Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 25 of the 25 references it cites were found in a registry or an index; none of the 21 sources its claims cite is checked for support yet.

Harder or Easier? Impact of the AI Exposure Index on China’s Online Labor Market: An Occupation-Based Perspective

Open access
Shi, Hao, Wang, Tianmei · 2026 · Proceedings of the Annual Hawaii International Conference on System Sciences

Abstract

This study explores the impact of artificial intelligence (AI) on the labor market from the perspective of occupations. We innovatively construct an AI exposure index based on local Chinese data by measuring the semantic similarity between AI capabilities and occupational task requirements using large language model and Sentence-BERT models. Using the retrieval-augmented generation method, we analyze online job postings in China from 2019 to 2024, examining the impact of AI exposure on employment opportunities, occupational wages, and job matching. Results show that overall, AI exposure has steadily increased across tasks and occupations in China. AI exposure shows significant substitution effect on traditional occupations but no significant creation effect on emerging ones. High AI exposure is associated with low wages, especially in occupations with high proportion of routine tasks, although task complexity mitigates this effect. In addition, AI exposure reduces job filling rates and prolongs recruitment time, indicating increased matching friction.

Overview

CoChat Summary

TL;DR

The paper builds a localized Chinese AI occupational exposure (AIOE) index from AI patent texts and occupational tasks, then links it to 2019–2024 online job postings. It finds AI exposure is associated with lower wages and lower fill rates, no significant net employment effect, substitution in traditional occupations, and limited creation in emerging ones.

Key findings

  • AI exposure rose steadily across tasks and occupations in China from 2019 to 2024; routine tasks are more exposed, while non-routine cognitive tasks have risen significantly since 2021 with generative AI. At the occupational level, categories 4 and 6 show the highest exposure, categories 2 and 3 lower exposure, and category 1 the lowest.
  • Baseline two-way fixed-effects regressions show lagged AIOE has a significant negative effect on log wages (−4.251*, SE 1.036) and fill rate (−4.050*, SE 1.201), but no significant effect on employment opportunity (−0.203, SE 0.127); N = 3,953, 3,953, and 3,248; R² = 0.110, 0.015, and 0.214. All proposed hypotheses are supported.
  • Heterogeneity: emerging digital occupations (97 newly added digital occupations, about 6% of all occupations) have average AIOE 0.558 versus 0.412 for traditional occupations. AIOE has a significant negative effect on employment opportunities for traditional occupations (−0.328, SE 0.166) but no significant effect for emerging occupations (0.209, SE 0.280).
  • Task structure and matching: AIOE × routine task share has a significant negative effect on wages (−4.445, SE 2.133), while AIOE × task complexity has a significant positive effect (4.848, SE 1.064). AIOE also significantly increases log hiring duration (0.118*, SE 0.056; N = 3,248; R² = 0.312).

Methods

  • AIOE construction: Chinese AI patent texts were processed with an LLM to extract core technical capability sentences; Sentence-BERT (all-mpnet-base-v2) encoded 9,985 occupational tasks and approximately 400,000 AI patent core technology sentences; cosine similarity produced task-level exposure, aggregated to occupations using the Classification of Occupations of the People’s Republic of China (2022). The final dataset includes 1,612 occupations and 9,985 tasks, and 430,175 AI patents.
  • Online labor data: web crawlers collected job postings from listed companies in China from 2019 to 2024 on platforms including 51job, BOSS, and Liepin. After cleaning an initial 4 million records, retrieval-augmented generation (RAG) matched job titles and descriptions to 1,639 seven-digit occupational subcategories; human annotators evaluated 5,000 postings, and overall system accuracy was 72%. Final analysis used 2,151,660 recruitment records covering 962 occupations.
  • Empirical analysis: occupation-year two-way fixed-effects regressions with occupation and year fixed effects; outcomes were normalized employment opportunity, log average wage, and fill rate (completed positions / total positions). AIOE and controls (education and experience requirements) were lagged one period. Robustness checks used gerund-pair matching with O*NET Detailed Work Activities (77% of DWAs produced gerund pairs; 660 matched to patent data), occupational classification merging, and an instrumental variable based on AI research intensity from HowNet publications and media coverage. Further analyses examined emerging vs. traditional occupations, routine task share, entropy-based task complexity, and hiring duration.

Limitations

  • Stated: the paper lacks a clear method for delineating generative AI patents; it references WIPO’s five core generative AI models and does not elaborate due to space constraints, leaving differentiated generative vs. traditional AI research for future work. Recruitment end dates for 2024 were unavailable, so 2024 was excluded from fill-rate analysis. The fill rate reflects completion of job search rather than matching quality or efficiency, so hiring duration was added to capture matching friction.
  • Stated/evident: AIOE is based on text-based semantic similarity and captures potential rather than realized effects. Low AIOE for managers and agricultural workers may partly reflect measurement limitations: vague task descriptions for managerial roles, and agricultural tasks depending on natural conditions, manual labor, and experiential judgment that are difficult to capture through text-based semantic similarity.
  • Evident from the text: RAG matching accuracy was 72%, and job titles on Chinese recruitment platforms are non-standard and semantically ambiguous, so occupational classification error is possible. The online recruitment data have industry concentration and occupational bias and cover 962 occupations, so they may not represent the entire labor market. The study period is short-term (2019–2024), and the AIOE cannot directly distinguish substitution from creation effects.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.24251/hicss.2026.023

Cites

25 of 25 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 Valid · 2 Not valid · 16 Not checked · 82 Own results and interpretations

38 more claims

Create a free account to see all 100 claims and to verify the 18 references still open.

See all claims — free account
  • It may state two separate claims as one. 30
  • Probably not a substantial claim. 40

Create a free account to see these claims.

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.

1 of 1

Citations

Cites

Show all 25 references

Research works printed but not linked: 1

  1. Restrepo, P., & Rozing, M. (2021). Tasks, automation, and the rise in U.S. wage inequality. American Economic Review, 111(5), 1737-1774.

Cited by

Still being checked.

Metadata

Authors
Shi, H., & Wang, T.
Affiliations
Central University of Finance and Economics (CN)
Venue
Proceedings of the Annual Hawaii International Conference on System Sciences
Published
2026
ISSN
  • 1530-1605 (ISSN-L)
  • 2572-6862
OpenAlex
W7154389855
Cited by
0
References
25
FWCI
0.00
Type
conference paper
Language
en
Open access
Licence
cc-by-nc-nd
Locations
  • Hawaii International Conference on System Sciences (HICSS) (publishedVersion, cc-by-nc-nd)
  • ScholarSpace (University of Hawaii at Manoa) (submittedVersion, cc-by-nc-nd)
Funders
National Natural Science Foundation of China (72072194)
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
  • Social Sciences › Business, Management and Accounting › Industrial relations › Legal and Labor Studies
Keywords
artificial intelligence exposure, online job postings, semantic similarity, large language models, Sentence-BERT, recruitment time, occupational wages, routine tasks, task complexity, China

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Harder or Easier? Impact of the AI Exposure Index on China’s Online Labor Market: An Occupation-Based Perspective. https://app.cochat.ai/papers/10.24251/hicss.2026.023
The paper
Shi, H., & Wang, T. (2026). Harder or Easier? Impact of the AI Exposure Index on China’s Online Labor Market: An Occupation-Based Perspective. Proceedings of the Annual Hawaii International Conference on System Sciences. https://doi.org/10.24251/hicss.2026.023

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.