Search and Filter

Little Impact Of ChatGPT On High School Test Scores

Date

AI Overview

Why it matters

Most evidence on generative artificial intelligence evaluates supervised classroom tools rather than unsupervised student use. Understanding how students' actual, independent use of large language models affects learning is an essential, unresolved empirical question.

Study at a glance
  • Question: How does unsupervised, school-related ChatGPT use affect aggregate U.S. high school test scores?
  • Setting and data: United States high schools and grades 3–8 across up to 43 states, 2017–2025, covering thousands of schools and districts.
  • Method: Fixed effects panel regression comparing test score changes across areas differing in summer-to-school-year ChatGPT activity drops.
  • AI tools: ChatGPT
Key findings
  • Higher educational ChatGPT use had an insignificant effect on high school test scores that was close to zero, bounded below roughly 0.04 standard deviations.
  • Across subject and grade-level exams, nearly all estimated test score impacts were small and statistically indistinguishable from zero.
  • Initial evidence suggests gains for lower-performing schools and negative impacts for higher-performing schools, but this pattern was not statistically robust.

Summary drafted with AI from the full paper and may contain errors. See the full paper for details.

In educational settings, AI can be used as a learning aid, but can also be used to avoid schoolwork, passing classes while learning little. Most existing evidence on AI's educational impact comes from trials of supervised instructor-encouraged AI usage, rather than on unsupervised usage which is of the most concern. This paper uses the dropoff in ChatGPT activity during non-school summer months to identify areas with heavy educational AI use, and then estimates the effect of that usage on aggregate United States high school test scores. Areas with higher educational AI use were not significantly different from other areas, with an insignificant effect estimated to be close to zero, and in the preferred specification bounded below roughly 0.04 standard deviations of test score performance in absolute terms.