Women in Sweden are concentrated in occupations predicted to be most exposed to generative AI, creating a risk that ChatGPT-style tools could widen the gender pay gap. Simulations based on exposure scores and assumed task complementarities suggest increased wage disparity, though the magnitude depends heavily on adoption patterns and modelling assumptions.

<scp>Pre‐AI</scp> Sorting, <scp>Post‐AI</scp> Inequality: Generative <scp>AI</scp> and the Gender Wage Gap

Malin Gardberg, F. Heyman, M. Olsson, Joacim Tåg · Fetched May 25, 2026 · Oxford Bulletin of Economics and Statistics

semantic_scholar correlational low evidence 7/10 relevance DOI Source

Using Swedish administrative data, the paper documents that women are overrepresented in occupations predicted to be highly exposed to generative AI and shows via mechanical simulations that, under plausible assumptions about exposure and task complementarity, generative AI could widen the gender wage gap.

We examine how gender‐based occupational sorting before the release of ChatGPT relates to predicted exposure to generative AI and its potential implications for the gender wage gap. Using Swedish administrative data, we document that women are overrepresented in occupations predicted to be more affected by generative AI. Mechanical partial‐equilibrium simulations, based on hypothesized deviations from the 2021 occupational and wage distribution and incorporating predicted AI exposure and task complementarity, indicate that generative AI may widen the gender wage gap through existing patterns of gender‐based occupational sorting.

Summary

Main Finding

Women in Sweden were disproportionately concentrated in occupations predicted to be more affected by generative AI (pre-ChatGPT). Mechanical partial‑equilibrium simulations that combine those exposure patterns with assumed task complementarity suggest generative AI could widen the gender wage gap via existing gender‑based occupational sorting.

Key Points

Pre‑ChatGPT occupational sorting: women are overrepresented in occupations with higher predicted exposure to generative AI.
Predicted exposure is occupation‑level and task‑based (i.e., measures of how much generative AI can affect the tasks typical for each occupation).
Simulations incorporate hypothesized deviations from the 2021 occupational and wage distribution and include assumptions about task complementarity (how AI affects productivity and wages depending on whether tasks complement or substitute human labor).
Under plausible combinations of exposure and complementarity assumptions, generative AI increases wage pressures more in occupations with higher female share, leading to a widening of the gender wage gap.
Results are from mechanical, partial‑equilibrium exercises — they do not model broader general equilibrium adjustments, labor reallocation, or policy/behavioral responses.

Data & Methods

Data: Swedish administrative data on employment, occupations, and wages (baseline occupational and wage distribution from 2021).
Exposure measure: occupation‑level predicted exposure to generative AI based on task content (constructed from external task‑AI mapping; specifics not provided in the summary).
Empirical finding: descriptive evidence of gender concentration across exposure categories showing female overrepresentation in high‑exposure occupations.
Simulations:
- Type: mechanical partial‑equilibrium simulations.
- Inputs: 2021 occupational and wage distribution, occupation AI exposure scores, assumed productivity/wage impacts by exposure and complementarity.
- Scenarios: counterfactual/hypothesized deviations to reflect varying degrees of AI impact and complementarity between AI and worker tasks.
- Outcome: simulated changes in occupation wages and overall gender wage gap under each scenario.
Limitations of methods: no dynamic reallocation of labor across occupations, no firm‑level heterogeneity in adoption, and reliance on predicted exposure and complementarity assumptions rather than observed post‑adoption outcomes.

Implications for AI Economics

Distributional risk: Generative AI can have distributional effects that depend on preexisting occupational segregation; gendered occupational structures matter for inequality outcomes.
Policy relevance: To avoid widening gender wage gaps, policies could prioritize:
- Targeted reskilling and upskilling for workers (especially women) in high‑exposure occupations.
- Support for transitions into occupations/tasks complementary to AI.
- Monitoring of AI adoption and wage outcomes by gender and occupation.
- Labor market interventions (bargaining support, wage subsidies, redistribution) if unequal impacts materialize.
Research priorities:
- Estimate actual (post‑adoption) impacts of generative AI on wages and employment by occupation and gender.
- Model general equilibrium effects and labor reallocation dynamics.
- Study firm heterogeneity in adoption and complementary investments that influence who benefits from AI.
- Cross‑country comparisons to evaluate how labor institutions and occupational structures mediate effects.
Caution: Mechanical simulations indicate risk but do not prove realized outcomes—actual effects will depend on how AI is adopted, which tasks are automated vs. augmented, and policy and behavioral responses.

Assessment

Paper Typecorrelational Evidence Strengthlow — The paper uses high-quality administrative data to document occupational sorting but does not provide causal identification of generative AI's effects; the counterfactual impact on wages is inferred from mechanical partial-equilibrium simulations that rely on predicted exposure scores and strong assumptions (task complementarity, uptake patterns), so empirical support for the central causal claim is limited. Methods Rigormedium — Empirical description uses comprehensive Swedish administrative registers (strong data provenance) and the simulation approach is transparent, but the exposure measures come from predictive mappings and the simulations omit equilibrium dynamics, firm responses, and potential measurement error, limiting internal validity. SampleSwedish administrative labor-market registers describing occupations and wages prior to widespread ChatGPT deployment (baseline occupational and wage distribution circa 2021); population-level or large sample of workers across occupations (national coverage typical of Swedish admin data). Themesinequality labor_markets GeneralizabilityFindings are Sweden-specific—high unionization, social safety nets, and sectoral composition limit transferability to other countries, Predicted AI exposure scores may not reflect real-world adoption, task change, or firm-level integration elsewhere, Partial-equilibrium simulations ignore general-equilibrium adjustments, firm reallocation, and labor supply responses, Results depend on assumed task complementarity parameters and counterfactual wage/occupational shifts that may not hold, Analysis uses pre-ChatGPT occupational structure and may not capture post-adoption re-skilling, redeployment, or emergent occupations

Claims (3)

Claim	Direction	Outcome	Confidence & Evidence	Details
Women are overrepresented in occupations predicted to be more affected by generative AI (using pre-ChatGPT occupational sorting). Automation Exposure	negative	predicted exposure to generative AI by occupation / gender representation in high-exposure occupations	Reading fidelity high Study strength medium	0.3
Mechanical partial-equilibrium simulations indicate that generative AI may widen the gender wage gap. Wages	negative	gender wage gap (changes in wages by gender)	Reading fidelity high Study strength speculative	0.05
The potential widening of the gender wage gap would operate through existing patterns of gender-based occupational sorting (i.e., because women are concentrated in occupations more exposed to generative AI). Wages	negative	mechanism linking occupational sorting to changes in gender wage gap	Reading fidelity high Study strength speculative	0.05