AI · Preregistered online randomized experiment with incentivized, occupation-specific writing tasks.
Experimental evidence on the productivity effects of generative artificial intelligence
Science · 2023
Summary
In this study, ChatGPT reduced average time by 40% and raised output quality by 18%; inequality fell, and exposed workers were 2 times as likely to report real-job use two weeks later and 1.6 times as likely two months later. Founder use: Evaluate AI features with randomized task-level tests that include quality and time, then measure sustained adoption rather than assuming a novelty effect equals retention.
For your marketing
Possible founder application
Evaluate AI features with randomized task-level tests that include quality and time, then measure sustained adoption rather than assuming a novelty effect equals retention.
Reported result
What the study found
ChatGPT reduced average time by 40% and raised output quality by 18%; inequality fell, and exposed workers were 2 times as likely to report real-job use two weeks later and 1.6 times as likely two months later.
Research question
What the authors investigated
What is the causal effect of ChatGPT access on productivity, quality, inequality, and subsequent workplace adoption in professional writing tasks?
Design
Preregistered online randomized experiment with incentivized, occupation-specific writing tasks.
Sample and setting
453 college-educated professionals performing midlevel professional writing tasks; half were randomly exposed to ChatGPT.
Scope and transfer
Limitations to keep in view
Short online tasks, selected professionals, and one model/version; measured writing performance is not the same as long-run business productivity, customer value, or safe deployment.
Research sources
Publisher page directly retrieved with Firecrawl; scrape reference 01a1227f-03d4-711b-90ec-fe5cebeebe65. Claims are limited to the reported article record. Correction notices were not separately checked.