The effect of large language models on worker retention among freelance writers: evidence from a difference-in-differences design

[SYNTHETIC full text — teaching stand-in for an open-access PDF.]

1. Introduction
This paper studies the labour-market impact of the technology described in the
title. We study how the adoption of large language models affects worker retention among freelance writers. Using a difference-in-differences design covering the 2024–2025 period, we estimate the impact of access to large language models on the outcome of interest. We find a modest but significant improvement. The results are robust to alternative specifications and are concentrated among less-experienced workers. We discuss implications for how large language models reshapes the tasks and skills that make up these jobs.

3. Data and design
Design: did. The sample is drawn from administrative and survey
sources over the study window. Table 3 reports the main estimate.

4. Results
Main estimate (Table 3, column 4): the point estimate is 0.171
(standard error 0.072) [SEE PAGE 7]. Effect class:
positive. Heterogeneity analysis on page 9 shows the effect concentrates
among less-experienced workers.

6. Conclusion
We discuss how the technology reshapes the tasks and skills composing these jobs.
