Evidence (16496 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
9875 claims
Filter claims →
Productivity
8807 claims
Filter claims →
Governance
7870 claims
Filter claims →
Human-AI Collaboration
7560 claims
Filter claims →
Org Design
4892 claims
Filter claims →
Innovation
4781 claims
Filter claims →
Labor Markets
4004 claims
Filter claims →
Skills & Training
3308 claims
Filter claims →
Inequality
2332 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 | 870 | 233 | 116 | 1066 | 2363 |
| Governance & Regulation | 976 | 451 | 218 | 133 | 1809 |
| Organizational Efficiency | 949 | 224 | 144 | 88 | 1416 |
| Technology Adoption Rate | 764 | 287 | 141 | 122 | 1325 |
| Research Productivity | 501 | 152 | 74 | 362 | 1101 |
| Output Quality | 542 | 216 | 69 | 69 | 896 |
| Decision Quality | 387 | 198 | 94 | 54 | 740 |
| Firm Productivity | 513 | 67 | 101 | 27 | 714 |
| AI Safety & Ethics | 249 | 303 | 73 | 36 | 667 |
| Market Structure | 190 | 192 | 134 | 27 | 548 |
| Task Allocation | 243 | 77 | 91 | 36 | 452 |
| Innovation Output | 291 | 33 | 55 | 20 | 401 |
| Skill Acquisition | 206 | 72 | 65 | 21 | 364 |
| Employment Level | 133 | 63 | 115 | 22 | 335 |
| Fiscal & Macroeconomic | 153 | 79 | 52 | 32 | 323 |
| Task Completion Time | 206 | 37 | 12 | 15 | 272 |
| Firm Revenue | 179 | 52 | 29 | 5 | 266 |
| Consumer Welfare | 130 | 76 | 47 | 13 | 266 |
| Inequality Measures | 48 | 137 | 51 | 6 | 242 |
| Worker Satisfaction | 101 | 81 | 25 | 13 | 220 |
| Error Rate | 84 | 110 | 11 | 5 | 210 |
| Wages & Compensation | 98 | 47 | 30 | 10 | 185 |
| Regulatory Compliance | 88 | 73 | 17 | 7 | 185 |
| Automation Exposure | 66 | 64 | 33 | 16 | 182 |
| Team Performance | 105 | 29 | 30 | 11 | 176 |
| Training Effectiveness | 109 | 22 | 14 | 21 | 168 |
| Developer Productivity | 114 | 21 | 14 | 8 | 158 |
| Job Displacement | 12 | 90 | 24 | 1 | 127 |
| Hiring & Recruitment | 57 | 9 | 9 | 5 | 80 |
| Skill Obsolescence | 6 | 56 | 9 | 1 | 72 |
| Social Protection | 43 | 17 | 8 | 2 | 70 |
| Creative Output | 35 | 21 | 9 | 4 | 70 |
| Labor Share of Income | 18 | 21 | 17 | 1 | 57 |
| Worker Turnover | 15 | 16 | — | 4 | 35 |
| Industry | — | — | — | 1 | 1 |
Aligning technical architecture with organizational governance structures (roles, approval workflows, risk committees) and following a lifecycle (design → validation → deployment → monitoring → decommissioning) is necessary for operationalizing automation governance.
Cross-case lessons and organizational integration recommendations derived from multi-sector case examples and best-practice synthesis; presented as prescriptive architecture and lifecycle processes.
Embedded governance features (access/data usage policy enforcement, model-output controls), human-in-the-loop checkpoints for high-risk decisions, continuous monitoring, and audit trails increase accountability and provide regulatory evidence.
Normative recommendations grounded in industry best practices and case examples; pattern specification enumerating governance controls. Evidence is qualitative rather than quantitative.
A practical reference pattern combining low-code development, RPA, generative AI, and a centralized governance layer can be deployed in mission-critical ERP/CRM landscapes.
Architectural pattern design and cross-case lessons from multi-sector enterprise implementations; qualitative synthesis of industry best practices and case examples. No large-scale quantitative deployment statistics provided.
Embedding policy enforcement, risk controls, human oversight, and continuous monitoring into the automation lifecycle enables organizations to scale automation while preserving data protection, regulatory compliance, operational stability, and long-term system integrity.
Conceptual framework synthesized from industry best practices and comparative analysis of multi-sector enterprise implementations and case examples; architectural pattern design. Methods: qualitative synthesis and pattern extraction. No randomized or large-sample empirical evaluation reported.
Design choices that combine transparency and explainable personalization materially increase consumer trust and purchase intention, making them important levers for firms seeking higher conversion in AI-mediated commerce.
Inference drawn from experimental findings showing transparency and empathetic personalization increased trust (and via trust, purchase intention); applied as an implication for firms.
Higher digital literacy weakens (attenuates) the negative link from perceived manipulation to purchase intention.
Moderator analysis in PLS-SEM including measured digital literacy as a moderator of the perceived manipulation → purchase intention path in the experimental sample (UAE, ages 18–25).
Trust is the primary (dominant) mediator through which transparency and empathetic personalization increase purchase intention.
Mediation analysis within PLS-SEM on experimental data (2 × 2 design); measures include trust and purchase intention; indirect paths from design cues to purchase intention were analyzed.
An empathetic, personalized conversational tone in chatbots increases trust among young consumers (UAE, ages 18–25).
2 × 2 between-subjects experiment manipulating conversational tone (empathetic/personalized vs. generic), same sample (UAE, ages 18–25); trust measured; analyzed with PLS-SEM.
Transparent AI identity disclosure increases trust among young consumers (UAE, ages 18–25).
2 × 2 between-subjects experiment manipulating identity disclosure (AI transparent vs. nondisclosed), sample: young consumers in the UAE aged 18–25; trust measured as a dependent variable; effects estimated using PLS-SEM.
Effective regulation can reshape market equilibria by mandating transparency/audits, enabling interoperability/identity portability, constraining high-risk personalization practices, and requiring privacy-preserving measurement standards.
Policy and economic modeling arguments combined with case examples; prescriptive claim based on plausibility and prior regulatory impacts rather than new causal estimates.
Regulatory interventions (e.g., limits on third-party cookies or profiling) will redirect long-term investments toward privacy-preserving measurement and contextual advertising solutions.
Policy analysis and plausibility argument based on past regulatory changes (cookie deprecation) and industry responses; predictive, not empirically validated within the paper.
Improvements in targeting raise advertiser willingness-to-pay, shifting surplus toward platforms unless competitive pressures or regulation change fee structures.
Economic theory and observed industry trends; no new cross-sectional or panel data regression in this paper to quantify the shift.
Interpretable models, causal evaluation of impact (not only prediction metrics), privacy-by-design, and governance mechanisms are central to sustainable adoption (resilience criteria).
Recommended evaluation framework based on methodological critique (attribution complexity, metric misalignment) and best-practice literature; no empirical validation sample provided.
Long-run viability requires moving beyond raw predictive performance toward resilient, interpretable, policy-aware, and socially legitimate systems.
Normative recommendation grounded in evaluation challenges and literature on trustworthy AI; not an empirically tested hypothesis within the paper.
Regulation shapes incentives for architectures (e.g., favoring first-party data architectures over third-party tracking) (Innovation vs regulatory compliance trade-off).
Policy analysis and observations about industry responses to cookie deprecation and privacy regulation; descriptive industry trend evidence rather than a single empirical trial.
Verifiable compliance (privacy budgets, provenance, auditability) becomes a key economic input; demand for standards, attestation services, and transparent governance frameworks will grow.
Policy/economic argumentation and proposed governance layer including audit logs and policy controllers. No empirical adoption or demand measurements provided.
Prototype simulations indicate that decentralized training with coordination protocols can approach centralized personalization performance under realistic constraints (communication budgets, DP noise, heterogeneity).
Prototype/simulation-based evaluation described qualitatively in the paper. The paper emphasizes illustrative experiments; specific simulation parameters, dataset sizes, and numeric performance comparisons are not reported in detail.
Re-conceptualizing federated learning as a socio-technical infrastructure (not merely a distributed optimizer) enables cross-platform personalized advertising that substantially reduces centralized data custody risks while retaining effective personalization, provided system design integrates secure aggregation, differential privacy, solutions for heterogeneous and delayed feedback, adversarial defenses, and explicit governance mechanisms.
High-level systems and conceptual design with a proposed multi-layer architecture; analytical discussion of privacy/accuracy trade-offs; prototype/simulation-based evaluation described qualitatively. No large-scale field deployment reported; simulations described without detailed sample sizes or numeric benchmarks.
Complementarities matter: digitalization increases AGTFP more when combined with complementary investments and institutions (mechanization, R&D, cooperative organization).
Findings from mediation analysis and interaction/heterogeneity checks indicating larger effects where complementary inputs/institutions are present.
Non-grain-producing provinces experience larger AGTFP gains from digital rural development than major grain-producing provinces.
Comparative sub-sample analysis (non-grain vs. major grain-producing regions) showing larger estimated effects in non-grain-producing areas.
Digital service capacity shows diminishing marginal returns: the marginal positive effect of digital services on AGTFP weakens at more advanced stages of digital-service development.
Panel threshold/modeling of nonlinearity indicating a decreasing marginal effect of the digital service sub-index on AGTFP at higher development levels.
Digitalization accelerates agricultural mechanization and the diffusion of agricultural R&D, which act as channels raising AGTFP.
Mediation analysis including mechanization rate and agricultural R&D input/technology diffusion indicators as mediators; reported significant indirect effects.
Digital rural development strengthens cooperative organizational forms (farmer cooperatives), and this organizational upgrading contributes to higher AGTFP.
Mediation tests showing digitalization is associated with greater cooperative organization indicators, which in turn are associated with higher AGTFP.
Digital rural development encourages larger-scale agricultural operations (land consolidation/scale expansion), which contributes to increases in AGTFP.
Mediation models that include farm scale/land transfer indicators as mediators and find significant indirect effects; analysis notes institutional constraints limit full realization.
Digital rural development raises AGTFP in part by promoting labor mobility and reallocating labor toward higher-productivity uses.
Mediation analysis using the same provincial panel (2012–2022) showing significant indirect effects through labor reallocation/factor allocation variables.
Productivity gains from WAPM are larger in hilly or more topographically complex areas.
Subgroup analysis by terrain (hilly vs. flat areas) reported in the paper based on the CLDS 2014–2018 sample showing stronger WAPM effects in hilly areas.
Productivity gains from WAPM are larger in major grain-producing regions of China.
Subgroup (heterogeneity) analysis by region reported in the paper using the CLDS panel; WAPM treatment effects are reported as larger and statistically stronger in major grain-producing regions.
WAPM offsets the productivity penalties associated with small farm size (i.e., reduces the negative scale effect on productivity for smallholders).
Interaction/heterogeneity analyses in the paper showing smaller negative associations between small farm size and productivity among WAPM adopters in the CLDS 2014–2018 sample.
The productivity advantages of WAPM operate mainly by easing labor constraints (i.e., WAPM mitigates labor shortages that limit productivity).
Mechanism analysis reported in the paper using mediation/interaction-style tests on the CLDS panel (authors report that labor-constraint indicators attenuate treatment effects and/or interact with WAPM adoption).
The productivity gain from WAPM is more than twice that of PAPM (WAPM effect ≈ 2.27× PAPM effect).
Direct comparison of reported regression coefficients (0.486 / 0.214 ≈ 2.27) from the TWFE models on the CLDS 2014–2018 panel; robustness checks with PSM.
Partial agricultural production chain management (PAPM) increases land productivity with an estimated effect (coefficient = 0.214).
Same CLDS 2014–2018 sample and two-way fixed-effects estimation as above; PAPM coefficient reported in the main regression results (PSM used for robustness).
Whole-process agricultural production chain management (WAPM) substantially increases land productivity for grain-producing households in China, with an estimated effect (coefficient = 0.486).
Analysis of a nationally representative panel of grain-producing households from the China Labor-force Dynamics Survey (CLDS), 2014–2018, using two-way fixed-effects (household and year) regression; propensity score matching (PSM) reported as a robustness check.
Empirical models of labor costs, productivity, and AI adoption should use total labor cost (wages + NWC) rather than wages alone; CFIL should be included when modeling transitions from informal to formal employment under automation scenarios.
Methodological recommendation based on the magnitude of measured non-wage and formalization costs (2023 estimates for 19 countries) and implications for correctly specifying empirical models; not an empirical test but a suggested best practice.
Robustness checks and sensitivity analyses (alternative mappings, sector aggregation, price/base-year choices) are performed or at least implied to assess the stability of VIS results.
Paper notes cross-checks with alternative mappings and sensitivity tests to examine stability; specifics depend on paper details.
VIS provides a framework to quantify cross-sectoral labor spillovers and dependencies.
Input–output based VIS construction attributes upstream labor requirements to final sectors, enabling accounting of cross-sector labor embodied in outputs (demonstrated in the electricity case study).
VIS enables robust estimation of productivity trends over time that can inform policy, planning, and comparative analysis across sectors.
VIS produces annual time-series productivity measures using 2014–2023 data; authors argue these trend estimates are suitable for policy and comparative use.
VIS captures interactions among generation, distribution, storage, and consumption consistent with Integrated Energy Systems concepts.
VIS mapping and analysis applied to electricity subsystem sectors (generation, distribution, storage, consumption) showing interconnections via input–output relationships.
Macroeconomic and fiscal gains (GDP growth and increased tax revenues) from platform-enabled productivity are quantitatively estimated via input–output/CGE-style simulations but remain sensitive to assumptions about adoption and policy.
Computed economy-wide estimates from input–output or computable general equilibrium simulations that scale micro productivity improvements; sensitivity analyses run under alternative adoption and policy scenarios.
Observed productivity and participation effects are attributable to AI-enabled capabilities using comparative or quasi-experimental contrasts (e.g., before/after rollouts, adopter vs non-adopter, geographic variation in fulfillment infrastructure).
Identification strategy described: comparative/quasi-experimental contrasts across time, sellers, and geographies; robustness and sensitivity checks reported to support causal attribution.
Algorithmic advertising, dynamic pricing, and demand-forecasting measurably improve ad-targeting outcomes and pricing responsiveness, increasing listing conversions and sales for adopting sellers.
Demand-side algorithmic performance measures (ad-targeting precision/CTR, conversion rates before/after dynamic pricing adoption) and seller sales metrics from platform data and quasi-experimental contrasts.
Platform services and fulfillment-as-a-service reduce fixed costs and complexity of cross-border and domestic sales, lowering market-entry barriers for sellers.
Platform-level service descriptions and seller metric comparisons (seller onboarding rates, cross-border listings, time-to-first-sale) using Amazon FBA case and seller-level data contrasts.
Aggregate micro-level productivity gains from platform AI and automated fulfillment translate into higher productivity-driven GDP growth and increased regional economic activity near logistics hubs.
Macroeconomic aggregation using input–output or computable general equilibrium style simulations that scale micro-level productivity changes to economy-wide GDP and regional spillovers; case analysis of regional activity near fulfillment infrastructure.
Real-time forecasting and automated warehousing increase supply-chain resilience and responsiveness to shocks (demand spikes, logistics disruptions) through faster replenishment and better buffer management.
Operational logistics and inventory metrics under shock scenarios; comparative/quasi-experimental contrasts across regions and time windows with/without AI-enabled forecasting and automated fulfillment; sensitivity analyses on buffer levels and replenishment times.
AI capabilities (demand forecasting, dynamic pricing, automated inventory, robotic fulfillment, algorithmic advertising) materially improve fulfillment speed, inventory turnover, and demand-response, raising seller- and platform-level productivity.
Operational warehousing metrics (pick/pack times, robot usage), inventory metrics (turnover rates), demand-side algorithmic performance measures (forecast accuracy, dynamic price responses), and seller performance metrics (conversion rates, sales) in case studies and comparative contrasts.
AI-enabled e-commerce platforms and automated warehousing (exemplified by Amazon FBA) lower entry and transaction costs for sellers, expanding SME market access and scale.
Case-based analysis using Amazon FBA as representative case; platform- and seller-level performance metrics comparing adopters vs non-adopters and before/after feature rollouts (metrics: seller participation rates, listing activity, fees/fulfilment costs).
Policy recommendation: invest in targeted upskilling and reskilling, strengthen active labor‑market policies, and design scalable safety nets to mitigate distributional harms of AI.
Synthesis of policy implications and repeated recommendations across the reviewed studies; formulated as actionable guidance in the paper.
AI often complements and raises productivity for skilled workers and high-skill tasks.
Synthesis of empirical results from the 17 included studies, several of which report productivity gains or complementary effects when AI is used alongside skilled labor (firm- and task-level analyses reported in the reviewed literature).
New-skill requirements tend to emerge first and most intensely in the United States.
Cross-country comparison of vacancy-level incidence of new-skill mentions (text-extracted) showing earlier and higher concentration in the U.S. relative to other countries in the sample.
Roughly 1 in 10 job vacancies in advanced economies request at least one new skill, and about 5% (roughly half that rate) in emerging economies do so.
Vacancy-level data across a set of advanced and emerging economies, with skills identified by text analysis of job postings; incidence measured as the fraction of vacancies requesting at least one skill labeled as "new" (including IT/AI).
Policy packages combining strengthened social safety nets, regulation of platform labor, investments in digital infrastructure, and incentives for inclusive AI adoption will better manage distributional risks from AI deployment.
Policy synthesis drawing on empirical literature on active labor market policies, social protection, infrastructure investments, and regulatory analyses in the review; the recommendation is inferential from aggregated evidence rather than demonstrated in a single causal study.