arxiv: 2604.18849 · v3 · submitted 2026-04-20 · 💰 econ.GN · q-fin.EC

Recognition: unknown

From Exposure to Adoption: Generative AI in European Workplaces

Golo Henseke

Authors on Pith no claims yet

Pith reviewed 2026-05-10 02:47 UTC · model grok-4.3

classification 💰 econ.GN q-fin.EC

keywords generative AIadoptiontask contentEuropean labor marketsshift-share designoccupational exposuregender gapdigitalization

0 comments

The pith

Generative AI adoption in Europe tracks occupational exposure plus skills and digital infrastructure, yet produces no measurable shift in job tasks.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper maps the early spread of generative AI tools across workers in 35 European countries and tests whether adoption has begun to alter the content of their jobs. Adoption varies sharply by occupation and country, rising where exposure is high and where workers possess abstract-task skills, organizational influence, or access to digital training. A persistent gender gap appears in the most exposed roles. A shift-share design that exploits differences in occupational exposure finds no detectable link between adoption and worker-reported changes in task composition, pointing to an initial phase of integration rather than immediate restructuring.

Core claim

The central finding is that occupational exposure to generative AI strongly predicts uptake, yet this relationship is sharpened at the worker level by skills, abstract task content, and employee influence, and at the country level by digitalization and workplace training. Despite these gradients and a concentrated gender gap in high-exposure occupations, a shift-share instrumental variable approach detects no causal effect of adoption on self-reported task restructuring, which the author interprets as evidence of an early integration stage in which tools have entered workplaces without yet prompting reorganization of work.

What carries the argument

Shift-share design instrumenting individual adoption with occupational exposure to generative AI, used to estimate effects on self-reported task content.

If this is right

Adoption rates range from under 3 percent to 25 percent, highest where exposure meets supporting skills and infrastructure.
Worker abstract-task content and organizational influence increase the probability of adoption conditional on exposure.
Country digitalization and training programs widen the exposure-adoption gap.
No detectable restructuring of tasks has occurred among early adopters.
A gender gap in adoption persists and is largest in the most exposed occupations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If task content remains stable, productivity gains from generative AI may require complementary changes in work organization that have not yet occurred.
The gender gap concentrated in exposed roles implies that closing it could expand the pool of AI users without altering exposure patterns.
Later stages of diffusion might produce observable task shifts once integration moves beyond initial experimentation.
Cross-country differences in digital infrastructure suggest that investments there could accelerate both adoption and eventual task adaptation.

Load-bearing premise

The shift-share design and self-reported task measures accurately isolate causal effects of adoption without confounding from unobserved worker or firm changes.

What would settle it

Panel data on the same workers showing statistically significant changes in task time allocation or descriptions after documented adoption would contradict the no-effect result.

Figures

Figures reproduced from arXiv: 2604.18849 by Golo Henseke.

**Figure 3.** Figure 3: AI adoption by levels of occupational AI susceptibility. [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗

**Figure 4.** Figure 4: Country AI adoption level and the exposure-adoption slope. [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗

**Figure 5.** Figure 5: Country-level partial correlates of the exposure–adoption slope, conditional on digital [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗

**Figure 6.** Figure 6: Job task restructuring and AI adoption across occupations. [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗

read the original abstract

This study examines who adopts generative AI and whether early adoption has begun to reshape the task content of jobs across 35 European countries. Adoption ranges from under 3% to 25%. Occupational exposure strongly predicts uptake, but AI does not diffuse passively along exposure lines. At the worker level, skills, abstract task content, and employee organisational influence steepen the exposure-adoption gradient; at the country level, so do digitalisation and workplace training. A gender gap persists, concentrated in the most exposed occupations. A shift-share design finds no detectable effect of adoption on worker-reported task restructuring, consistent with an initial integration phase.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Descriptive cross-country map of gen AI adoption in Europe with exposure gradients and a null on task changes, but identification stays preliminary.

read the letter

The main thing to know is that this paper gives a broad descriptive picture of generative AI uptake across 35 European countries, with adoption running from under 3% to 25%, occupational exposure as a strong but incomplete predictor, and a shift-share result showing no detectable impact on worker-reported task restructuring yet. It layers in worker-level factors like skills and organizational influence plus country-level digitalization and training, and flags a gender gap in the most exposed jobs. That combination at this scale is new enough to be useful right now. The null on tasks fits the story of an early integration phase and supplies a benchmark other studies can build on. The paper does a clean job of separating exposure from actual adoption and showing how individual and institutional factors shape the gap between the two. The shift-share approach is a straightforward way to probe causality on the task side without overclaiming. The soft spots are measurement and identification. Everything rests on self-reports for adoption and task content, which can be noisy or slow to reflect real changes, and the shift-share needs the exposure measure to satisfy the usual exclusion restrictions. If unobserved firm or worker shifts are correlated with adoption, the null could be fragile. The abstract is thin on sample sizes, exact variable definitions, and robustness tables, so the strength of the evidence is still hard to judge fully even with the full text. This is for labor economists and policy readers who need current facts on AI diffusion in Europe rather than a finished causal story. A reader tracking workforce impacts will get value from the patterns and the scale. It deserves peer review because the questions are timely and the cross-country scope adds incremental evidence, even if the design will need tightening on identification and checks.

Referee Report

0 major / 3 minor

Summary. This paper examines generative AI adoption across 35 European countries and tests whether early adoption has begun to reshape job task content. Adoption rates range from under 3% to 25%. Occupational exposure strongly predicts uptake, but diffusion is heterogeneous: at the worker level, skills, abstract task content, and organizational influence moderate the exposure-adoption link; at the country level, digitalization and workplace training do likewise. A gender gap is concentrated in high-exposure occupations. A shift-share design yields no detectable effect of adoption on worker-reported task restructuring, interpreted as consistent with an initial integration phase.

Significance. If the null result is robust, the paper supplies timely cross-country evidence on the early diffusion of generative AI and its limited immediate effects on task content. The descriptive gradients on who adopts add value for understanding barriers to diffusion, while the design-based approach to the null finding on restructuring is a constructive contribution to labor economics work on new technologies. Reproducible patterns from large-scale European survey data strengthen the descriptive component.

minor comments (3)

Abstract: report sample sizes, exact data source (survey name and year), and brief definitions of key variables such as 'task restructuring' and the exposure measure to allow immediate assessment of the reported adoption rates and null result.
Methods section: provide the precise shift-share instrument construction, the first-stage strength, and any robustness checks (e.g., alternative exposure measures or placebo tests) so readers can evaluate the exclusion restriction for the null effect on task restructuring.
Results: clarify whether the gender gap analysis interacts exposure with gender or is purely descriptive, and report effect sizes or confidence intervals alongside the 'no detectable effect' statement.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive and accurate summary of our paper, the assessment of its significance for labor economics, and the recommendation for minor revision. The referee's description correctly captures the range of adoption rates, the role of occupational exposure, the moderating factors at worker and country levels, the gender gap, and the null finding on task restructuring from the shift-share design.

Circularity Check

0 steps flagged

No circularity: empirical shift-share analysis with no self-referential derivations

full rationale

The paper is a descriptive and design-based empirical study using survey data across European countries. It reports adoption rates, correlates of uptake (occupational exposure, skills, etc.), and applies a shift-share design to test for effects on task restructuring. No equations, fitted parameters, or derivations are presented that reduce the central null finding to a tautology or to the inputs by construction. The shift-share approach is a standard econometric tool whose validity rests on external identifying assumptions rather than internal redefinition. No self-citation chains, ansatzes smuggled via prior work, or renaming of known results appear in the provided abstract or described methods. The analysis is self-contained against external benchmarks and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No formal model or derivation; the analysis rests on standard survey data assumptions and the validity of the shift-share identification strategy.

pith-pipeline@v0.9.0 · 5392 in / 1021 out tokens · 44219 ms · 2026-05-10T02:47:16.981259+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

59 extracted references · 38 canonical work pages · 1 internal anchor

[1]

2025 , author =

What Drives AI and Robot Adoption? , subtitle =. 2025 , author =

2025
[2]

Digital Monitoring, Algorithmic Management and the Platformisation of Work in Europe , publisher =

Gonz. Digital Monitoring, Algorithmic Management and the Platformisation of Work in Europe , publisher =. 2025 , doi =

2025
[3]

, title =

Bick, Alexander and Blandin, Adam and Deming, David J. , title =. 2024 , number =

2024
[4]

2026 , journal =

Liu, Yan and Wang, He , pages =. 2026 , journal =. doi:10.1016/j.worlddev.2025.107260 , issn =

work page doi:10.1016/j.worlddev.2025.107260 2026
[5]

Proceedings of the National Academy of Sciences of the United States of America , volume =

Humlum, Anders and Vestergaard, Emilie , title =. Proceedings of the National Academy of Sciences of the United States of America , volume =. 2025 , doi =

2025
[6]

doi:10.3386/w34995 , keywords =

2026 , author =. doi:10.3386/w34995 , keywords =

work page doi:10.3386/w34995 2026
[7]

GPTs Are GPTs: Labor Market Impact Potential of LLMs

Eloundou, Tyna and Manning, Sam and Mishkin, Pamela and Rock, Daniel , volume =. 2024 , journal =. doi:10.1126/science.adj0998 , url =

work page doi:10.1126/science.adj0998 2024
[8]

The Quarterly Journal of Economics , author =

Generative AI at Work , year =. The Quarterly Journal of Economics , author =. doi:10.1093/qje/qjae044 , url =

work page doi:10.1093/qje/qjae044
[9]

Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI , year =

Humlum, Anders and Vestergaard, Emilie , number =. Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI , year =. doi:10.3386/w33777 , url =

work page doi:10.3386/w33777
[10]

and Cranney, Julie and Delecourt, Sol

Otis, Nicholas G. and Cranney, Julie and Delecourt, Sol. The Gender Gap in Generative AI Use , year =
[11]

Economic Policy , author =

The Simple Macroeconomics of AI , year =. Economic Policy , author =. doi:10.1093/epolic/eiae042 , url =

work page doi:10.1093/epolic/eiae042
[12]

Artificial Intelligence and Jobs: Evidence from Online Vacancies,

Artificial Intelligence and Jobs: Evidence from Online Vacancies , year =. Journal of Labor Economics , author =. doi:10.1086/718327 , url =

work page doi:10.1086/718327
[13]

Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives , year =

Baslandze, Salom. Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives , year =
[14]

doi:10.2139/ssrn.6120067 , institution =

2026 , author =. doi:10.2139/ssrn.6120067 , institution =

work page doi:10.2139/ssrn.6120067 2026
[15]

Generative ai practices, literacy, and divides: An empirical analysis in the italian context,

2025 , author =. 2512.03671 , archivePrefix =

work page arXiv 2025
[16]

How Exposed Are UK Jobs to Generative AI? Developing and Applying a Novel Task-Based Index

Henseke, Golo and Davies, Rhys and Felstead, Alan and Gallie, Duncan and Green, Francis and Zhou, Ying , doi =. How Exposed Are UK Jobs to Generative AI? Developing and Applying a Novel Task-Based Index , year =. 2507.22748 , archiveprefix =

work page internal anchor Pith review Pith/arXiv arXiv
[17]

Gallie, Duncan and Zhou, Ying , title =
[18]

Journal of Economic Perspectives , author =

A Practical Guide to Shift-Share Instruments , year =. Journal of Economic Perspectives , author =. doi:10.1257/JEP.20231370 , issn =

work page doi:10.1257/jep.20231370
[19]

doi:10.3386/w33509 , keywords =

2025 , author =. doi:10.3386/w33509 , keywords =

work page doi:10.3386/w33509 2025
[20]

2020 , journal =

Goldsmith-Pinkham, Paul and Sorkin, Isaac and Swift, Henry , number =. 2020 , journal =. doi:10.1257/AER.20181047 , keywords =

work page doi:10.1257/aer.20181047 2020
[21]

2026 , doi =

The Use of Artificial Intelligence (AI) Technologies in the European Union: Key Results, 2026 Edition , publisher =. 2026 , doi =

2026
[22]

and Brynjolfsson, Erik and Hitt, Lorin M

Bresnahan, Timothy F. and Brynjolfsson, Erik and Hitt, Lorin M. , number =. 2002 , journal =. doi:10.1162/003355302753399526 , issn =

work page doi:10.1162/003355302753399526 2002
[23]

2021 , journal =

Brynjolfsson, Erik and Rock, Daniel and Syverson, Chad , number =. 2021 , journal =. doi:10.1257/MAC.20180386 , issn =

work page doi:10.1257/mac.20180386 2021
[24]

Frank and Brynjolfsson, Erik and Kroff, Zachary and Dinlersoz, Emin and Foster, Lucia and Zolas, Nikolas , number =

McElheran, Kristina and Li, J. Frank and Brynjolfsson, Erik and Kroff, Zachary and Dinlersoz, Emin and Foster, Lucia and Zolas, Nikolas , number =. 2024 , journal =

2024
[25]

doi:10.2139/ssrn.4414065 , keywords =

2023 , author =. doi:10.2139/ssrn.4414065 , keywords =

work page doi:10.2139/ssrn.4414065 2023
[26]

doi:10.2139/ssrn.4739904 , institution =

2024 , author =. doi:10.2139/ssrn.4739904 , institution =

work page doi:10.2139/ssrn.4739904 2024
[27]

Journal of Political Economy , author =

Robots and Jobs: Evidence from US Labor Markets , year =. Journal of Political Economy , author =. doi:10.1086/705716 , issn =

work page doi:10.1086/705716
[28]

Science 381(6654): 187–92

Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence , year =. Science , author =. doi:10.1126/science.adh2586 , issn =

work page doi:10.1126/science.adh2586
[29]

ILR Review , author =

The Impact of Technological Change on Older Workers: Evidence from Data on Computer Use , year =. ILR Review , author =. doi:10.1177/001979390305600309 , issn =

work page doi:10.1177/001979390305600309
[30]

OECD Skills Outlook 2013: First Results from the Survey of Adult Skills , year =

2013
[31]

2009 , journal =

Autor, David and Dorn, David , number =. 2009 , journal =. doi:10.1257/AER.99.2.45 , issn =

work page doi:10.1257/aer.99.2.45 2009
[32]

AEA Papers and Proceedings , author =

What Can Machines Learn and What Does It Mean for Occupations and the Economy? , year =. AEA Papers and Proceedings , author =. doi:10.1257/pandp.20181019 , url =

work page doi:10.1257/pandp.20181019
[33]

ILR Review , author =

Employee Involvement, Technology and Evolution in Job Skills: A Task-Based Analysis , year =. ILR Review , author =. doi:10.1177/001979391206500103 , issn =

work page doi:10.1177/001979391206500103
[34]

2010 , journal =

Felstead, Alan and Gallie, Duncan and Green, Francis and Zhou, Ying , number =. 2010 , journal =. doi:10.1080/09585192.2010.500489 , issn =

work page doi:10.1080/09585192.2010.500489 2010
[35]

, editor =

Appelbaum, Eileen and Bailey, Thomas and Berg, Peter and Kalleberg, Arne L. , editor =. Do high-performance work systems pay off? , booktitle =. 2001 , doi =

2001
[36]

Comin, B

Comin, D. and Hobijn, Bart , number =. 2004 , journal =. doi:10.1016/j.jmoneco.2003.07.003 , keywords =

work page doi:10.1016/j.jmoneco.2003.07.003 2004
[37]

Comin, B

Comin, Diego and Hobijn, Bart , number =. 2010 , journal =. doi:10.1257/AER.100.5.2031 , keywords =

work page doi:10.1257/aer.100.5.2031 2010
[38]

and Dughera, S

Brunetti, I. and Dughera, S. and Ricci, A. , pages =. 2025 , journal =. doi:10.1080/10438599.2025.2573062 , keywords =

work page doi:10.1080/10438599.2025.2573062 2025
[39]

2004 , journal =

Keller, Wolfgang , number =. 2004 , journal =. doi:10.1257/0022051042177685 , keywords =

work page doi:10.1257/0022051042177685 2004
[40]

A Framework for Assessing AI Exposure and Output Quality Across UK Occupations and Regions , year =

Ransom, James and Whittle, Richard , doi =. A Framework for Assessing AI Exposure and Output Quality Across UK Occupations and Regions , year =
[41]

Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality

Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality , year =. Organization Science , author =. doi:10.1287/orsc.2025.21838 , keywords =

work page doi:10.1287/orsc.2025.21838 2025
[42]

2025 , journal =

Teutloff, Ole and Einsiedler, Johanna and K. 2025 , journal =. doi:10.1016/j.jebo.2024.106845 , keywords =

work page doi:10.1016/j.jebo.2024.106845 2025
[43]

Management Science , author =

Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms , year =. Management Science , author =. doi:10.1287/mnsc.2024.05420 , keywords =

work page doi:10.1287/mnsc.2024.05420 2024
[44]

Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence , year =

Brynjolfsson, Erik and Chandar, Bharat and Chen, Ruyu , url =. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence , year =
[45]

2025 , doi =

Generative AI and Labor Market Outcomes: Evidence from the United Kingdom , author =. 2025 , doi =

2025
[46]

2025 , doi =

Shifting Work Patterns with Generative AI , author =. 2025 , doi =

2025
[47]

European Working Conditions Survey, 2024 , year =

2024
[48]

2024 , journal =

Aldasoro, Iñaki and Armantier, Olivier and Doerr, Sebastian and Gambacorta, Leonardo and Oliviero, Tommaso , pages =. 2024 , journal =. doi:10.1016/j.econlet.2024.111814 , keywords =

work page doi:10.1016/j.econlet.2024.111814 2024
[49]

and Inklaar, Robert and Timmer, Marcel P

Feenstra, Robert C. and Inklaar, Robert and Timmer, Marcel P. , title =. American Economic Review , year =
[50]

Artificial Intelligence for Europe , year =
[51]

doi:10.2806/8618665 , note =

European Working Conditions Survey 2024: Overview report , year =. doi:10.2806/8618665 , note =

work page doi:10.2806/8618665 2024
[52]

Handa, A

Handa, Kunal and Tamkin, Alex and McCain, Miles and Huang, Saffron and Durmus, Esin and Heck, Sarah and Mueller, Jared and Hong, Jerry and Ritchie, Stuart and Belonax, Tim and Troy, Kevin K. and Amodei, Dario and Kaplan, Jared and Clark, Jack and Ganguli, Deep , doi =. Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversati...

work page arXiv
[53]

doi:10.1787/3639d1e2-en , url =

Survey of Adult Skills -- Reader's Companion: 2023 , year =. doi:10.1787/3639d1e2-en , url =

work page doi:10.1787/3639d1e2-en 2023
[54]

2022 , journal =

Holm, Jacob Rubæk and Lorenz, Edward , number =. 2022 , journal =. doi:10.1111/NTWE.12215 , keywords =

work page doi:10.1111/ntwe.12215 2022
[55]

2026 , booktitle =

Murgia, Madhumita and Burn-Murdoch, John , url =. 2026 , booktitle =

2026
[56]

, number =

Berg, Janine and Green, Francis and Nurski, Laura and Spencer, David A. , number =. 2023 , journal =. doi:10.1177/09596801231178904 , issn =

work page doi:10.1177/09596801231178904 2023
[57]

2025 , journal =

Cnossen, Femke , number =. 2025 , journal =

2025
[58]

and Prettner, Klaus and Saadaoui, Jamel and Veruete, Mario , pages =

Bloom, David E. and Prettner, Klaus and Saadaoui, Jamel and Veruete, Mario , pages =. 2025 , journal =. doi:10.1016/J.FRL.2025.107401 , issn =

work page doi:10.1016/j.frl.2025.107401 2025
[59]

Industrial Relations: A Journal of Economy and Society , year =

Lewandowski, Piotr and Szymczak, Wojciech , title =. Industrial Relations: A Journal of Economy and Society , year =