AIORG-W010 Peer-reviewed research

Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence

Noy and Zhang 2023 · Science adh2586

peer reviewed journal article

Recorded claims
2
coded from the inspected source
Evidence classes
1
independent-empirical
Recorded limitations
4
stated, never hidden

From the wave-0 evidence ledger

Recorded claims

Each claim states what the inspected source says, at the recorded location—bounded by its scope and graded by its confidence. Nothing here is a synthesis across works.

In a randomized experiment with 453 college-educated professionals completing occupation-specific writing tasks, ChatGPT access reduced completion time by about 40% and increased evaluated output quality by about 18%.

confidence: high within study independent-empirical

Scope: Short professional writing tasks; task productivity, not realized firm performance.

Benefits were larger for initially lower-performing participants, compressing performance inequality on the studied tasks, but the experiment did not test whether this convergence persists after workers stop using the tool or across interdependent workflows.

confidence: high for immediate heterogeneity; low for persistence beyond study independent-empirical

Scope: Experimental task performance and the study window only.

Boundaries

Limitations & independence

Recorded at coding time, carried with the work forever. A claim without its limits is not evidence.

Recorded limitations

  • 453 college-educated professionals on short writing tasks
  • Participants and evaluators operated outside a continuing organization
  • No long-run learning, wage, or firm outcome
  • Early ChatGPT model

Source independence

Independent academic randomized online experiment; simulated professional tasks rather than embedded firm outcomes.