Experimental evidence on the productivity effects of generative artificial intelligence

Shakked Noy, Whitney Zhang · Science · 13 Jul 2023

Where this sits in the debate

Will AI actually boost productivity and growth? · its finding supports the claim (why studies differ)
Will AI widen inequality? · its finding goes against the claim (why studies differ)

Direction of the finding as read from the abstract by an AI model (AI-written).

Key takeaway AI-written

A preregistered experiment assigning 453 college-educated professionals writing tasks found that access to ChatGPT cut average completion time by 40% and raised output quality by 18%, with the gap between weaker and stronger writers narrowing. Two-thirds of exposed workers reported using ChatGPT in their actual jobs weeks later, making this one of the earliest rigorous demonstrations that generative AI both raises output and reduces within-occupation inequality for a common white-collar task.

The paper in three steps AI-written

Written by an AI agent from the paper's abstract, which is the primary source and is quoted below.

The question

Does using a generative AI chatbot change how fast and how well professionals complete everyday writing tasks?

The approach

A preregistered online experiment: 453 college-educated professionals were given occupation-specific, incentivized writing tasks, with half randomly given access to ChatGPT and the rest working without it.

The findings

Access to ChatGPT raised productivity substantially: average task time fell by 40% and output quality rose by 18%. Gaps between stronger and weaker writers narrowed, and exposed workers were about twice as likely to use the tool in their real jobs two weeks later.

How to read it: experiment. Measures a causal effect in a controlled or randomized setting — strong internal validity; ask how far it generalizes.

Limitations and counterpoints AI-written

The tasks were short, self-contained writing assignments done online for pay, not full jobs — effects may differ for longer or more specialized work. The study measures task output, not employment or wages, and the follow-up window is a couple of months.

Abstract

We examined the productivity effects of a generative artificial intelligence (AI) technology, the assistive chatbot ChatGPT, in the context of midlevel professional writing tasks. In a preregistered online experiment, we assigned occupation-specific, incentivized writing tasks to 453 college-educated professionals and randomly exposed half of them to ChatGPT. Our results show that ChatGPT substantially raised productivity: The average time taken decreased by 40% and output quality rose by 18%. Inequality between workers decreased, and concern and excitement about AI temporarily rose. Workers exposed to ChatGPT during the experiment were 2 times as likely to report using it in their real job 2 weeks after the experiment and 1.6 times as likely 2 months after the experiment.

About this publication

Academic journal. Published in an academic journal — typically peer-reviewed before publication, though indexing alone doesn't guarantee review depth.

Published: 2023 (40 months ago)
Citations: 1426 (OpenAlex)

1426 citations recorded by OpenAlex — too few 2023 papers in this collection for a fair comparison.

□ added to this collection in 2026.

Shakked Noy, Whitney Zhang signed the wemustactnow.ai letter · author affiliation matches seed institution massachusetts institute of technology — part of why this paper is in the collection.

Cited elsewhere in this guide

State of the debate: Will AI actually boost productivity and growth?