Evidence (677 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 |
Among 11,429 European SMEs, AI adoption was associated with a higher probability of revenue growth; the effect was stronger when AI was combined with the Internet of Things and big-data analytics and more moderate when AI was adopted in isolation.
The paper summarizes Ardito et al. (2025), reporting an empirical analysis of 11,429 European SMEs and heterogeneous effects depending on complementary technologies.
Incentive allocation must balance incentive cost promised in advance against downstream advertising revenue observed after an exposure is generated.
The paper’s formal business-process description and MDP reward definition, which specifies reward as RTB revenue minus the incentive payout.
Privacy rules and design standards affecting minors will change firms' revenue models and incentives to develop child-focused AI products.
The article discusses limits on collecting minors' data, transparency requirements, restrictions on targeted advertising, and their effects on firm incentives; no causal regulatory estimate is reported.
Waiting is relatively safe for a firm whose existing product cannot be replaced by the emerging technology, but it causes substantial value loss for a firm whose product is directly substitutable by that technology.
Agent-based model with demand-side substitution eroding the legacy revenue of non-adapters according to product exposure; stated as Proposition P4.
The effects of peer-driven digital transformation vary across industries and market structures.
Reported heterogeneity and supplementary analyses using industry and market-structure splits.
The experiment used self-reported click intention rather than logged behavioral taps as the primary click outcome because a tap could be mechanically confused with initiating a swipe gesture.
Participants navigated sequential screenshots by swiping, and the authors treated click intention as a dichotomous post-task outcome while using logged taps only descriptively.
The liquidity provider's PnL is affine in the provider's own liquidity depth: fee revenue and impermanent loss are linear in liquidity, while gas costs enter as a liquidity-independent constant.
Theorem 3.3 combines the preceding lemmas and the fixed-cost rebalancing formulation. The result is theoretical and assumes the provider is small enough that its actions do not affect future liquidity-taking activity.
In risk-neutral settings, narrow-position agents achieve higher fee income but incur higher impermanent loss and gas costs because they rebalance more frequently.
The claim is stated in the paper's contribution summary and follows from comparisons between PPO_narrow and wider-position strategies. The experimental design evaluates agents across 1,000 simulations per evaluation and varies volatility and gas costs, but the supplied excerpt does not report numerical effect estimates.
Big Tech’s surplus profits derive from a combination of industrial and commercial profit categories rather than from a single rent channel.
The paper develops a preliminary theoretical and classificatory framework distinguishing profits from production, intellectual property, capital investment, distribution, market intermediation, advertising, and platform monetization.
The intensity, or depth and amount at a point in time, of top management team digital attention has an inverted U-shaped relationship with export performance: moderate intensity improves performance, while excessive intensity harms it.
Linear and nonlinear two-way fixed-effects regressions on publicly listed Chinese firms from 2013 to 2023, including an inverted-U specification for attention intensity.
The dispatch policy produced heterogeneous outcomes across DSPs: DSP-A had 13.3% fewer requests, 4.4% more responses, and 7.4% higher net revenue, while DSP-B had 4.2% fewer requests and 18.1% higher net revenue.
Per-DSP auction-funnel analysis for the recent 7-day Mid-RPM window.
The benefit of Hermes depended on the underlying model: gains in mean final net assets ranged from 11.5% for Claude Opus 4.8 to 187.8% for Qwen3.7-Max, while Kimi K2.6 performed 4.1% worse under Hermes than under ReAct.
Framework analysis reporting model-specific percentage differences in mean final net assets.
The paper characterizes the current market condition as concentration rather than a literal market collapse: aggregate platform revenue continued to grow while median per-title outcomes weakened.
Comparison of computed attention-concentration measures with secondary estimates of Steam aggregate revenue and reported median-revenue brackets; the paper explicitly notes that the revenue data are secondary.
We analyze economic implications of mitigating detected group bias for profit-maximizing personalized targeting, characterizing when bias correction alters targeting decisions and profits and the trade-offs involved.
Theoretical economic analysis linking bias correction to targeting decisions and profits; likely models and comparative statics presented in paper (described in abstract). No sample size reported in abstract.
Publishers contend these summaries substitute for source pages and cannibalize traffic, while platforms argue they are complementary by directing users through included links.
Statement of stakeholder positions reported in paper (qualitative claim about opinions/arguments from publishers and platforms; no quantified empirical evidence cited in the excerpt).
The negative effect of AI integration on firm performance is significantly moderated by upper-echelon characteristics.
Longitudinal empirical analysis reporting a statistically significant moderation effect of upper-echelon (top management) characteristics on the AI integration — firm performance relationship (specific moderator variables and sample size not provided in the excerpt).
Models focused on financial returns achieved the highest profitability but lower sustainability metrics.
Experimental comparisons reported in the results showing that RL variants emphasizing financial objectives produced higher returns while scoring worse on ESG/emission measures.
The effects of R&D expenditures on export performance are stronger in countries with low and medium levels of export performance, and they weaken at higher export levels.
Quantile regression (MMQR) findings showing larger R&D coefficients at lower and middle quantiles of the export distribution and smaller coefficients at higher quantiles for the 15-country panel (2013–2023).
The relationship between AI investments and export performance is neither linear nor homogeneous and differs depending on countries’ export performance levels (i.e., effects vary across quantiles of the export distribution).
Quantile regression analysis (MMQR) applied to the panel of 15 OECD countries (2013–2023), reporting heterogeneous effects across conditional distribution (different quantiles).
A formal structural break test confirms parameter instability after 2023 (Wald test p-value = 0.028).
Formal structural break testing reported in the paper; Wald test statistic and associated p-value (p = 0.028) are given.
Regression results indicate a reversal in interest-rate sensitivity: the 10-year Treasury yield had a negative effect on NVDA returns before 2023 but a strong positive association thereafter.
Regression analysis (period-split or interaction terms) reported in the paper showing coefficient sign change on the 10-year Treasury yield pre- and post-2023.
The results reveal a U-shaped relationship between AI investments and revenue growth.
Main empirical finding from regressions on the panel data (multi-way clustered CGM estimation), as reported in the Findings; robustness checks (alternative AI proxies, 2SLSIV) are noted.
AI adoption affects firms' pricing and quantity decisions.
Theoretical treatment in the paper (index entries on price decisions, pricing, quantity decisions, pages 180-181, 196-197) — economic models and analysis.
Returns from adopting AI often depend on how many other firms adopt AI, even in a competitive market, due to externalities that alter prices and volatility.
Chapter synthesis noting externalities in AI adoption that affect individual firm returns depending on aggregate adoption.
Profitability had a significant influence on deposit behavior.
Profitability included as a control in the panel regressions; reported as statistically significant in affecting deposits in the random effects model on the 2019–2023 balanced panel of 12 banks.
Public sector banks show stronger effects of voluntary AI disclosure on deposits; the ownership dummy had a negative coefficient, suggesting private banks face a credibility gap.
Regression models include an ownership dummy; results reported that public sector banks exhibit stronger disclosure–deposit relationships and the ownership dummy coefficient is negative. Same balanced panel (12 banks, 2019–2023) and random effects specification.
I find the optimal black box targeting policy and compare it to the optimal comprehensible policy.
Modeling and comparative analysis reported in the paper: derivation/estimation of optimal black box policy and comparison to the comprehensible policy (theoretical and/or empirical comparison described).
The impact of AI on bilateral trade is asymmetric: the promoting effect on exports is stronger than on imports.
Comparative coefficients/results reported from the extended gravity model using panel data for 2020–2024 between China and the ten ASEAN countries; the abstract explicitly notes this asymmetry.
Customer credit risk assessment, AI-enabled credit scoring, digital transaction analytics, and dynamic pricing influence revenue performance in commercial banks.
Paper frames a quantitative study examining the influence of these four factors on revenue performance using primary data from banking professionals and digital banking customers analyzed with SEM. No sample size provided.
Unprecedented AI capital expenditure coexists with persistent operating losses, speculative valuations, and fragile revenue models.
Empirical characterization asserted in the paper (references implied); the provided excerpt does not state specific datasets, firms counted, dates, or sample size.
The modality gap (weaker penalty for visual vs. textual AI-use disclosure) widens when AI is used in final products but narrows when AI is used in marketing materials.
Interaction analyses across application stages (final product vs. marketing material) within the 41,073 Kickstarter projects, using LLM-assisted classification to label both modality and application stage and entropy balancing for covariate control.
A standard learning agent can obtain near-reference revenue per available room (RevPAR) while failing to learn market-like yield management: it sells too aggressively, undercuts, or collapses to modal price buckets.
Experiments in a two-hotel revenue-management simulator where Hotel A is trained against a fixed rule-based competitor (Hotel B); comparison of learned agent behavior to market-like yield management patterns observed in traces.
AI adoption moves value creation away from physical resources and human collaboration toward continuous token flows produced through data refinement loops.
Theoretical/analytical claim within the Structural Dissolution Framework and illustrative discussion; no empirical quantification provided in the text excerpt.
Only a small subset of LLM retailers can consistently achieve capital appreciation, while many hover around the break-even point.
Empirical results from the 20-agent benchmark experiments reported in the paper, contrasting capital appreciation for winners vs break-even for many agents.
Benchmarking on 20 open- and closed-source LLM agents reveals significant performance disparities and a winner-take-most phenomenon.
Empirical evaluation described in the paper using 20 LLM agents (open- and closed-source); results reported show uneven performance distribution.
Reimbursement models (fee-for-service vs. capitation) will influence whether cost savings from GenAI are realized or offset by increased service volume.
Economic incentive framework and prior health-economics literature cited; the paper does not provide direct empirical tests but references plausible incentive channels.
The vendor triple bind is associated with rising license and consumption costs, fewer effective procurement concessions, and higher enterprise exit costs.
Market pricing observations, reported CIO experiences and complaints, documented contracting practices, and procurement case examples.
The paper argues that reporting financial infrastructure is likely to be more effective than reporting communication infrastructure because financial accounts are more costly and difficult for scammers to replace.
Model-based supply-elasticity argument supplemented by cited black-market price estimates for financial accounts and communication accounts.
If scam conversion is halved, reporting is doubled, and the ban threshold is halved simultaneously, the model predicts that scam revenue per channel decreases by a factor of at least 8.
Analytical revenue-ratio calculation based on Theorem 1, illustrated using the parameter changes s′ = s/2, r′ = 2r, and t′ = t/2.
Reverting from MB-IQL to plain IQL reduces per-user net profit by 6.56%.
Online A/B testing of the learned policies; the abstract reports p<0.0001 for the comparison.
Exposure to ESG controversies is negatively associated with firm value.
Fixed-effects panel regressions modeling ESG controversies and firm value for firms in the global ESG Leaders Index over 2018–2023; controversies are described as a proxy for ESG washing or credibility problems.
For firms exposed to substitution, waiting reduces value without necessarily causing insolvency.
Model outcome under the demand-side substitution mechanism, distinguishing loss of firm value from exit or solvency failure.
Cybersecurity incidents can damage firm valuation through reputational damage, customer churn, lost future revenue, and increased borrowing costs.
Qualitative identification of economic-impact channels based on literature and illustrative incident and industry evidence; no quantified valuation estimate is provided.
Same-trait ad-app pairings were associated with lower effectiveness across click-intention and recall outcomes.
The authors compared personality pairings in logistic-regression analyses of awareness-conditioned click intention, brand recall, and category recall in the randomized online experiment.
MIT's NANDA initiative found that 95% of the generative-AI pilots it analyzed delivered no measurable P&L impact.
The paper cites MIT Project NANDA's analysis of 300 public deployments, based on executive interviews, employee surveys, and deployment analysis. The paper notes that the report was preliminary and treats the figure as directional evidence rather than a precise estimate.
Technology-focused sponsors pay substantially lower company-level price-to-revenue multiples at entry than generalist sponsors.
Comparative deal-level analysis of specialist- and generalist-sponsored technology-unicorn acquisitions in the U.S. sample.
Impermanent loss for a concentrated-liquidity position is linear in the position's own liquidity depth.
Lemma 3.2 uses the positive homogeneity of the concentrated-liquidity reserve formulas in the liquidity-depth parameter. The paper notes that this holds whether or not the position is in range.
Assignment to AI Mode Search reduced the fraction of users clicking Reddit by 21.2 percentage points.
Intent-to-treat estimate from the randomized field experiment; 95% CI [-27.5pp, -14.8pp], p < 0.001; robust to using clicks per day.
Assignment to AI Mode Search reduced the fraction of users clicking through to news sites by 12.5 percentage points.
Intent-to-treat estimate from the randomized field experiment; 95% CI [-18.7pp, -6.3pp], p < 0.001; findings were robust to using clicks per day and an alternative news-domain classification.
Assignment to AI Mode Search reduced click-through rate to external sites by 18.8 percentage points relative to Current Search.
Preregistered randomized field experiment on Google Search; intent-to-treat estimate among 1,100 participants eligible for the post-survey; 95% CI [-22.2pp, -15.3pp], p < 0.001.