Skip to content
SEARCHIT
Menu
SearchIT
HUMAN PROGRESS

In an RCT with 1,053 students, ChatGPT improved the assignment while critical-thinking training broadened ideas

On one marketing task, ChatGPT Edu raised scores by about 0.86 points out of five; causal training produced more diverse ideas.

1 min readBy NOWPOSSIBLE editorial team
NO IMAGE / EVIDENCE FIRST

IN BRIEF

ChatGPT improved evaluated output; reasoning training improved a different dimension, originality.

WHY IT MATTERS

It shows that LLM access and cognitive training can expand different capabilities rather than substitute for one another.

WHAT MADE IT POSSIBLE

Students in the AI condition could use ChatGPT Edu with GPT-4o during the task; the study did not measure durable skill without AI.

THE STORY

Students were randomized to AI, causal training, both, or control. ChatGPT raised scores by about 0.86 points relative to an estimated 2.09/5 control score, while causal training broadened idea diversity.

WHAT WE KNOW

The randomized design included 1,053 students. [1][2][3]

WHAT WE DON'T KNOW

It is unknown whether the advantage persists without AI or over time.

WHAT HAPPENS NEXT

Replicate across tasks and add delayed AI-free assessments.

SOURCE TRACE

  1. The trial assigned 1,053 first-year undergraduates to causal-reasoning training, ChatGPT Edu access, both, or neither.

    [1][2]

  2. ChatGPT Edu access increased the task score by about 0.86 points relative to an estimated 2.09 control score on a five-point rubric.

    [2]

  3. Causal-reasoning training increased idea diversity, while ChatGPT improved rubric scores, coherence and similarity to expert submissions.

    [2][3]

SOURCES

  1. DP21882 Training Novices to Think, or Giving Them LLMs? Evidence from an RCTCEPR · 2026-08-27 · Class A · discussion paper · PRIMARY_EVIDENCEOpen source
  2. Better evaluations with ChatGPT. Critical thinking broadens ideasBocconi University · 2026-08-27 · Class B · institutional research report · RESULT_DETAILOpen source
  3. Better answers, broader thinking: What students gain from ChatGPT and critical-thinking trainingOpenAI · 2026-08-27 · Class B · collaborator research summary · COLLABORATOR_CONTEXTOpen source

CURRENT EVIDENCE

EARLY EVIDENCE

single-task randomized field experiment with 1,053 first-year students; no longitudinal learning or transfer assessment

WHAT WOULD CHANGE THIS?

  • If other tasks fail to reproduce the advantage or delayed testing shows worse unaided learning, the interpretation should be downgraded.

RELATED PROGRESS