CRBC News
Culture
Society
Security
Health
Conflict
Science
Technology
Politics
Environment
Lifestyle
Economy

Learning with ChatGPT Produces Shallower Understanding, Large Study Finds

A PNAS Nexus analysis of seven experiments with over 10,000 participants found that people who relied on AI chatbots like ChatGPT developed shallower knowledge than those who used traditional web searches. AI users produced shorter, more generic advice even when shown the same facts, indicating reduced engagement with material. Researchers warn that LLM summaries can turn active learning into a passive process, and other studies link heavy AI reliance to declines in critical thinking, memory, and academic performance.

02:54 AM, 11/23/2025Technology
Learning with ChatGPT Produces Shallower Understanding, Large Study Finds

A new analysis published in PNAS Nexus suggests that relying on AI chatbots like ChatGPT to “learn” a topic can produce shallower knowledge than doing the work yourself via a traditional web search. The study, co-led by Wharton professor Shiri Melumad, pooled seven experiments with more than 10,000 participants to compare learning from AI-generated summaries versus active searching and synthesis from web links.

Study design

Volunteers were asked to learn about a topic and were randomly assigned to research using either an AI chatbot (e.g., ChatGPT) or a conventional search engine (e.g., Google). Afterward, participants wrote advice to a friend summarizing what they had learned. In some experiments the researchers even controlled the exact facts shown to each group to isolate the effect of the format and process of learning.

Main findings

The pattern was consistent: participants who used AI-generated summaries produced shorter, more generic advice containing fewer factual details, while those who searched and synthesized information themselves wrote longer, more detailed and thoughtful recommendations. This difference persisted even when both groups were exposed to the same underlying facts, suggesting the format and level of user engagement — not just access to information — shaped learning outcomes.

Why this matters

One core principle of durable learning is active engagement: navigating sources, evaluating claims, and integrating evidence helps build deeper understanding. When an LLM synthesizes and delivers a concise answer, it removes much of that “friction,” which appears to convert an active learning process into a more passive one and reduce retention and depth of knowledge.

Broader context and implications

Other research raises complementary concerns. High-profile studies have reported links between overreliance on AI tools and declines in critical thinking, memory, and academic performance among some users. Meanwhile, AI adoption in education is accelerating: companies are investing in teacher training, and some universities are creating campus-specific chatbots. These trends make it important to balance convenience with strategies that preserve active learning.

Takeaway: AI tools can speed up information access, but they shouldn’t replace the effortful work of finding, evaluating and synthesizing sources if the goal is deep, durable learning.

Practical advice: Use chatbots as research assistants (to summarize or brainstorm), but practice retrieving and synthesizing information yourself—read original sources, compare perspectives, and explain concepts in your own words to strengthen understanding.

Similar Articles

Trending

Learning with ChatGPT Produces Shallower Understanding, Large Study Finds - CRBC News