Do Chatbots Erode Critical Thinking?

The Headline You've Definitely Seen

"Chatbots are lowering critical thinking." Within a week, the line was everywhere, from The Guardian to the tech news feeds. The image is effective: a generation that asks ChatGPT for everything and unlearns how to think.

The healthy reflex when faced with a hook like that is exactly the one the study is supposed to measure. Before nodding along, you look at what the research actually says. Because if you swallow the conclusion without checking it, you fall straight into the trap the article claims to denounce.

The Study Behind the Headlines

The most cited paper comes from Michael Gerlich, at SBS Swiss Business School, published in early 2025 in the journal Societies. 666 participants were surveyed on their use of AI tools, their tendency to delegate mental tasks, and their level of critical thinking. 50 of them sat for an in-depth interview.

The result fits in one sentence: the more intensively you use AI, the lower your critical thinking score. And the gap widens among the young. The 17-to-25 group delegates the most to the machine and posts the weakest scores. The 46-and-over crowd does the opposite.

One more detail matters: education level softens the blow. At equal usage, the more educated keep a sharper critical eye. The tool doesn't hit everyone the same way.

Delegating a Task, or Delegating Judgment

The mechanism Gerlich points to has a name: cognitive offloading. You hand a machine a mental effort you used to do yourself. We did it with the calculator for mental arithmetic, with GPS for our sense of direction. Nobody really complains.

The thing is, this time the delegated effort isn't a task, it's a judgment. Assessing whether information is credible, spotting shaky reasoning, weighing two options: these are operations you don't outsource without consequences.

The difference with GPS is cruel. A GPS that gets it wrong, you notice by missing your exit. A chatbot that serves you fragile reasoning with total confidence, you only catch it if you challenge it. And challenging it is precisely the muscle that's falling asleep.

The Moment You Let Go

A second study sheds light on the when. Researchers at Microsoft and Carnegie Mellon surveyed 319 professionals across 936 real work situations involving AI. Their finding is finer than a flat "AI makes you passive."

What makes critical thinking slip is trust. The more you trust the AI, the less you reread its answer. The more confidence you have in your own skills, the more you stay in control. And it's on the tasks deemed low-stakes that people ease off the most.

Take a mundane case. You ask a chatbot to summarize a report you don't have time to read. The summary looks tidy, the tone is assured, you paste it into your email.

You've just made a decision without reading the source. Not out of laziness, but because the task felt minor. That's exactly where judgment slips into autopilot.

Now, the Part the Headlines Forget

Here's where you need to slow down. All these studies show a correlation. None demonstrates a cause.

The difference carries weight. A correlation says two things move together. It doesn't say which one pushes the other.

It's entirely possible that the people with the weakest critical thinking are simply the ones who lean most on AI. In that case, AI isn't deepening the problem: it's revealing it. The arrow would point the other way.

It's the classic trap of causal reasoning. You observe that ice cream sales and drownings rise at the same time, and you conclude that ice cream makes you sink. The real cause is summer pushing both. With AI and critical thinking, we haven't yet found the equivalent of summer: the hidden variable that would raise both together could be age, education level, or self-confidence.

Gerlich says it himself, in black and white: his study "does not establish a direct causal relationship." The data is self-reported, meaning people estimate their own usage and their own critical thinking. And all this research is cross-sectional: a snapshot at one moment, not a follow-up over time. To settle which way the arrow points, you'd need longitudinal studies tracking the same people over several years. They don't exist yet.

At Data & Society, researcher Briana Vecchione sums up the real state of the file: the results are "very preliminary." They show mental processes that cost less effort. Nothing more, for now.

What Stays True Once You Strip Out the Hype

Acknowledging a study's limits isn't burying it. Several independent works point in the same direction, and the cognitive offloading mechanism is credible: we've long known that a capacity you stop exercising goes dull. That holds for memory, and probably for judgment.

The genuinely usable takeaway sits in a nuance. According to Microsoft, the decisive variable is the relationship you have with the tool, more than the tool itself. Delegating after you've checked, with confidence, keeps judgment switched on. Delegating because it's faster and you don't feel like looking switches it off without a sound. The same prompt, two opposite trajectories depending on what's going on in your head the moment you hit send.

A study can measure how many people eased off their attention. It can't decide for you to switch it back on the next time a chatbot answers you with a little too much comfortable confidence.