Evidence (215 claims)
Search and filter individual claims pulled from the papers. Looking for a specific finding ("what's the effect on wages?"), you're in the right place. Want to compare whole outcome categories against each other instead? Use the Evidence Explorer.
The board below groups claims two ways: by broad theme (nine paper-level topics) and by outcome category (the 34 claim-level outcomes that the Explorer and Syntheses also use).
Browse by theme
Nine broad, paper-level topics. Click one to filter the claims below.
Adoption
21267 claims
Filter claims →
Productivity
17978 claims
Filter claims →
Governance
17038 claims
Filter claims →
Human-AI Collaboration
16914 claims
Filter claims →
Org Design
11104 claims
Filter claims →
Innovation
11087 claims
Filter claims →
Labor Markets
6711 claims
Filter claims →
Skills & Training
5616 claims
Filter claims →
Inequality
4343 claims
Filter claims →
Claims by outcome category
Counts by direction of finding. These are the same 34 outcome categories the Explorer compares and the Syntheses are written for. A linked row has a published synthesis.
| Outcome | Positive | Negative | Mixed | Null | Total |
|---|---|---|---|---|---|
| Other | 1880 | 496 | 296 | 1854 | 4721 |
| Organizational Efficiency | 2906 | 665 | 438 | 180 | 4210 |
| Governance & Regulation | 2162 | 929 | 480 | 247 | 3866 |
| Technology Adoption Rate | 1533 | 545 | 278 | 210 | 2593 |
| Decision Quality | 1391 | 534 | 321 | 173 | 2429 |
| Output Quality | 1298 | 472 | 231 | 145 | 2153 |
| AI Safety & Ethics | 682 | 821 | 230 | 90 | 1837 |
| Research Productivity | 855 | 253 | 121 | 425 | 1675 |
| Firm Productivity | 1105 | 171 | 175 | 73 | 1531 |
| Task Allocation | 735 | 229 | 361 | 99 | 1433 |
| Market Structure | 457 | 461 | 251 | 47 | 1222 |
| Innovation Output | 673 | 94 | 108 | 36 | 913 |
| Task Completion Time | 499 | 118 | 43 | 38 | 702 |
| Firm Revenue | 458 | 130 | 61 | 26 | 677 |
| Skill Acquisition | 381 | 122 | 113 | 34 | 650 |
| Consumer Welfare | 316 | 176 | 115 | 39 | 648 |
| Employment Level | 223 | 143 | 177 | 53 | 600 |
| Error Rate | 246 | 282 | 44 | 19 | 594 |
| Fiscal & Macroeconomic | 283 | 142 | 78 | 52 | 562 |
| Inequality Measures | 103 | 329 | 106 | 13 | 552 |
| Worker Satisfaction | 225 | 185 | 63 | 30 | 503 |
| Automation Exposure | 158 | 155 | 72 | 37 | 426 |
| Regulatory Compliance | 186 | 126 | 35 | 14 | 362 |
| Team Performance | 193 | 56 | 51 | 24 | 326 |
| Developer Productivity | 224 | 58 | 27 | 13 | 323 |
| Wages & Compensation | 148 | 108 | 50 | 17 | 323 |
| Training Effectiveness | 218 | 44 | 21 | 27 | 313 |
| Job Displacement | 23 | 159 | 53 | 5 | 240 |
| Hiring & Recruitment | 109 | 61 | 32 | 11 | 215 |
| Skill Obsolescence | 16 | 107 | 26 | 6 | 155 |
| Creative Output | 71 | 44 | 28 | 6 | 150 |
| Social Protection | 58 | 31 | 12 | 3 | 104 |
| Labor Share of Income | 29 | 43 | 25 | 2 | 99 |
| Worker Turnover | 45 | 29 | 6 | 4 | 84 |
| Industry | — | — | — | 1 | 1 |
AI-related hiring demand is unevenly distributed across urban labor markets and across occupations within cities.
Descriptive analysis of online job-posting data from Chinese listed firms, covering cities and occupations over 2016–2024.
Employers assess graduate value by comparing expected productivity with the graduate's wage and other employment costs, while also considering whether the work could be performed by lower-cost labour or AI-enabled alternatives.
Conceptual and integrative synthesis of employability, human capital, signalling, global labour arbitrage, strategic HRM, and AI-enabled knowledge-work literature; no original empirical data.
Blockchain may reduce demand for lower-skill transaction-checking labor while increasing demand for specialists in cryptography, software engineering, smart-contract auditing, oracle validation, and regulatory compliance.
Task-based labor-reallocation argument derived from the conceptual analogy between blockchain automation and narrow AI automation; no labor-market data are reported.
Adoption of multimodal AI is expected to increase demand for machine-learning systems, data-engineering, and multimodal-annotation roles while reorienting some traditional data-management work toward model-centric workflows.
Conceptual assessment of likely occupational and skill-demand changes; no labor-market data or employment estimates are reported.
Potential skill shortages vary substantially across occupations and countries.
The study applies its shortage measure across occupations and countries using worker skill data from ESJS and vacancy requirements from Lightcast job advertisements.
Gender-related salary disparities persist across 15 of the 20 sectors examined.
Shortlisting and pipeline outcomes were stratified by salary level and sector across the operational hiring data.
Micro-credentials are more likely to be recognised for upskilling and internal mobility than as substitutes for formal degrees in initial hiring.
Cross-study synthesis of employer recognition patterns in the systematic review; the supplied text does not report a pooled effect estimate.
Employer acceptance of micro-credentials is heterogeneous: some employers view them as indicators of job-ready and adaptive skills, while many remain sceptical because of inconsistent standards, uncertain verifiability, and concerns about assessment rigour.
Systematic review synthesis of 86 included studies using thematic coding and cross-study triangulation, including employer surveys and qualitative studies.
AI-powered talent systems reduce the time and cost of talent acquisition, but algorithmic hiring systems can disadvantage candidates from underrepresented groups by reproducing historical hiring patterns.
Conceptual synthesis of HR applications and cited studies, including Tambe et al. (2019) and Raghavan et al. (2020); no original sample or effect estimate is provided.
Yapay zekâ destekli analitik sistemler ve algoritmik yönetim uygulamaları işe alım, performans izleme, zamanlama ve denetim süreçlerini dönüştürmektedir.
ILO (2025) raporu ile De Stefano ve Taes (2021) çalışmasına dayalı literatür değerlendirmesi.
Models select candidates from different demographic groups at different rates, occasionally prioritizing historically-marginalized candidates.
Empirical demographic analysis reported in the paper comparing selection rates across demographic groups on the constructed dataset; the abstract reports directional results but no quantitative selection-rate numbers.
Because the tool requires supervision from the worker's own time and attention budget, adoption is a team-production decision, similar to hiring a coworker.
Model structure and interpretation in the paper: supervision is modeled as an attention/time budget constraint, leading to adoption decisions analogous to adding a team member.
While AI holds promise for reducing overt discrimination, significant challenges remain in addressing subtle biases and ensuring AI systems do not perpetuate existing inequities.
Paper conclusion synthesizing benefits and limitations based on conceptual analysis and review of potential applications and risks; no empirical effect sizes or trials reported in the supplied text.
Recruiters' own background and AI usage significantly moderate the effect of candidate AI skills on invitation decisions.
Moderator analyses in the experiment linking recruiter characteristics (background, AI usage) to differential responses to AI skill attributes in resumes.
BLV job seekers practiced 'strategic refusal', choosing to avoid certain AI systems to regain their agency.
Empirical finding from interviews (participants described avoiding some AI-mediated hiring systems).
O desafio central reside em conciliar o uso de tecnologias de IA com a proteção dos direitos fundamentais dos candidatos, nomeadamente o direito à igualdade de oportunidades no acesso ao emprego.
Posição normativa/exortativa do texto fundamentada em princípios constitucionais e convenções internacionais mencionadas; sem evidência empírica.
Possessing GenAI skills influences the gender gap in IT job applications.
Field experiment on Upwork (stated in paper summary); empirical comparison of application behavior by gender with and without GenAI skills.
AIGC can narrow entry-level pathways while increasing demand for supervisory and pipeline roles.
Paper's analytic argument using the production-chain framework about how task automation and augmentation change hiring/demand patterns; no quantitative labor-market data given in the excerpt.
Deep RL agentic workflows affect personnel retention and recruitment, staff flexibility and autonomy, and job performance and satisfaction.
Reported associations and case discussions in the reviewed literature (2024–2025) summarized in the article.
Improvements in model-predicted success do not guarantee more referrals in the real world, but they provide low-cost signals for promising features before running higher-stakes experiments on real users.
Methodological caveat and recommendation stated in the paper: the authors note the distinction between model-predicted outcomes and real-world user behavior and suggest using these predictions as preliminary signals.
Revisions suggested by the LLM increase predicted success rates for weaker requests while reducing them for stronger requests.
Experimental comparison reported in the paper: applying LLM-generated edits to requests and evaluating pre- vs post-edit predicted referral probabilities using the evaluator model, stratified by initial request strength (weaker vs stronger).
Participant race was associated with systematic differences in selection patterns across professional domains.
Reported association in experimental data between participants' self-reported race and their selection patterns of artificial agents across the tested occupational domains.
When candidate quality is heterogeneous, prompt injection is less effective on average, but can occasionally allow lower-quality candidates to outrank higher-quality ones, raising fairness concerns.
Controlled experiments comparing homogeneous vs. heterogeneous candidate quality conditions and tracking ranking outcomes; specific experimental counts not included in the abstract.
The LLM fallacy has implications for education, hiring, and AI literacy.
Implications and argumentation presented in the paper; these are prospective and conceptual rather than supported by empirical data in the abstract.
Small differences in managerial incentives can determine which skill path a worker takes (whether they realize full potential or deskill).
Comparative statics / theoretical sensitivity analysis in the dynamic model indicating tipping behavior based on managerial incentives.
Demand for labor will shift toward data scientists, ML engineers, and interdisciplinary scientists, while wet-lab expertise and translational teams remain crucial.
Workforce trend analysis and employer hiring patterns summarized in the paper; interviews/case studies indicating changes in team composition.
AI adoption is associated with a marked decline in entry-level job postings, which traditionally serve as an entry point for new graduates.
The paper cites workforce analyses and reports this as an observed labor-market trend associated with employers adopting AI-enabled screening and productivity tools.
Historical training data can cause automated recruitment models to reproduce or amplify existing underrepresentation and discrimination, such as interpreting female demographic indicators as predictors of poor performance when historical hiring data underrepresents female engineers.
Conceptual and empirical literature synthesized by the review, including the cited discussion of historical training data and algorithmic selection.
A survey of executives found that 88% believed qualified, high-skilled candidates are screened out because their résumés do not exactly match job-description criteria.
Survey of executives conducted for the Harvard Business School–Accenture Hidden Workers project, as reported by Fuller et al. (2021). The paper cautions that this is a belief measure, not a measured error rate.
Suspicion that a freelancer used AI negatively affects the likelihood that evaluators will hire the freelancer.
Experiments measuring hiring likelihood in gender, race, and nationality conditions; Chapter 3 includes figures specifically reporting the negative effect of AI suspicion on hiring likelihood.
Software-development job postings remained approximately one-third below pre-pandemic or 2020 levels through late 2025, with junior postings declining more sharply than senior postings.
Job-posting analysis cited in the report's labor-market synthesis.
Junior recruitment in the wholesale and retail industry decreased by 40%.
The paper attributes this estimate to Harvard University research but does not report the sample size or methods.
Housekeeping was the most critical unfilled hotel role, accounting for 38% of reported staffing shortages in early 2025.
Industry staffing survey statistic reported in Table 1 and attributed to AHLA and Hireology.
In early 2025, 71% of hotels had unfilled job openings despite actively searching for workers.
Industry staffing survey statistic reported in Table 1 and attributed to AHLA and Hireology.
Native-speaker ideologies and standardized language expectations can privilege particular forms of professional communication and marginalize legitimate multilingual practices in algorithmically mediated hiring.
Literature synthesis on language ideology and algorithmic recruitment, citing Flores and Rosa (2015), Holliday (2006), Roberts (2013), and Koenecke et al. (2020).
Applicant Tracking Systems, predictive hiring algorithms, and automated résumé screening may unevenly recognize multilingual candidates' credentials and communicative practices when they privilege dominant linguistic, cultural, and institutional norms.
Integrative synthesis of scholarship on algorithmic hiring, language ideology, multilingualism, and algorithmic inequality, including Bogen and Rieke (2018), Benjamin (2019), Noble (2018), Flores and Rosa (2015), Holliday (2006), Roberts (2013), and Koenecke et al. (2020).
The study found that AI-driven applicant tracking systems can act as a structural barrier to diverse recruitment when algorithmic bias is not addressed.
Interpretive conclusion based on the reported negative associations between gender bias, ethnicity-linked ranking distortions, training-data bias, and diversity hiring outcomes.
Each of the three examined algorithmic-bias dimensions individually showed a statistically significant negative association with diversity hiring outcomes.
Three separate simple linear regressions were used to test one hypothesis per bias dimension, using responses from 243 practitioners across three major oil and gas firms in Rivers State, Nigeria.
Bias in training data has a statistically significant negative effect on diversity hiring outcomes in the studied oil and gas firms.
Cross-sectional survey of 243 HR practitioners, recruitment officers, and diversity managers; simple linear regression testing the training-data-bias hypothesis.
Ethnicity-linked ranking distortions have a statistically significant negative effect on diversity hiring outcomes in the studied oil and gas firms.
Cross-sectional survey of 243 HR practitioners, recruitment officers, and diversity managers; simple linear regression testing the ethnicity-linked ranking hypothesis.
Gender-biased screening algorithms have a statistically significant negative effect on diversity hiring outcomes in the studied oil and gas firms.
Cross-sectional survey of 243 HR practitioners, recruitment officers, and diversity managers; simple linear regression testing the gender-bias hypothesis.
Skill shortages are defined as vacancies that cannot be filled because suitably qualified or skilled candidates are lacking.
This is the study's stated operational definition of skill shortages.
Roughly 2% of vacancies in the EU are likely to experience skill shortages.
The authors combine the 2021 European Skills and Jobs Survey with Lightcast job-advertisement data, comparing vacancy skill requirements with the observed distribution of workforce skills.
If AI substantially improves the surface quality of scholarly writing, writing quality becomes a less reliable screening signal of author competence.
Conceptual signaling argument and thought experiment concerning AI-mediated improvements in prose quality.
La construction d’un « auditeur idéal » à partir de normes culturelles et non objectives contribue à reproduire des biais dans les processus organisationnels des cabinets comptables.
Synthèse de travaux sur les biais organisationnels, les normes professionnelles et la gestion des talents; le texte ne fournit pas d’estimation quantitative.
About 75% of employers struggle to find appropriate talent.
Employer-demand statistic attributed to the World Economic Forum's Future of Jobs Report (2025); no survey sample size or methodology is reported in the paper.
Candidates aged 55 and over are entirely absent from the hiring pipeline, despite representing 15.6% of Barcelona's labor force.
Comparison of the candidate-pipeline age distribution with Barcelona labor-force benchmarks from official labor-force statistics.
Non-binary candidates are shortlisted at less than one-third the rate of men.
Disparate impact analysis comparing non-binary candidates with men; the estimate is based on 285 non-binary candidates.
Women aged 46–55 face compounded disadvantage in the hiring pipeline, with a disparate impact ratio of 0.77.
Intersectional analysis combining gender and age groups, using shortlisting or pipeline selection outcomes.
Women experience adverse impact in shortlisting for mid-salary vacancies, with a disparate impact ratio of 0.786.
Disparate impact analysis of shortlisting outcomes stratified by salary level; the reported DIR is below the paper's 0.80 adverse-impact benchmark and is statistically significant.