Does AI lower your IQ? No study has shown it

Where the claim comes from

Three pieces of research get quoted as if they were about IQ. None of them is.

• The MIT Media Lab essay study (Kosmyna and others, 2025, a preprint, 54 people). It recorded brain activity while people wrote essays and found lower engagement in the group using ChatGPT. It contains no IQ test. Its authors publish a list of words they ask the media not to use about their work, including "dumb", "stupid" and "brain rot". What that study did and did not find.
• A UK survey (Gerlich, Societies, 2025, 666 adults). People who said they relied on AI more also scored lower on critical thinking questions. That is a correlation from a questionnaire. It cannot say which came first, and it did not measure IQ.
• A survey of office workers (Lee and others, Microsoft Research and Carnegie Mellon, CHI 2025, 319 people). Workers who trusted the AI more reported doing less critical thinking on a task. Workers who trusted themselves more reported doing more. It measured what people said about their own effort, not their ability.

The one study that did use an intelligence test

One small preprint from Argentina (Benitez and others, 30 people) gave four subtests of a standard adult intelligence scale before and after four weeks of using ChatGPT. It found faster task work and no change on the four subtests.

That is not proof that AI is harmless either. Thirty people, four weeks and four subtests is too small and too short to settle anything. It is simply the only direct test so far, and it did not find a fall.

What about falling IQ scores in general?

Some countries have reported falling test scores in recent decades. Those falls were measured before chatbots existed. A Norwegian study in PNAS (Bratsberg and Rogeberg, 2018) found declines among men born after 1975. Whatever caused them, it was not a tool released in late 2022.

What studies do show

The real findings are narrower than "AI makes you less intelligent", and more useful.

• When AI hands over the answer, people learn less from that task. In a field trial with about 1,000 high school students (Bastani and others, PNAS, 2025), students who practised maths with plain GPT-4 scored 17% lower on the later exam than students who never had it. The details of that trial.
• Hints behaved differently from answers. In the same trial, a tutor version that gave hints largely avoided the exam drop. In three experiments with 1,222 adults (Liu, Christian and others, COLM 2026, peer reviewed), people given direct answers solved fewer problems alone afterwards, and people who used the AI for hints did not.
• Order mattered. In the MIT preprint, people who wrote on their own first and then used ChatGPT showed higher brain connectivity than people who used it from the start.

So the question worth asking is not "is AI lowering my IQ?" It is "am I doing the thinking, or is it?" That one you can change today.

What to do with this

Before you ask an AI something you want to learn, write your own attempt first. Then ask it to check your thinking. That is the pattern the studies above point to, and it costs you about a minute.

Questions