Does AI automatically weaken children's thinking?

No. The review finds positive, negative and mixed effects. Task, age, guidance and the specific use all matter.

What does the decline in Google use mean?

It shows a change in how information is accessed. It proves neither that AI is the sole cause nor that learning performance has declined.

Why is the evidence gap so large?

Eighty percent of the reviewed studies examined people in higher education. Younger children were substantially underrepresented.

Children use AI instead of Google: what 173 studies show

What this is about

A review of 173 international studies presented on September 22, 2026 reaches an uncomfortable conclusion: generative AI can strengthen or weaken critical thinking and problem solving, depending on how it is used. Yet robust evidence is especially scarce for younger pupils.

At the same time, Norway's media authority shows a sharp behavioral shift. Among 9- to 18-year-olds, the share using Google fell from 72% in 2024 to 47% in 2026. Already 39% of 11- to 12-year-olds use AI services. These figures measure usage, not learning outcomes or harm.

What the research actually shows

Hilde Storrøsæter of the Norwegian University of Science and Technology and a Czech-Norwegian team reviewed studies published from 2022 through spring 2025. Of 139 studies covering critical thinking, 67 reported only positive effects, 33 only negative effects and 39 mixed effects. For problem solving, 57 of 103 studies found exclusively positive effects, 23 negative effects and 23 mixed effects.

The apparently clear tally has a major gap: in 80% of the reviewed research, participants were university or college students. Findings about adults cannot simply be transferred to 10- or 14-year-olds whose reasoning and evaluation skills are still developing. Many studies also relied on self-reporting and were geographically imbalanced.

Why it matters

Children increasingly seek answers from systems that provide a finished response while not always making sources visible or reliable. This can help them compare perspectives and ask follow-up questions. It can also skip the part of an assignment in which learners search, organize, doubt and argue for themselves.

Public information services also feel the shift. Visits to Norway's youth portal ung.no fell from 24 million in 2024 to 18 million in 2025; chatbots and Google's AI summaries both influenced the decline. Schools therefore need to protect specific learning processes, not merely allow or ban a tool.

In plain language

AI in learning is like a calculator in mathematics class. Once a pupil understands a method, it can speed up the work. If it is used before a child has built the basics, it may produce the right answer without the necessary reasoning path.

A practical example

An eighth-grade class analyzes a text about climate policy. Group A immediately asks a chatbot for a finished answer. Group B first writes three original claims, checks two primary sources and then asks AI for counterarguments. Both groups submit 500 words, but only the second workflow makes source criticism and independent reasoning visible and assessable.

Scope and limits

The drop in Google use does not prove that AI alone caused it or that learning became worse.
The 173 studies differ widely in method, region, age and task, and many rely on self-reporting.
Because 80% studied higher-education students, conclusions about younger children are especially uncertain.

The review supports neither blanket bans nor uncritical adoption. It points toward age-appropriate rules, assignments with visible reasoning and new longitudinal studies involving schoolchildren.

SEO & GEO keywords

children and AI, generative AI in schools, critical thinking, problem solving, Google Search, Norway, media literacy, AI literacy, classroom, NTNU

Children are switching from Google to AI, with little evidence on learning