Evidence (3308 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
Filtered →
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 |
Skills Training
Remove filter
Scale of experiments: seven agent–model configurations and 7,308 execution trajectories were used to compute pass rates and deltas.
Reported experimental scale in Methods: 7 agent–model configurations and a total of 7,308 agent execution traces collected and analyzed across tasks/conditions.
Each task was evaluated under three conditions: (1) no Skills, (2) curated (human-authored) Skills, and (3) self-authored (model-generated) Skills.
Experimental protocol described in Methods: three-arm evaluation per task across the SkillsBench benchmark.
SkillsBench benchmark: evaluates 86 tasks spanning 11 domains with deterministic, automated verifiers.
Dataset and benchmark description in the paper: SkillsBench contains 86 tasks across 11 domains and uses deterministic pass/fail verifiers for objective evaluation.
Research should prioritize dynamic, task-based models that include transitional frictions, heterogeneous agents, and sectoral structure to better measure AI exposure and impacts.
Methodological recommendation grounded in the paper's theoretical critique of static occupation-level automation metrics and noted empirical gaps.
Timing uncertainty and measurement challenges make forecasting the pace and scale of AI-induced employment change inherently uncertain.
Methodological limitations section noting uncertainty in AI adoption speed and difficulties mapping capabilities to tasks and predicting new occupation emergence.
Research agenda: there is a need for causal studies on AI’s impact on accounting labor demand and firm performance, analyses of distributional effects across firm sizes and industries, and evaluation of regulatory frameworks for reliable, interpretable AI in financial reporting.
Author-stated research priorities drawn from gaps identified in the literature review; not an empirical finding.
Policy implications include workforce retraining, standards for AI auditability and transparency, and regulation balancing innovation and controls (privacy, fraud prevention).
Policy recommendations based on identified risks and barriers discussed in the paper rather than empirical policy evaluation.
For stronger causal evidence, recommended empirical methods include difference-in-differences on adopting firms vs. controls, matched samples, and randomized pilots for particular tools, supplemented by qualitative interviews.
Methodological recommendations stated in the paper (not an empirical finding); no implementation/sample reported in the abstract.
Empirical approach measured and compared expectation formation, innovation responses, and pipeline outcomes across local exposure to closures and across distinct entrepreneurial identity groups.
Methodological description: survey-based, cross-country quantitative approach using measures of local exposure (nearby closures), identity classification (family/purpose-driven vs. wealth-driven), and outcomes (expectations, perceived impediments, self-reported innovation, pipeline transitions) in a sample >27,000.
The study analyzes a cross-country sample of more than 27,000 entrepreneurs across 43 countries (survey-based, comparative).
Descriptive claim about the dataset used in the paper: survey-based sample size >27,000 spanning 43 countries as reported in Data & Methods.
The paper's evidence is policy‑oriented, qualitative and analytical; it does not report causal estimates from new field data and produces testable propositions and an empirical agenda instead.
Explicit methods statement in the paper: structured desk review, corridor process mapping, governance gap analysis; absence of field experiments or causal quantitative analysis.
Calibration via Method of Simulated Moments (MSM) matches six empirical moments to discipline mechanism magnitudes.
Model calibration procedure reported in the paper: MSM matching six chosen empirical moments that summarize key pre/post-AI patterns (paper states six moments were used).
The paper highlights governance risks requiring transparency about LLM-derived mappings, mitigation of model biases, privacy-preserving data practices, and careful communication of uncertainty to avoid overconfident policy recommendations.
Explicit discussion of risks and governance considerations in the paper; this is an acknowledgment rather than an empirical claim. No implementation or audit evidence is provided.
Backtesting the architecture on historical automation waves and recent AI introductions will validate model design and calibration.
Paper explicitly proposes backtesting and holdout validation using historical automation episodes and recent AI adoption events; does not report completed backtests or empirical sample sizes.
Evaluations reporting outcomes predominantly relied on learner surveys, knowledge/skill tests, or self‑reported behavior change measures.
Methods of evaluation extracted from the included studies: most used surveys, tests, or self-report measures to assess Kirkpatrick‑Barr levels 1–3.
The study used a cross-sectional quantitative survey (purposive sampling) of pharmaceutical-sector employees in Karnataka, India (N = 350) and analyzed relationships using PLS-SEM (SmartPLS 4.0).
Study design and methods as reported in the paper summary: cross-sectional survey, purposive sampling, N = 350, analysis via Partial Least Squares Structural Equation Modeling (SmartPLS 4.0).
Policy recommendations include: invest in open metadata standards; fund pilot programs to evaluate ROI (earnings, placement, employer satisfaction); require model governance and periodic external audits for AI-assisted curriculum tools; and support smaller providers via shared infrastructure or accreditation hubs.
Explicit policy recommendations in paper (prescriptive).
Careful attention is needed to validity/reliability of assessments and to selection bias in employment outcome measurement.
Paper's methodological caveat (prescriptive); no empirical bias analysis provided.
Suggested evaluation metrics include placement rates, wage premiums, competency attainment, compliance scores, cost per qualification, and update latency.
Paper's recommended evaluation metrics (prescriptive).
Implementation requires integration with information systems for documentation, versioning, metadata, and audit trails, and benefits from continuous monitoring dashboards.
Paper's technical implementation recommendations (prescriptive).
Recommended analysis methods are qualitative (semi-structured interviews, focus groups, document review) and quantitative (surveys, competency mapping, statistical analysis of outcomes), plus systematic audit methods including traceability checks.
Paper's methods section (methodological specification).
Data inputs for the framework should include competency taxonomies, labor-market signals, regulatory requirements, learner assessment results, and stakeholder interviews.
Paper's data-input specification (descriptive).
Management principles emphasised are transparency, traceability of outcomes, IT integration for documentation, and continuous monitoring/evaluation.
Explicit management principles in paper (prescriptive).
Research and audit should emphasise validity, reliability, and compliance using mixed methods (qualitative interviews/focus groups; quantitative surveys/statistics) and systematic curriculum audits.
Recommended research & audit approach in paper (methodological guidance).
Tools recommended include logigrams (visual decision/compliance flows) and algorigram (algorithmic step-flows for planning, assessment, audit).
Tool definitions and recommendations in paper (descriptive).
Core components of the framework are inputs (learner needs, industry requirements, regulatory standards), processes (curriculum mapping, competency alignment, career assessment), and outputs (structured lesson plans, compliance-ready frameworks, career-path documentation).
Framework component list provided in paper (descriptive).
Scope of the program includes curriculum design, organisational management, career-alignment, and audit/compliance processes.
Explicit scope statement in paper (descriptive).
The framework foregrounds logical modelling (logigrams, algorigrams) and mixed-methods data analysis to support design, auditability, and alignment with industry and regulatory standards.
Paper's methodological design and tool recommendations (conceptual). No empirical implementation data reported.
The program offers a comprehensive curriculum-engineering framework linking organizational orientation, management systems, lesson planning, and career assessment into traceable, compliance-ready curriculum products.
Paper's program description and framework specification (conceptual); no empirical evaluation or sample size reported.
The paper calls for subsequent quantitative validation (using task-based, matched employer-employee, and provider-level panel data) to estimate causal impacts on productivity, health outcomes, wages, and employment composition across the three interaction levels.
Stated research agenda and measurement recommendations in the paper's discussion section.
The study is qualitative and small-sample (four case) and therefore interpretive and illustrative rather than statistically generalizable.
Explicit methodological statement in the paper: design = qualitative multiple case study, sample = four AI healthcare applications.
The study identifies a three-level taxonomy of human–AI interaction in healthcare: AI-assisted, AI-augmented, and AI-automated.
Conceptual taxonomy derived from multiple qualitative case studies (n=4) using cross-case comparison and Bolton et al. (2018)'s three-dimensional service-innovation framework.
Few longitudinal or randomized studies were found, which limits the evidence base for causal claims about digital transformation's effect on productivity.
Review recorded a limited number of longitudinal analyses and quasi-experimental designs among the 145 studies; randomized studies were scarce or absent.
Measurement heterogeneity across studies includes self-reported productivity, output-per-worker metrics, and process efficiency indicators.
Extraction of productivity indicators from included studies (detailed in Methods/Extraction fields) showed multiple distinct measurement approaches.
There is a lack of standardized instruments and inconsistent controls for confounding factors across studies, limiting causal inference about the effect of digital transformation on productivity.
Review extraction documented varied instruments/measures and inconsistent adjustment for confounders across the included studies; few randomized or robust longitudinal designs were found.
Heterogeneous definitions of 'digital transformation' and a variety of productivity measurement approaches prevented a formal quantitative meta-analysis.
Extraction found wide variation in how digital transformation and productivity were defined and measured across the 145 studies (self-reported productivity, output per worker, process efficiency metrics, etc.), leading authors to forgo meta-analysis.
535 records were identified across Scopus, Web of Science, ScienceDirect, IEEE Xplore, and Google Scholar, of which 145 met PRISMA 2020 inclusion criteria.
Search and screening procedure documented in the review: initial database searches yielded 535 records → duplicates removed → screening → full-text evaluation → 145 included studies.
Non-probability sampling and self-reported measures limit claims about prevalence and causality; cross-sectional design cannot capture dynamics of skill acquisition over time.
Study limitations explicitly reported by authors: non-probability sampling, self-reported measures, and cross-sectional design.
The study is primarily diagnostic and prescriptive rather than empirical: no explicit empirical dataset, causal identification strategy, or statistical estimation is reported.
Methods section of the paper explicitly characterizes the work as conceptual, systems-oriented, and not reporting empirical evaluation data.
Research recommendation: invest in longer-run, rigorous impact evaluations (RCTs, panel studies) and system-level assessments to capture spillovers and sustainability outcomes.
Authors' stated research agenda based on identified methodological gaps (limited long-term and system-level evidence) in the review.
There is variation in study design and quality in the evidence base (RCTs, quasi-experimental studies, observational case studies, pilots).
Methodological caveats noted by the authors summarizing the diversity of designs reported across reviewed studies.
The review used a structured literature review with thematic synthesis and a comparative effect-size analysis to quantify ranges for yield, cost, and efficiency outcomes.
Authors' description of review approach and analytical methods in the Data & Methods section.
The evidence base reviewed comprises more than 60 peer-reviewed articles and institutional reports from 2020–2025, primarily focusing on Sub-Saharan Africa.
Statement in the paper's Data & Methods section describing the scope and composition of the review sample.
Effect sizes and impacts vary substantially across contexts—by crop, farm size, and institutional setting.
Comparative synthesis across studies showing heterogeneity in reported outcomes and authors' methodological caveats highlighting context dependence.
Technologies assessed in the review include predictive analytics, digital advisory systems, smart irrigation, pest/disease detection, and precision fertilization.
Descriptive synthesis of the types of AI and digital technologies evaluated across the >60 reviewed articles and reports (2020–2025).
These quantitative performance figures come from case‑level, high‑performer pilots and should not be treated as typical industry benchmarks.
Authors' caveat based on the composition of evidence in the review (skew towards pilots and selected advanced implementations; limited longitudinal/multi‑project empirical studies).
Inter‑rater reliability for the study selection/encoding was Cohen’s κ = 0.83 (substantial agreement).
Reported inter‑rater reliability statistic from the review's quality control step (Cohen's kappa = 0.83).
The review screened 463 Scopus records (2018–2026) and selected 160 peer‑reviewed studies using a PRISMA‑guided process.
Systematic literature review described in paper: Scopus search (2018–2026), PRISMA screening and eligibility filtering; initial n=463, final n=160.
The study has potential selection and ecological-validity constraints because it was conducted at two institutions across six courses, limiting generalizability.
Authors note limitations regarding sample scope (two institutions, six courses) and the ecological validity of the experimental tasks/settings.
The study employed a multi-method approach combining experimental quantitative analysis (descriptives, GLM, non-parametric robustness checks) with qualitative topic-based coding of open-ended survey responses.
Methods description: randomized/experimental assignment; quantitative analyses using GLM and non-parametric tests; qualitative topic-based coding of student responses; sample N = 254 across six courses at two institutions.