The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market
Open accessThe DOI matches its Crossref record; 41 of the 41 references it cites were found in a registry or an index; none of the 21 sources its claims cite is checked for support yet.
Abstract
Generative Artificial Intelligence (AI) holds the potential to either complement knowledge workers by increasing their productivity or substitute them entirely. We examine the short-term effects of the recent release of the large language model (LLM), ChatGPT, on the employment outcomes of freelancers on a large online platform. We find that freelancers in highly affected occupations suffer from the introduction of generative AI, experiencing reductions in both employment and earnings. We find similar effects studying the release of other image-based, generative AI models. Exploring the heterogeneity by freelancers' employment history, we do not find evidence that high-quality service, measured by their past performance and employment, moderates the adverse effects on employment. In fact, we find suggestive evidence that top freelancers are disproportionately affected by AI. These results suggest that in the short term generative AI reduces overall demand for knowledge workers of all types, and may have the potential to narrow gaps among workers.
Overview
CoChat Summary
TL;DR
This paper studies short-term effects of generative AI on freelancers in an online labor market, using ChatGPT’s November 2022 release as a shock. It finds that freelancers in more affected writing-related occupations experienced lower employment and earnings, with similar negative effects after image-based generative AI releases.
Key findings
- Following ChatGPT’s release, freelancers in more affected occupations saw monthly jobs decrease by 2% (s.e.=0.004) and monthly earnings decrease by 5.2% (s.e.=0.016), statistically significant at the 1% level.
- Extensive margin: the probability of receiving at least one job in a month fell by 1.2 percentage points, approximately a 10% drop compared to baseline employment. Intensive margin: conditional on working, the number of jobs fell by 4.7% (s.e.=0.011), and income fell by 5.1% (s.e.=0.037, noisier).
- Image-based generative AI releases (DALL-E 2 and Midjourney, coded post-April 2022) showed qualitatively similar negative effects on design, image-editing, and art freelancers’ jobs and earnings.
- Heterogeneity: the paper does not find evidence that high-quality service moderates adverse effects, and finds suggestive evidence that top freelancers are disproportionately affected; for image-based services, above-median prior earnings had reductions of 7% in jobs and 14% in income, both significant at the 1% level.
Methods
- Data were scraped from Upwork’s freelancers API, covering January 2022 through April 2023 at the freelancer-month level. The final data set consists of 92,547 freelancers (text; Table 1 lists 92,457), restricted to freelancers offering services in a single occupation (approximately 85%). Outcomes are monthly number of jobs and monthly income, winsorized at the 1% and 99% levels.
- Main design is a difference-in-differences specification: Post (post-November 2022) × treatment (writing, proofreading, copy-editing), with freelancer and month fixed effects, time-varying freelancer controls, and standard errors clustered at the occupation level. An event-study specification tests pre-trends.
- Robustness checks include alternative controls and clustering, wild bootstrap standard errors, coarsened exact matching on pre-release characteristics, and a separate analysis of image-based generative AI releases.
Limitations
- The paper states it cannot observe off-platform employment, so it cannot determine whether reduced platform employment reflects reduced total employment or substitution to other employment; it cites Horton (2017) that multi-homing is rare and online/offline hiring are not very substitutable. It also cannot evaluate whether high-skilled Upwork workers are highly skilled in the general population.
- The paper states data include only searchable freelancers available through the API, with no estimate of the share of non-searchable freelancers; fraudulent jobs are omitted; only active profiles at data collection are observed, so terminated accounts are missing; profile changes over time are unobserved; and hours data are poorly populated.
- The paper states the analysis is short-term only, long-term implications may differ, and it focuses entirely on negative effects on workers; total welfare implications for all stakeholders are beyond scope. Evident from the text: the design is observational, platform-specific, and uses occupation-level treatment over a short post period.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
0 Valid · 0 Not valid · 11 Not checked · 55 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For example, observing Column 2 we find that freelancers with above median prior earnings suffer an additional reduction of 1.2% in the number of jobs (s.e.=0.007) and a 2.9% reduction in monthly income, though the result is not statistically significant.”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 4.2 Heterogeneous Treatment Effects
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Following the release, the monthly number of jobs on the platform for freelancers in more affected occupations decreases by 2% (s.e.=0.004), and total monthly compensation decreases by 5.2% (s.e.=0.016).”
- Also stated in: § 1 Introduction · plain
“Freelancers in more affected occupations experienced a decrease of 2% in the number of monthly jobs and a decrease of 5.2% in monthly earnings on the platform, following the release of ChatGPT, compared to freelancers in less-affected occupations.”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 4.1 Main Results
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on 2 assumptions not shown here.
- Also stated in: § 1 Introduction · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For image-based services, we estimate reductions of 7% and 14% for freelancers with above median prior earnings, both significant at the 1% level.”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 4.2 Heterogeneous Treatment Effects
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In general, across the various measures of freelancers' quality, we can generally rule out that freelancers' quality moderates the adverse employment effects of generative AI.”
- Also stated in: § 5 Conclusion · hedged
“In addition, we find that offering high-quality service does not mitigate the negative effect of AI on freelancers, and in fact present suggestive evidence that top employees are disproportionately hurt by AI.”
- Also stated in: § 5 Conclusion · plain
“Importantly, we find that top performance and high-quality service do not help mitigate or harness the introduction of generative AI.”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 4.2 Heterogeneous Treatment Effects
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 5 Conclusion · hedged
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The number of jobs decreases by approximately 4.7% (s.e.=0.011), which is substantially larger than the total effect, and the effect of income is approximately the same size, a decrease of 5.1%, though it is much noisier (s.e.=0.037).”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 4.1 Main Results
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on 2 assumptions not shown here.
26 more claims
Create a free account to see all 66 claims and to verify the 13 references still open.
- It may state two separate claims as one. 15
- The quote only partly states this claim. 1
- Probably not a substantial claim. 26
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“previous literature, which focused primarily on a limited subset of skill clusters in the online freelancing market (Hui et al., 2023; Demirci et al., 2023; Qiao et al., 2023; Liu et al., 2023)”
From Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand (2025) · cites this paper as Hui et al., 2023
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 3.1 Data
“Our final data set consists of 92,547 freelancers.”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > Additional information
“No. of Freelancers 92,457”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 4.1 Main Results
“a decrease of 5.1%, though it is much noisier (s.e.=0.037)”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 5 Conclusion
“| Post × Treatment | -0.020*** | -0.052*** | -0.012*** | -0.047*** | -0.051 | | | (0.004) | (0.016) | (0.002) | (0.011) | (0.033) |”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 1 Introduction
“freelancers are 1.2% less likely to receive any employment in a given month”
§ The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market > 4.1 Main Results
“the probability that a freelancer offering writing-related services starts a new job in a given month decreases by 1.2 percentage points”
Citations
Cites
How Much Should We Trust Differences-in-Differences Estimates?
Bertrand, M., Duflo, E., & Mullainathan, S. (2004) · The Quarterly Journal of Economics · doi:10.1162/003355304772839588 · Cited by 10,641
The Skill Content of Recent Technological Change: An Empirical Exploration
Autor, D. H., Levy, F., & Murnane, R. J. (2003) · The Quarterly Journal of Economics · doi:10.1162/003355303322552801 · Cited by 6,205
Bootstrap-Based Improvements for Inference With Clustered Errors
Cameron, A. C., Gelbach, J. B., & Miller, D. L. (2008) · Review of Economics and Statistics · doi:10.1162/rest.90.3.414 · Cited by 4,417
Robots and Jobs: Evidence from US Labor Markets
Acemoglu, D., & Restrepo, P. (2020) · Journal of Political Economy · doi:10.1086/705716 · Cited by 4,088
The growth of low-skill service jobs and the polarization of the US labor market
Autor, D. H., & Dorn, D. (2013) · American Economic Review · doi:10.1257/aer.103.5.1553 · Cited by 3,824
Skills, Tasks and Technologies: Implications for Employment and Earnings
Acemoglu, D., & Autor, D. (2011) · Handbook of Labor Economics · doi:10.1016/s0169-7218(11)02410-5 · Cited by 3,799
GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models
Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023) · arXiv · doi:10.48550/arxiv.2303.10130 · Cited by 2,330
Experimental evidence on the productivity effects of generative artificial intelligence
Noy, S., & Zhang, W. (2023) · Science · doi:10.1126/science.adh2586 · Cited by 2,150
Skill‐Biased Technological Change and Rising Wage Inequality: Some Problems and Puzzles
Card, D., & DiNardo, J. E. (2002) · Journal of Labor Economics · doi:10.1086/342055 · Cited by 1,714
Artificial Intelligence in Human Resources Management: Challenges and a Path Forward
Tambe, P., Cappelli, P., & Yakubovich, V. (2019) · California Management Review · doi:10.1177/0008125619867910 · Cited by 1,609
Research works printed but not linked: 5
- ACEMOGLU, D. AND P. RESTREPO (2018): "Artificial intelligence, automation, and work," in The economics of artificial intelligence: An agenda, University of Chicago Press, 197-236.
- CAO, X. AND D. ZHANG (2020): "The impact of forced intervention on AI adoption," Available at SSRN 3640862.
- CHENG, Z., D. LEE, AND P. TAMBE (2022): "Innovae: Generative ai for understanding patents and innovation," Available at SSRN 3868599.
- ---(2023): "Occupational heterogeneity in exposure to generative ai," Available at SSRN 4414065.
- ---(2017): "Price floors and employer preferences: Evidence from a minimum wage experiment," Available at SSRN 2898827.
Cited by
Foundational Challenges in Assuring Alignment and Safety of Large Language Models
Anwar, U., Saparov, A., Ramírez, J. R., Paleka, D., Turpin, M., Hase, P., Lubana, E. S., Jenner, E., Casper, S., Sourbut, O., et al. (2024) · arXiv (Cornell University) · doi:10.3929/ethz-b-000706962 · Cited by 17
A scoping review of ChatGPT research in accounting and finance
Dong, M., Stratopoulos, T. C., & Wang, V. X. (2024) · International Journal of Accounting Information Systems · doi:10.1016/j.accinf.2024.100715 · Cited by 239
The power of generative marketing: Can generative AI create superhuman visual marketing content?
Hartmann, J., Exner, Y., & Domdey, S. (2025) · International Journal of Research in Marketing · doi:10.1016/j.ijresmar.2024.09.002 · Cited by 229
The Uneven Impact of Generative AI on Entrepreneurial PerformancePreprint
Otis, N. G., Clarke, R. P., Delecourt, S., Holtz, D., & Koning, R. M. (2023) · OSF Preprints (OSF Preprints) · doi:10.31219/osf.io/hdjpk · Cited by 206
Artificial Intelligence and the Transformation of Higher Education Institutions: A Systems Approach
Katsamakas, E., Pavlov, O. V., & Saklad, R. (2024) · Sustainability · doi:10.3390/su16146118 · Cited by 195
Metadata
- Authors
- Hui, X., Reshef, O., & Zhou, L.
- Affiliations
- Washington University in St. Louis (US)
- New York University (US)
- Corresponding author
- Hui, Xiang
- Venue
- SSRN Electronic Journal
- Publisher
- Elsevier BV
- Published
- 2023
- ISSN
- 1556-5068 (ISSN-L)
- OpenAlex
- W4385450019
- Semantic Scholar
- 47bef342b9e1204a1360ef087556bd1243c65e59
- Cited by
- 116
- References
- 33
- Citations by year
- 2026: 21 · 2025: 43 · 2024: 43 · 2023: 8
- Type
- posted content
- Language
- en
- Open access
- green
- hdl.handle.net
- Locations
- SSRN Electronic Journal (submittedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Econstor (Econstor) (submittedVersion)
- Topics
- Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Labor market dynamics and wage inequality
- Social Sciences › Sociology and Political Science › Digital Economy and Work Transformation
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Keywords
- generative AI, ChatGPT, large language models, freelancers, digital labour markets, employment effects, earnings, knowledge workers, generative image models, labor demand, wage inequality
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-11-17, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-17 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.