi.org/10.1186/s40561-024-00316-7