This site is a work in progress and has not been widely shared. Content may contain errors. Feedback is welcome.
This site is undergoing review. Some annotations were human-generated, some AI-generated — all are being verified.
Back to papers

Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms

Demirci, Hannane, Zhu

2024Management Science (noted in prior annotations)20 citations
Observational labor marketCausal
LLM / Generative AIComputer Vision / Image AIPlatforms / gig economyWriting / contentSoftware / coding
DOI: 10.2139/ssrn.4991774
Summary

Demirci, Hannane, and Zhu use difference-in-differences analysis on 1.4 million job posts from a global freelancing platform (July 2021-July 2023) to study the heterogeneous short-term impacts of ChatGPT and image-generating AI on demand for different types of freelance work.

Main Finding

The introduction of ChatGPT led to a 21% decrease in job posts for automation-prone jobs (writing, software/web development, engineering) compared to manual-intensive jobs within 8 months, with writing jobs experiencing the largest decline (30%), and image-generating AI led to a 17% decrease in graphic design and 3D modeling job posts.

Primary Datasets

Google Trends

Other

Proprietary data from undisclosed global online labor market platform via API (1,388,711 job posts from July 2021 to July 2023); Google Trends Search Volume Index

Secondary Datasets

Felten et al. (2021)

AI-focused

AI Occupational Exposure Index (AIOE) from Felten et al. (2021, 2023)

Key Methods
Difference-in-differences comparing automation-prone vs manual-intensive job clusters before/after GenAI releases; multiple DiD estimators including TWFE, Synthetic DiD, Callaway-Sant'Anna DiD, and negative binomial models
Sample Period
2021-2023
Geographic Coverage
International
Sample Size
1,388,711 job posts from 541,828 employers; final analysis sample of 1,218,463 job posts across 61 countries aggregated to cluster-week-country level
Level of Analysis
Occupation, Country
Occupation Classification
Custom job clusters created via Louvain algorithm on skill co-occurrence network, mapped to O*NET via AIOE index
Industry Classification
None
Notes
Management Science [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments. [Claude classification]: Uses Louvain clustering algorithm (network analysis/ML) to categorize job posts into clusters. Multiple robust DiD estimators employed including Synthetic DiD (Arkhangelsky et al. 2021) and doubly robust DiD (Callaway-Sant'Anna 2021). Google Trends data used as external validation. The paper treats GenAI releases as natural experiments.