{
  "name": "Studies of AI at work, with how each was done",
  "description": "The studies the practices and assessments rest on, one row per study. Each row gives the main result, the authors, year and publication, how the study was done, whether it was peer-reviewed, who took part and where, and the practices it supports or limits.",
  "url": "https://www.ai-first.company/evidence/studies/",
  "version": "2026-10-17",
  "licence": "Data: CC BY 4.0, Paweł Szczęsny, ai-first.company",
  "licenceUrl": "https://creativecommons.org/licenses/by/4.0/",
  "count": 53,
  "rows": [
    {
      "id": "ai-tutor-without-guardrails",
      "url": "https://www.ai-first.company/evidence/studies/ai-tutor-without-guardrails/",
      "title": "Students did better with ChatGPT in practice and worse without it on the exam",
      "headline": "During practice, students with an ordinary GPT-4 chat interface got 48% higher grades and those with a hints-only tutor 127% higher. On the exam without AI, the ordinary-chat group scored 17% below students who never had AI. The hints-only group scored about the same as them.",
      "authors": "Bastani, Bastani, Sungu, Ge, Kabakcı and Mariman",
      "year": 2025,
      "paper": "Generative AI without guardrails can harm learning: Evidence from high school mathematics",
      "source": "Proceedings of the National Academy of Sciences, 122(26), e2422633122",
      "doi": "10.1073/pnas.2422633122",
      "source_url": "https://doi.org/10.1073/pnas.2422633122",
      "design": "Randomised experiment",
      "design_note": "Students were assigned by chance to practise with an ordinary GPT-4 chat interface, with a GPT-4 tutor built to give hints rather than answers, or with no AI. All then sat an exam without AI.",
      "peer_reviewed": true,
      "sample": "Nearly 1,000 high-school mathematics students.",
      "setting": "High schools in Turkey.",
      "practices_supported": [
        "train-staff-on-the-tools-you-chose",
        "check-what-staff-can-still-do-without-the-tool",
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [
        "give-ai-assistance-to-newer-staff-first"
      ],
      "reviewed": "2026-10-17"
    },
    {
      "id": "shadow-learning",
      "url": "https://www.ai-first.company/evidence/studies/shadow-learning/",
      "title": "Shadow learning: how surgical trainees lost their way to learn",
      "headline": "In traditional surgery, trainees learned by doing more of the operation over time. Robotic surgery greatly reduced the trainee’s part in the operation. Only a minority became competent, through unofficial practice.",
      "authors": "Beane",
      "year": 2018,
      "paper": "Shadow Learning: Building Robotic Surgical Skill When Approved Means Fail",
      "source": "Administrative Science Quarterly, 64(1), 87–123",
      "doi": "10.1177/0001839217751692",
      "source_url": "https://doi.org/10.1177/0001839217751692",
      "design": "Case study",
      "design_note": "Two studies: two years of observation at five hospitals, comparing robotic with traditional surgery, and interviews at 13 leading teaching hospitals.",
      "peer_reviewed": true,
      "sample": "Surgical trainees and surgeons at five hospitals, plus interviews at 13 teaching hospitals.",
      "setting": "Hospitals in the United States.",
      "practices_supported": [
        "check-what-staff-can-still-do-without-the-tool",
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [
        "give-ai-assistance-to-newer-staff-first"
      ],
      "reviewed": "2026-10-17"
    },
    {
      "id": "developers-slower-with-ai",
      "url": "https://www.ai-first.company/evidence/studies/developers-slower-with-ai/",
      "title": "Experienced developers were slower with AI and believed they were faster",
      "headline": "Tasks took 19% longer when AI was allowed. Before the study, the developers expected AI to save them 24% of the time. Afterwards they still believed it had saved about 20%.",
      "authors": "Becker, Rush, Barnes and Rein (METR)",
      "year": 2025,
      "paper": "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity",
      "source": "METR, published 10 July 2025 (arXiv 2507.09089); follow-up published 24 February 2026",
      "doi": "",
      "source_url": "https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/",
      "design": "Randomised experiment",
      "design_note": "Each task was assigned by chance to be done with AI tools allowed or not allowed. The researchers recorded the developers’ screens and timed the work, and asked the developers for their own estimates before and after.",
      "peer_reviewed": false,
      "sample": "16 experienced open-source developers working on 246 real tasks. The 2026 follow-up had 57 developers and more than 800 tasks.",
      "setting": "Mature open-source software projects the developers had worked on for years, with AI tools available in early 2025.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "us-census-ai-use-by-firms",
      "url": "https://www.ai-first.company/evidence/studies/us-census-ai-use-by-firms/",
      "title": "US Census: 18% of firms use AI in a business function",
      "headline": "18% of US firms used AI in a business function in November 2025 to January 2026, or 32% when firms are weighted by their number of employees. 66% of users used AI only to help staff with tasks, and 2% of firms cut jobs because of AI.",
      "authors": "Bonney, Breaux, Dinlersoz, Foster, Haltiwanger and Pande",
      "year": 2026,
      "paper": "The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks",
      "source": "US Census Bureau, Center for Economic Studies Working Paper CES-WP-26-25 (April 2026)",
      "doi": "",
      "source_url": "https://www2.census.gov/library/working-papers/2026/adrm/ces/CES-WP-26-25.pdf",
      "design": "Survey",
      "design_note": "Firms answered extra questions on AI in the Census Bureau's regular Business Trends and Outlook Survey. The answers were weighted to represent all US firms.",
      "peer_reviewed": false,
      "sample": "More than 117,000 US firms that answered between 17 November 2025 and 8 February 2026, weighted to represent all US firms.",
      "setting": "United States; all sectors and sizes. About 75% of US firms have fewer than 10 employees.",
      "practices_supported": [
        "run-two-tracks-automate-and-augment"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "young-workers-in-ai-exposed-jobs",
      "url": "https://www.ai-first.company/evidence/studies/young-workers-in-ai-exposed-jobs/",
      "title": "Fewer young people hired into AI-exposed jobs: US payroll data",
      "headline": "Employment of 22–25-year-olds in AI-exposed occupations is 19% below where it would be had it kept pace with less exposed peers. Experienced workers show no such gap. The gap comes mainly from fewer young people being hired.",
      "authors": "Brynjolfsson, Chandar and Chen",
      "year": 2025,
      "paper": "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence",
      "source": "Stanford Digital Economy Lab working paper (revised 12 August 2026)",
      "doi": "",
      "source_url": "https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/",
      "design": "Data analysis",
      "design_note": "The authors used monthly payroll records from ADP, a large US payroll company, and compared employment by age group in occupations more and less exposed to AI.",
      "peer_reviewed": false,
      "sample": "Payroll records for millions of US workers, up to June 2026.",
      "setting": "United States.",
      "practices_supported": [
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [
        "map-tasks-before-automating"
      ],
      "reviewed": "2026-10-17"
    },
    {
      "id": "generative-ai-at-work",
      "url": "https://www.ai-first.company/evidence/studies/generative-ai-at-work/",
      "title": "An AI assistant for 5,172 customer-support agents",
      "headline": "Agents with an AI assistant resolved 15% more customer issues per hour on average. Less experienced agents gained the most. The most experienced gained little and their quality dipped slightly.",
      "authors": "Brynjolfsson, Li and Raymond",
      "year": 2025,
      "paper": "Generative AI at Work",
      "source": "The Quarterly Journal of Economics, 140(2), 889–942",
      "doi": "10.1093/qje/qjae044",
      "source_url": "https://doi.org/10.1093/qje/qjae044",
      "design": "Field study",
      "design_note": "The company gave the assistant to different groups of agents at different times. The researchers compared agents who already had it with agents who did not yet.",
      "peer_reviewed": true,
      "sample": "5,172 customer-support agents.",
      "setting": "Customer support for a company studied by researchers in the United States; some agents worked from other countries.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline",
        "start-with-one-costly-process",
        "give-ai-assistance-to-newer-staff-first"
      ],
      "practices_limited": [
        "keep-entry-level-learning-tasks"
      ],
      "reviewed": "2026-10-13"
    },
    {
      "id": "suitability-for-machine-learning",
      "url": "https://www.ai-first.company/evidence/studies/suitability-for-machine-learning/",
      "title": "Which tasks can machines learn? 18,156 tasks scored",
      "headline": "Most occupations include at least some tasks that machine learning could do. Few occupations could be fully automated. Getting the benefit usually means redesigning what the job contains.",
      "authors": "Brynjolfsson, Mitchell and Rock",
      "year": 2018,
      "paper": "What Can Machines Learn and What Does It Mean for Occupations and the Economy?",
      "source": "AEA Papers and Proceedings, 108, 43–47",
      "doi": "10.1257/pandp.20181019",
      "source_url": "https://doi.org/10.1257/pandp.20181019",
      "design": "Data analysis",
      "design_note": "The authors scored each task in O*NET, the United States government’s database of jobs and tasks, on how well machine learning could do it.",
      "peer_reviewed": true,
      "sample": "18,156 tasks across the occupations in the database.",
      "setting": "United States job and task data.",
      "practices_supported": [
        "map-tasks-before-automating",
        "redesign-the-workflow"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "endoscopist-skill-after-ai",
      "url": "https://www.ai-first.company/evidence/studies/endoscopist-skill-after-ai/",
      "title": "After three months with AI, endoscopists found fewer growths without it",
      "headline": "In colonoscopies done without AI, the share in which doctors found at least one pre-cancerous growth fell from 28.4% to 22.4% within three months of AI being introduced. The doctors lost part of a skill that the tool was doing for them.",
      "authors": "Budzyń, Romańczyk, Kitala and colleagues",
      "year": 2025,
      "paper": "Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study",
      "source": "The Lancet Gastroenterology & Hepatology, 10(10), 896–903",
      "doi": "10.1016/S2468-1253(25)00133-5",
      "source_url": "https://doi.org/10.1016/S2468-1253(25)00133-5",
      "design": "Data analysis",
      "design_note": "The authors took routine quality records from four clinics and compared colonoscopies done without AI in the three months before AI was introduced with those done without AI in the three months after.",
      "peer_reviewed": true,
      "sample": "1,443 colonoscopies done without AI: 795 before AI was introduced and 648 after.",
      "setting": "Four endoscopy centres in Poland taking part in the ACCEPT trial, where AI that marks possible growths on screen was introduced at the end of 2021.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline",
        "check-what-staff-can-still-do-without-the-tool",
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [
        "keep-a-person-checking-what-goes-out"
      ],
      "reviewed": "2026-10-17"
    },
    {
      "id": "organizational-readiness-for-change-review",
      "url": "https://www.ai-first.company/evidence/studies/organizational-readiness-for-change-review/",
      "title": "Does readiness for change predict success? A review of 47 studies",
      "headline": "Whether an organization’s readiness for change leads to a successful rollout, and under what conditions, remains unclear. Studies define and measure readiness in many different ways, and most measure it only once.",
      "authors": "Caci, Nyantakyi, Blum, Sonpar, Schultes, Albers and Clack",
      "year": 2025,
      "paper": "Organizational readiness for change: A systematic review of the healthcare literature",
      "source": "Implementation Research and Practice, 6",
      "doi": "10.1177/26334895251334536",
      "source_url": "https://doi.org/10.1177/26334895251334536",
      "design": "Systematic review",
      "design_note": "The authors searched five databases, screened 2,907 studies and kept those that measured readiness for change with numbers alongside at least one other factor.",
      "peer_reviewed": true,
      "sample": "47 studies.",
      "setting": "Healthcare organizations, internationally.",
      "practices_supported": [
        "do-not-treat-a-maturity-score-as-a-forecast",
        "measure-readiness-and-resistance-separately"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "ai-exposure-and-occupational-mobility",
      "url": "https://www.ai-first.company/evidence/studies/ai-exposure-and-occupational-mobility/",
      "title": "AI exposure and career moves in Brazil and the UK",
      "headline": "Young graduates often start in jobs that AI is likely to replace and move up to jobs that AI is likely to assist. AI could widen that path. It could also remove the starting jobs.",
      "authors": "Cazzaniga, Pizzinelli, Rockall and Tavares",
      "year": 2025,
      "paper": "Exposure to Artificial Intelligence and Occupational Mobility: A Cross-Country Analysis",
      "source": "Economía, 24(1), 314–339",
      "doi": "10.31389/eco.451",
      "source_url": "https://doi.org/10.31389/eco.451",
      "design": "Data analysis",
      "design_note": "The authors tracked how workers have moved between occupations over their careers, and sorted occupations by how exposed they are to AI and whether AI is likely to help or replace the work.",
      "peer_reviewed": true,
      "sample": "Workers’ past moves between occupations, by age and education.",
      "setting": "Brazil and the United Kingdom.",
      "practices_supported": [
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-technostress-and-adoption",
      "url": "https://www.ai-first.company/evidence/studies/ai-technostress-and-adoption/",
      "title": "The same AI pressure can raise or lower willingness to use AI",
      "headline": "When employees saw the demands of AI as a challenge they could meet, they felt positive and were more willing to use AI. When they saw the demands as obstacles, they felt anxious and were less willing.",
      "authors": "Chang, Zhang, Cai and Guo",
      "year": 2024,
      "paper": "Does AI-Driven Technostress Promote or Hinder Employees’ Artificial Intelligence Adoption Intention? A Moderated Mediation Model of Affective Reactions and Technical Self-Efficacy",
      "source": "Psychology Research and Behavior Management, 17, 413–427",
      "doi": "10.2147/prbm.s441444",
      "source_url": "https://doi.org/10.2147/prbm.s441444",
      "design": "Repeated survey",
      "design_note": "A questionnaire answered in three rounds. Participants were recruited by referral from earlier participants.",
      "peer_reviewed": true,
      "sample": "301 employees from 11 organizations.",
      "setting": "Guangdong Province, China.",
      "practices_supported": [
        "measure-readiness-and-resistance-separately"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "who-captures-ai-coding-gains",
      "url": "https://www.ai-first.company/evidence/studies/who-captures-ai-coding-gains/",
      "title": "Across 160,097 developers, experienced programmers gained from AI coding tools and early-career ones did not",
      "headline": "AI wrote an estimated 29% of Python functions in the United States. Quarterly output, measured in code contributions, rose an estimated 3.6%. Experienced developers produced more and moved into new kinds of software; early-career developers showed no clear benefit.",
      "authors": "Daniotti, Wachs, Feng and Neffke",
      "year": 2026,
      "paper": "Who is using AI to code? Global diffusion and impact of generative AI",
      "source": "Science, 391(6787), 831–835",
      "doi": "10.1126/science.adz9311",
      "source_url": "https://doi.org/10.1126/science.adz9311",
      "design": "Data analysis",
      "design_note": "The authors trained a program to recognise Python code written by AI. They ran it over public code on GitHub to track how fast AI coding spread, and estimated how developers' output changed as they took it up.",
      "peer_reviewed": true,
      "sample": "More than 30 million commits to GitHub by 160,097 software developers.",
      "setting": "Public Python code on GitHub, worldwide, with country comparisons.",
      "practices_supported": [],
      "practices_limited": [
        "give-ai-assistance-to-newer-staff-first"
      ],
      "reviewed": "2026-10-17"
    },
    {
      "id": "jagged-frontier",
      "url": "https://www.ai-first.company/evidence/studies/jagged-frontier/",
      "title": "The jagged frontier: AI helped consultants on some tasks and hurt on another",
      "headline": "On 18 tasks that AI handles well, consultants using AI completed 12.2% more tasks and worked 25.1% faster, with better quality. On one task chosen to be beyond the AI’s ability, they were 19 percentage points less likely to get the right answer.",
      "authors": "Dell’Acqua, McFowland, Mollick, Lifshitz-Assaf, Kellogg, Rajendran, Krayer, Candelon and Lakhani",
      "year": 2026,
      "paper": "Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality",
      "source": "Organization Science, 37(2), 403–423",
      "doi": "10.1287/orsc.2025.21838",
      "source_url": "https://doi.org/10.1287/orsc.2025.21838",
      "design": "Randomised experiment",
      "design_note": "An experiment registered in advance. After a baseline task, consultants were assigned by chance to work without AI, with GPT-4, or with GPT-4 plus a short introduction to prompting.",
      "peer_reviewed": true,
      "sample": "758 consultants at Boston Consulting Group.",
      "setting": "A global management consulting firm; realistic consulting tasks designed with the firm.",
      "practices_supported": [
        "keep-a-person-checking-what-goes-out",
        "check-human-plus-ai-by-task-type"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "freelance-postings-after-chatgpt",
      "url": "https://www.ai-first.company/evidence/studies/freelance-postings-after-chatgpt/",
      "title": "Freelance job posts for writing and coding fell 21% after ChatGPT",
      "headline": "Within eight months of ChatGPT's release, posts for writing and coding jobs fell 21% compared with jobs that need manual skills. Posts for image creation fell 17% after image-generating AI arrived.",
      "authors": "Demirci, Hannane and Zhu",
      "year": 2025,
      "paper": "Who Is AI Replacing? The Impact of Generative AI on Online Freelancing Platforms",
      "source": "Management Science, 71(10), 8097–8108",
      "doi": "10.1287/mnsc.2024.05420",
      "source_url": "https://doi.org/10.1287/mnsc.2024.05420",
      "design": "Data analysis",
      "design_note": "The authors compared the number of job posts for work AI can do, such as writing and coding, with posts for work that needs manual skills, before and after ChatGPT and image-generating AI were released.",
      "peer_reviewed": true,
      "sample": "A large data set of job posts on one leading global freelancing platform.",
      "setting": "One global online freelancing platform. Freelancers are sole traders; many of the buyers are small firms.",
      "practices_supported": [
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "copilot-field-experiment-66-firms",
      "url": "https://www.ai-first.company/evidence/studies/copilot-field-experiment-66-firms/",
      "title": "Office AI assistant saved email time but did not change the work",
      "headline": "The 80% of workers given the tool who used it spent about two hours less on email each week and worked less outside regular hours. Apart from those time savings, the study found no change in how much work people did or what kind.",
      "authors": "Dillon, Jaffe, Immorlica and Stanton",
      "year": 2025,
      "paper": "Shifting Work Patterns with Generative AI",
      "source": "NBER Working Paper 33795; arXiv 2504.11436, version 4 (November 2025)",
      "doi": "10.3386/w33795",
      "source_url": "https://arxiv.org/abs/2504.11436v4",
      "design": "Randomised experiment",
      "design_note": "Within each firm, workers were assigned by chance to get Microsoft 365 Copilot or to carry on without it, for six months. The outcomes come from the software’s own activity records, not from what workers said.",
      "peer_reviewed": false,
      "sample": "7,137 knowledge workers at 66 large firms, 3,684 of them given the tool.",
      "setting": "Large firms in several industries, most operating in more than one country, from September 2023 to October 2024.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline",
        "start-with-one-costly-process",
        "start-with-ai-in-software-you-already-pay-for",
        "redesign-the-workflow"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "genai-sales-experiments-online-retail",
      "url": "https://www.ai-first.company/evidence/studies/genai-sales-experiments-online-retail/",
      "title": "Randomised experiments on a retail platform: AI raised sales most where there had been no service",
      "headline": "Sales effects ranged from none to 16.3%. The largest gain came from a pre-sale chatbot where shoppers had previously got no live help. Gains came mostly from more shoppers buying, and returns and ratings did not get worse.",
      "authors": "Fang, Yuan, Zhang, Donati and Sarvary",
      "year": 2025,
      "paper": "Generative AI and Sales Productivity: Field Experiments in Online Retail",
      "source": "arXiv 2510.12049, version 6 (29 June 2026)",
      "doi": "",
      "source_url": "https://arxiv.org/abs/2510.12049",
      "design": "Randomised experiment",
      "design_note": "The platform built generative AI into seven of its workflows. In each, shoppers or products were assigned by chance to the AI version or the usual one, and the researchers compared sales.",
      "peer_reviewed": false,
      "sample": "Millions of shoppers and products, in seven experiments run between September 2023 and June 2024.",
      "setting": "One of the world’s largest cross-border online retail platforms, not named in the paper, serving third-party sellers.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline",
        "start-with-one-costly-process",
        "start-with-ai-in-software-you-already-pay-for",
        "check-human-plus-ai-by-task-type"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "us-small-employer-firms-ai-survey",
      "url": "https://www.ai-first.company/evidence/studies/us-small-employer-firms-ai-survey/",
      "title": "46% of US small employer firms use AI; 31% of users report higher sales",
      "headline": "46% of small employer firms said the business or its employees used AI for work. Among users, 71% reported higher productivity and 31% higher sales. 77% saw no change in labour costs, and 18% spent less on outside services.",
      "authors": "Federal Reserve Banks",
      "year": 2026,
      "paper": "Small Business Credit Survey: 2026 Report on Employer Firms",
      "source": "Federal Reserve Banks, Small Business Credit Survey (published March 2026)",
      "doi": "10.55350/sbcs-20260303",
      "source_url": "https://www.fedsmallbusiness.org/-/media/project/clevelandfedtenant/fsbsite/reports/2026/2026-report-on-employer-firms/2026-report-on-employer-firms.pdf",
      "design": "Survey",
      "design_note": "Firms answered the survey from 3 September to 14 November 2025, with an optional set of questions on AI. The firms were not picked at random. Answers are weighted to match the US population of small employer firms.",
      "peer_reviewed": false,
      "sample": "6,525 firms with 1–499 employees; 5,248 of them answered the questions on AI.",
      "setting": "United States, all sectors.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-occupational-exposure",
      "url": "https://www.ai-first.company/evidence/studies/ai-occupational-exposure/",
      "title": "A measure of how exposed each occupation is to AI",
      "headline": "The study built and tested a score of how far each occupation relies on abilities where AI has made progress. The same data can be turned into scores for industries, regions and single firms.",
      "authors": "Felten, Raj and Seamans",
      "year": 2021,
      "paper": "Occupational, industry, and geographic exposure to artificial intelligence: A novel dataset and its potential uses",
      "source": "Strategic Management Journal, 42(12), 2195–2217",
      "doi": "10.1002/smj.3286",
      "source_url": "https://doi.org/10.1002/smj.3286",
      "design": "Data analysis",
      "design_note": "The authors linked the abilities each occupation needs to the areas where AI has advanced. They then built matching measures for industries and regions.",
      "peer_reviewed": true,
      "sample": "Occupations, industries and counties.",
      "setting": "United States.",
      "practices_supported": [
        "map-tasks-before-automating"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "euro-area-firms-ai-use-by-size",
      "url": "https://www.ai-first.company/evidence/studies/euro-area-firms-ai-use-by-size/",
      "title": "Small firms try AI less but use it as heavily: euro-area firm survey",
      "headline": "41% of micro firms used no AI, against 13% of large firms. But the share of firms making significant use of AI was about the same at every size, at around 7–8%. Size goes with whether a firm starts, not with how far it goes.",
      "authors": "Ferrando, Lamboglia, Rariga and Schmidt",
      "year": 2026,
      "paper": "Adoption and investment in AI across the euro area: Insights from harmonised firm-level data",
      "source": "European Central Bank, Occasional Paper Series No 395",
      "doi": "10.2866/6856108",
      "source_url": "https://www.ecb.europa.eu/pub/pdf/scpops/ecb.op395.en.pdf",
      "design": "Survey",
      "design_note": "Senior executives answered the European Central Bank’s regular survey of firms by telephone or online. In December 2025 they rated their firm’s use of AI on a scale from not in use to significant use.",
      "peer_reviewed": false,
      "sample": "5,067 firms in the December 2025 round, from micro firms (fewer than 10 employees) to large firms (more than 250). Results are weighted by the number of employees.",
      "setting": "Twelve euro-area countries; manufacturing, construction, trade and services.",
      "practices_supported": [
        "assess-readiness-from-several-sources"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-exposure-and-employment-oecd",
      "url": "https://www.ai-first.company/evidence/studies/ai-exposure-and-employment-oecd/",
      "title": "AI exposure and employment in 23 countries",
      "headline": "There was no clear link between how exposed an occupation was to AI and how its employment changed. In occupations with heavy computer use, more exposure went with faster employment growth.",
      "authors": "Georgieff and Hyee",
      "year": 2022,
      "paper": "Artificial Intelligence and Employment: New Cross-Country Evidence",
      "source": "Frontiers in Artificial Intelligence, 5, 832736",
      "doi": "10.3389/frai.2022.832736",
      "source_url": "https://doi.org/10.3389/frai.2022.832736",
      "design": "Data analysis",
      "design_note": "The authors took an existing score of how exposed each occupation is to AI, extended it to 23 countries, and compared it with changes in employment and hours.",
      "peer_reviewed": true,
      "sample": "Occupations in 23 OECD countries.",
      "setting": "23 member countries of the OECD, a group of mostly high-income economies.",
      "practices_supported": [
        "map-tasks-before-automating"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "generative-ai-and-jobs-ilo",
      "url": "https://www.ai-first.company/evidence/studies/generative-ai-and-jobs-ilo/",
      "title": "Generative AI and jobs: a global estimate from the ILO",
      "headline": "The study predicts that generative AI will mostly assist people in their jobs, not replace them. Clerical work is the most exposed, so richer countries and women are affected most.",
      "authors": "Gmyrek, Berg and Bescond",
      "year": 2023,
      "paper": "Generative AI and jobs: A global analysis of potential effects on job quantity and quality",
      "source": "International Labour Organization",
      "doi": "10.54394/fhem8239",
      "source_url": "https://doi.org/10.54394/fhem8239",
      "design": "Data analysis",
      "design_note": "The authors scored the tasks in each occupation for how far a generative AI model (GPT-4) could do them, to estimate how exposed occupations are around the world.",
      "peer_reviewed": false,
      "sample": "Occupations and their tasks, across countries at all income levels.",
      "setting": "Worldwide.",
      "practices_supported": [
        "map-tasks-before-automating"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "genai-and-startup-employment",
      "url": "https://www.ai-first.company/evidence/studies/genai-and-startup-employment/",
      "title": "Startups most exposed to AI cut staff, mostly juniors, and grew faster",
      "headline": "Startups whose work was fully exposed to AI cut employment by 8% within two quarters of ChatGPT's release, mostly in junior and implementation roles. Their code contributions rose 64%, and they became more likely to rank among the startups with the highest sales per employee.",
      "authors": "Gupta, Qian, Simintzi and Sun",
      "year": 2026,
      "paper": "Generative AI and Entrepreneurship",
      "source": "NBER conference paper, version of 14 April 2026",
      "doi": "",
      "source_url": "https://conference.nber.org/conf_papers/f232872.pdf",
      "design": "Data analysis",
      "design_note": "The authors measured how much of each startup's work AI could do, from its mix of jobs just before ChatGPT's release. They then compared more and less exposed startups before and after the release, using workforce, code and sales records.",
      "peer_reviewed": false,
      "sample": "94,789 US startups founded between 2018 and 2021 and listed in Crunchbase or PitchBook. On average they employ 11 workers, have raised $3 million and generate about $700,000 a year in revenue.",
      "setting": "United States; young, fast-growing firms of the kind that seek venture capital, in all sectors. Information technology is the largest sector, at 36.8% of firms.",
      "practices_supported": [
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-triage-in-breast-screening",
      "url": "https://www.ai-first.company/evidence/studies/ai-triage-in-breast-screening/",
      "title": "AI decided which scans got one reader or two: more cancers found, 44% less reading",
      "headline": "Screening with AI found 6.4 cancers per 1,000 women, against 5.0 with two radiologists reading every scan. False positives did not rise. Radiologists did 44.2% fewer readings.",
      "authors": "Hernström, Josefsson, Sartor, Schmidt, Larsson, Hofvind, Andersson, Rosso, Hagberg and Lång",
      "year": 2025,
      "paper": "Screening performance and characteristics of breast cancer detected in the Mammography Screening with Artificial Intelligence trial (MASAI): a randomised, controlled, parallel-group, non-inferiority, single-blinded, screening accuracy study",
      "source": "The Lancet Digital Health, 7(3), e175–e183",
      "doi": "10.1016/S2589-7500(24)00267-X",
      "source_url": "https://doi.org/10.1016/S2589-7500(24)00267-X",
      "design": "Randomised experiment",
      "design_note": "Women were assigned by chance to AI-supported screening or to the standard, in which two radiologists read every scan. In the AI group, the AI sent each scan to one or two readers by risk and marked suspicious areas.",
      "peer_reviewed": true,
      "sample": "105,934 women eligible for breast screening, assigned between 12 April 2021 and 7 December 2022.",
      "setting": "Four screening sites in southwest Sweden (Malmö, Lund, Landskrona and Trelleborg), within the national screening programme.",
      "practices_supported": [
        "keep-a-person-checking-what-goes-out",
        "redesign-the-workflow"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "genai-adopters-and-junior-hiring",
      "url": "https://www.ai-first.company/evidence/studies/genai-adopters-and-junior-hiring/",
      "title": "Firms that put generative AI into their work hired fewer juniors",
      "headline": "From early 2023, junior employment at adopting firms fell about 9% relative to other firms within six quarters. Senior employment was largely unchanged. The fall came from slower hiring, not from people leaving.",
      "authors": "Hosseini and Lichtinger",
      "year": 2025,
      "paper": "Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data",
      "source": "SSRN working paper 5425555 (version of 6 May 2026)",
      "doi": "10.2139/ssrn.5425555",
      "source_url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5425555",
      "design": "Data analysis",
      "design_note": "The authors marked a firm as an adopter once it advertised a job for someone to build generative AI into its work. They compared junior and senior employment at adopters and other firms, before and after adoption.",
      "peer_reviewed": false,
      "sample": "Résumés of 65 million workers at more than 280,000 US firms, 2015–2025, and their job ads. 10,433 firms counted as adopters.",
      "setting": "United States. Adopters were concentrated in professional services (28%) and information (24%).",
      "practices_supported": [
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "chatbots-and-the-labour-market-denmark",
      "url": "https://www.ai-first.company/evidence/studies/chatbots-and-the-labour-market-denmark/",
      "title": "AI chatbots, earnings and hours: evidence from Denmark",
      "headline": "Two years after ChatGPT appeared, AI chatbots were widely used and workers reported benefits. Yet the study found no effect on earnings or recorded hours, and rules out effects larger than 2%.",
      "authors": "Humlum and Vestergaard",
      "year": 2025,
      "paper": "Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI",
      "source": "NBER Working Paper 33777 (revised March 2026)",
      "doi": "10.3386/w33777",
      "source_url": "https://www.nber.org/papers/w33777",
      "design": "Field study",
      "design_note": "The authors linked large surveys on chatbot use to official records of pay and hours. They compared workers and workplaces that adopted chatbots with those that did not, before and after.",
      "peer_reviewed": false,
      "sample": "Workers and workplaces in occupations where AI chatbots are relevant.",
      "setting": "Denmark.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline",
        "start-with-one-costly-process",
        "redesign-the-workflow"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "ai-job-insecurity-and-knowledge-hiding",
      "url": "https://www.ai-first.company/evidence/studies/ai-job-insecurity-and-knowledge-hiding/",
      "title": "Fear of losing a job to AI and the hiding of knowledge",
      "headline": "Employees who feared AI would cost them their job were more likely to hide what they know from colleagues. Part of the reason was lower psychological safety. Confidence in one’s ability to learn AI softened the effect.",
      "authors": "Kim and Kim",
      "year": 2024,
      "paper": "How artificial intelligence-induced job insecurity shapes knowledge dynamics: the mitigating role of artificial intelligence self-efficacy",
      "source": "Journal of Innovation & Knowledge, 9(4), 100590",
      "doi": "10.1016/j.jik.2024.100590",
      "source_url": "https://doi.org/10.1016/j.jik.2024.100590",
      "design": "Repeated survey",
      "design_note": "A survey of the same employees at three points in time.",
      "peer_reviewed": true,
      "sample": "402 employees from various industries.",
      "setting": "South Korea.",
      "practices_supported": [
        "protect-psychological-safety-during-rollout"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "ai-adoption-and-psychological-safety",
      "url": "https://www.ai-first.company/evidence/studies/ai-adoption-and-psychological-safety/",
      "title": "AI adoption, psychological safety and depression among employees",
      "headline": "Where employees reported more AI adoption at their company, they later reported lower psychological safety, and then more symptoms of depression. Ethical leadership weakened the first link.",
      "authors": "Kim, Kim and Lee",
      "year": 2025,
      "paper": "The dark side of artificial intelligence adoption: linking artificial intelligence adoption to employee depression via psychological safety and ethical leadership",
      "source": "Humanities and Social Sciences Communications, 12(1)",
      "doi": "10.1057/s41599-025-05040-2",
      "source_url": "https://doi.org/10.1057/s41599-025-05040-2",
      "design": "Repeated survey",
      "design_note": "An online survey in three rounds, spaced apart in time. The researchers used a statistical model to test how the answers were linked.",
      "peer_reviewed": true,
      "sample": "381 employees of companies.",
      "setting": "South Korea.",
      "practices_supported": [
        "protect-psychological-safety-during-rollout"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "oecd-ai-surveys-employers-workers",
      "url": "https://www.ai-first.company/evidence/studies/oecd-ai-surveys-employers-workers/",
      "title": "OECD surveys of employers and workers on AI at work",
      "headline": "Workers and employers were generally positive about AI’s effect on performance and working conditions, with real worries about job loss. Where employers trained workers and consulted them, outcomes for workers were better.",
      "authors": "Lane, Williams and Broecke",
      "year": 2023,
      "paper": "The impact of AI on the workplace: Main findings from the OECD AI surveys of employers and workers",
      "source": "OECD Social, Employment and Migration Working Papers, No. 288",
      "doi": "10.1787/ea0a0fe1-en",
      "source_url": "https://doi.org/10.1787/ea0a0fe1-en",
      "design": "Survey",
      "design_note": "Two matching surveys, one of employers and one of workers, in the same sectors and countries.",
      "peer_reviewed": false,
      "sample": "Employers and workers in manufacturing and finance.",
      "setting": "Seven OECD countries.",
      "practices_supported": [
        "protect-psychological-safety-during-rollout",
        "budget-for-people-and-process"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "critical-thinking-and-trust-in-ai",
      "url": "https://www.ai-first.company/evidence/studies/critical-thinking-and-trust-in-ai/",
      "title": "Workers who trust AI more report thinking less critically about its output",
      "headline": "Workers who were more confident in generative AI reported less critical thinking about its output. Workers who were more confident in their own ability to do the task reported more. With AI, the thinking shifted towards checking information and fitting AI output into the work.",
      "authors": "Lee, Sarkar, Tankelevitch, Drosos, Rintel, Banks and Wilson",
      "year": 2025,
      "paper": "The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers",
      "source": "Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, Article 1121",
      "doi": "10.1145/3706598.3713778",
      "source_url": "https://doi.org/10.1145/3706598.3713778",
      "design": "Survey",
      "design_note": "Knowledge workers answered questions once, each describing real examples of using AI at work, and said how much critical thinking each task took.",
      "peer_reviewed": true,
      "sample": "319 knowledge workers, who gave 936 first-hand examples of using generative AI in their work.",
      "setting": "Knowledge work in many jobs; a survey by researchers at Microsoft Research.",
      "practices_supported": [
        "budget-for-the-work-ai-creates"
      ],
      "practices_limited": [
        "keep-a-person-checking-what-goes-out"
      ],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-literacy-scales-review",
      "url": "https://www.ai-first.company/evidence/studies/ai-literacy-scales-review/",
      "title": "A review of 16 questionnaires that measure AI knowledge",
      "headline": "Of 16 questionnaires that measure AI literacy, 13 ask people to rate their own knowledge. Only 3 test it. None had been checked for whether it works the same way in different cultures.",
      "authors": "Lintner",
      "year": 2024,
      "paper": "A systematic review of AI literacy scales",
      "source": "npj Science of Learning, 9, 50",
      "doi": "10.1038/s41539-024-00264-4",
      "source_url": "https://doi.org/10.1038/s41539-024-00264-4",
      "design": "Systematic review",
      "design_note": "The author found every published study that tested an AI-literacy questionnaire and scored each questionnaire against a standard checklist for measurement quality (COSMIN).",
      "peer_reviewed": true,
      "sample": "22 studies testing 16 questionnaires, aimed at the general public, students and teachers.",
      "setting": "International.",
      "practices_supported": [
        "test-ai-knowledge-not-self-ratings"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "ai-scribe-randomised-trial",
      "url": "https://www.ai-first.company/evidence/studies/ai-scribe-randomised-trial/",
      "title": "AI scribes in a randomised trial: one tool cut note time by 9.5%, and doctors used them in about a third of visits",
      "headline": "Doctors given Nabla spent 9.5% less time writing notes than doctors without a scribe. Doctors given DAX Copilot showed no clear change. Doctors used the scribes in about a third of their visits.",
      "authors": "Lukac, Turner, Vangala, Chin, Khalili, Shih, Sarkisian, Cheng and Mafi",
      "year": 2025,
      "paper": "Ambient AI Scribes in Clinical Practice: A Randomized Trial",
      "source": "NEJM AI, 2(12)",
      "doi": "10.1056/AIoa2501000",
      "source_url": "https://doi.org/10.1056/AIoa2501000",
      "design": "Randomised experiment",
      "design_note": "Doctors were assigned by chance to one of two AI scribes, Microsoft's DAX Copilot or Nabla, or to carry on as usual. Time spent writing notes came from the health-record system.",
      "peer_reviewed": true,
      "sample": "238 outpatient doctors in 14 specialties.",
      "setting": "Outpatient clinics of UCLA Health in Los Angeles, United States, from 4 November 2024 to 3 January 2025.",
      "practices_supported": [
        "check-human-plus-ai-by-task-type"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "self-rated-vs-tested-digital-skills",
      "url": "https://www.ai-first.company/evidence/studies/self-rated-vs-tested-digital-skills/",
      "title": "Self-rated digital skills versus test results in 629 students",
      "headline": "Students’ ratings of their own digital skills differed from their test results by more than chance. Some were overconfident, others underconfident. Problem solving and handling information were weak even among the best.",
      "authors": "Luna, Ocaña, Ortiz-Colón and Rodríguez Moreno",
      "year": 2026,
      "paper": "From Self-Perception to Reality in Digital Competence: Findings from the DigComp 2.2 Framework",
      "source": "The Electronic Journal of e-Learning, 24(3), 137–153",
      "doi": "10.34190/ejel.24.3.4611",
      "source_url": "https://doi.org/10.34190/ejel.24.3.4611",
      "design": "Survey",
      "design_note": "Students rated their own digital skills and also took a test of the same skills. The researchers compared the two.",
      "peer_reviewed": true,
      "sample": "629 students at a military university.",
      "setting": "Spain.",
      "practices_supported": [
        "test-ai-knowledge-not-self-ratings",
        "assess-readiness-from-several-sources"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "state-of-ai-2025-survey",
      "url": "https://www.ai-first.company/evidence/studies/state-of-ai-2025-survey/",
      "title": "McKinsey’s 2025 global survey on AI in organizations",
      "headline": "88% of organizations use AI in at least one business function, but only about 6% are “high performers” that credit AI with 5% or more of their operating profit. High performers are nearly three times as likely to have rebuilt workflows from the ground up.",
      "authors": "McKinsey & Company",
      "year": 2025,
      "paper": "The State of AI: Global Survey 2025",
      "source": "McKinsey Global Survey, published 5 November 2025",
      "doi": "",
      "source_url": "https://web.archive.org/web/20251215/https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai",
      "design": "Survey",
      "design_note": "An online survey run from 25 June to 29 July 2025. Respondents answered about their own organizations.",
      "peer_reviewed": false,
      "sample": "1,993 participants from 105 countries, across regions, industries and company sizes. 38% work for organizations with more than $1 billion in annual revenue.",
      "setting": "Worldwide.",
      "practices_supported": [
        "name-one-accountable-ai-owner",
        "redesign-the-workflow"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-capability-and-firm-performance",
      "url": "https://www.ai-first.company/evidence/studies/ai-capability-and-firm-performance/",
      "title": "AI capability and company performance: a survey of 143 executives",
      "headline": "Companies whose executives reported stronger AI capability also reported more organizational creativity and better performance. The study produced a 46-question measure of AI capability.",
      "authors": "Mikalef and Gupta",
      "year": 2021,
      "paper": "Artificial intelligence capability: Conceptualization, measurement calibration, and empirical study on its impact on organizational creativity and firm performance",
      "source": "Information & Management, 58(3), 103434",
      "doi": "10.1016/j.im.2021.103434",
      "source_url": "https://doi.org/10.1016/j.im.2021.103434",
      "design": "Survey",
      "design_note": "The authors defined what makes up a company’s AI capability, built a 46-question survey to measure it, and tested how the score related to creativity and performance.",
      "peer_reviewed": true,
      "sample": "143 senior executives.",
      "setting": "United States.",
      "practices_supported": [],
      "practices_limited": [
        "do-not-treat-a-maturity-score-as-a-forecast"
      ],
      "reviewed": "2026-10-13"
    },
    {
      "id": "chatgpt-writing-experiment",
      "url": "https://www.ai-first.company/evidence/studies/chatgpt-writing-experiment/",
      "title": "ChatGPT and professional writing tasks: an experiment",
      "headline": "Professionals who used ChatGPT finished writing tasks 40% faster, and the quality of their work rose by 18%. The gap between stronger and weaker writers narrowed.",
      "authors": "Noy and Zhang",
      "year": 2023,
      "paper": "Experimental evidence on the productivity effects of generative artificial intelligence",
      "source": "Science, 381(6654), 187–192",
      "doi": "10.1126/science.adh2586",
      "source_url": "https://doi.org/10.1126/science.adh2586",
      "design": "Randomised experiment",
      "design_note": "An online experiment, registered in advance. Half of the participants were chosen by chance to use ChatGPT for a paid writing task typical of their job.",
      "peer_reviewed": true,
      "sample": "453 college-educated professionals.",
      "setting": "Online; participants worked on writing tasks matched to their occupation.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline",
        "give-ai-assistance-to-newer-staff-first"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "ai-business-mentor-kenya",
      "url": "https://www.ai-first.company/evidence/studies/ai-business-mentor-kenya/",
      "title": "An AI business adviser helped strong small firms and hurt weak ones",
      "headline": "On average the AI adviser made no detectable difference to revenue or profit. Owners whose businesses were doing poorly at the start did nearly 10% worse with it. Those doing well may have done more than 15% better.",
      "authors": "Otis, Clarke, Delecourt, Holtz and Koning",
      "year": 2026,
      "paper": "The Uneven Impact of Generative AI on Entrepreneurial Performance: Evidence from a Field Experiment in Kenya",
      "source": "Management Science (published 10 July 2026)",
      "doi": "10.1287/mnsc.2024.06909",
      "source_url": "https://escholarship.org/uc/item/0fb001r8",
      "design": "Randomised experiment",
      "design_note": "Owners were assigned by chance to get a GPT-4 business adviser on WhatsApp or links to standard business training guides. The researchers compared revenue and profit, as reported by the owners, over about five months.",
      "peer_reviewed": true,
      "sample": "640 small-business owners from 44 of Kenya’s 47 counties. Most ran food, clothing or farming businesses.",
      "setting": "Kenya, from late May to early November 2023.",
      "practices_supported": [
        "check-human-plus-ai-by-task-type",
        "check-ai-advice-before-you-act-on-it"
      ],
      "practices_limited": [
        "give-ai-assistance-to-newer-staff-first"
      ],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-knowledge-and-delegation",
      "url": "https://www.ai-first.company/evidence/studies/ai-knowledge-and-delegation/",
      "title": "Knowing how AI works improves decisions about what to hand over",
      "headline": "People who were taught how AI works handed over the tasks that suited AI and kept those that suited them. Their results improved. They also said they were less likely to use AI in future.",
      "authors": "Pinski, Adam and Benlian",
      "year": 2023,
      "paper": "AI Knowledge: Improving AI Delegation through Human Enablement",
      "source": "Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems",
      "doi": "10.1145/3544548.3580794",
      "source_url": "https://doi.org/10.1145/3544548.3580794",
      "design": "Randomised experiment",
      "design_note": "Participants were split into two groups. One group was first taught how AI works and where human judgement goes wrong. Both then chose which tasks to hand to an AI.",
      "peer_reviewed": true,
      "sample": "111 participants in two groups.",
      "setting": "A controlled experiment run by a university research team in Germany.",
      "practices_supported": [
        "train-staff-on-the-tools-you-chose",
        "check-human-plus-ai-by-task-type",
        "test-ai-knowledge-not-self-ratings"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "social-penalty-for-using-ai",
      "url": "https://www.ai-first.company/evidence/studies/social-penalty-for-using-ai/",
      "title": "Colleagues judge people who use AI as less competent",
      "headline": "People expected to be judged as less competent and less motivated if they used AI at work, and evaluators did judge AI users that way. The penalty also affected how evaluators rated job candidates.",
      "authors": "Reif, Larrick and Soll",
      "year": 2025,
      "paper": "Evidence of a social evaluation penalty for using AI",
      "source": "Proceedings of the National Academy of Sciences, 122(19), e2426766122",
      "doi": "10.1073/pnas.2426766122",
      "source_url": "https://doi.org/10.1073/pnas.2426766122",
      "design": "Randomised experiment",
      "design_note": "Four online experiments, registered in advance. Participants were assigned by chance to conditions in which a worker did or did not use AI, and either predicted how they would be judged or judged others.",
      "peer_reviewed": true,
      "sample": "4,439 participants across four experiments.",
      "setting": "Online experiments run by researchers at Duke University's Fuqua School of Business, United States.",
      "practices_supported": [
        "protect-psychological-safety-during-rollout",
        "write-a-short-ai-use-policy"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "change-readiness-and-resistance",
      "url": "https://www.ai-first.company/evidence/studies/change-readiness-and-resistance/",
      "title": "Ready for change, resistant, or both?",
      "headline": "Readiness for change and resistance to change are usually treated as two ends of one scale. The authors argue they are separate. A person can be high on both, or low on both.",
      "authors": "Repovš, Drnovšek and Kaše",
      "year": 2019,
      "paper": "Change Ready, Resistant, or Both? Exploring the Concepts of Individual Change Readiness and Resistance to Organizational Change",
      "source": "Economic and Business Review, 21(2)",
      "doi": "10.15458/85451.82",
      "source_url": "https://doi.org/10.15458/85451.82",
      "design": "Literature review",
      "design_note": "The authors reviewed how research has defined and used the two ideas, and compared them side by side.",
      "peer_reviewed": true,
      "sample": "Published research on readiness for change and resistance to change.",
      "setting": "International literature; research team in Slovenia.",
      "practices_supported": [
        "measure-readiness-and-resistance-separately"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "vicious-circles-of-skill-erosion",
      "url": "https://www.ai-first.company/evidence/studies/vicious-circles-of-skill-erosion/",
      "title": "How automation eroded skills at an accounting firm",
      "headline": "Relying on automated software wore down accountants’ skills over several years. The loss went unnoticed by staff and managers. It became a problem because staff were still answerable for work they no longer fully understood.",
      "authors": "Rinta-Kahila, Penttinen, Salovaara, Soliman and Ruissalo",
      "year": 2023,
      "paper": "The Vicious Circles of Skill Erosion: A Case Study of Cognitive Automation",
      "source": "Journal of the Association for Information Systems, 24(5), 1378–1412",
      "doi": "10.17705/1jais.00829",
      "source_url": "https://doi.org/10.17705/1jais.00829",
      "design": "Case study",
      "design_note": "The researchers studied one firm in depth and mapped how reliance on automation, complacency and loss of skill fed each other.",
      "peer_reviewed": true,
      "sample": "One accounting firm that had relied on its software’s automated functions for a number of years.",
      "setting": "An accounting firm; the research team was based in Finland and Australia.",
      "practices_supported": [
        "check-what-staff-can-still-do-without-the-tool",
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "digital-skills-employees-vs-supervisors",
      "url": "https://www.ai-first.company/evidence/studies/digital-skills-employees-vs-supervisors/",
      "title": "Digital skills at a financial company: employees versus supervisors",
      "headline": "Employees and their supervisors gave different accounts of the employees’ digital skills. The gaps were in cybersecurity, in using digital tools to streamline work, and in work-life balance.",
      "authors": "Rogacka",
      "year": 2022,
      "paper": "Digital skills in the workplace – from the employees’ and supervisors’ perspective",
      "source": "e-mentor, 5(97), 53–64",
      "doi": "10.15219/em97.1591",
      "source_url": "https://doi.org/10.15219/em97.1591",
      "design": "Case study",
      "design_note": "The author assessed employees’ digital skills in set areas, then compared what employees said about their skills with how their supervisors described them.",
      "peer_reviewed": true,
      "sample": "Employees and supervisors at one financial company.",
      "setting": "Poland.",
      "practices_supported": [
        "test-ai-knowledge-not-self-ratings",
        "assess-readiness-from-several-sources"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "exposure-is-not-adoption",
      "url": "https://www.ai-first.company/evidence/studies/exposure-is-not-adoption/",
      "title": "Exposure is not adoption: what AI could do versus what it is used for",
      "headline": "Scores of how exposed a job is to AI measure what AI could do. A measure of what AI is actually used for predicted demand for workers better.",
      "authors": "Sedai",
      "year": 2026,
      "paper": "Exposure Is Not Adoption: Measuring Generative AI and Labor Demand",
      "source": "SSRN working paper",
      "doi": "10.2139/ssrn.7528420",
      "source_url": "https://doi.org/10.2139/ssrn.7528420",
      "design": "Data analysis",
      "design_note": "The author compared two kinds of measure against demand for workers: scores of what AI could do in each occupation, and a measure of what people actually use AI for.",
      "peer_reviewed": false,
      "sample": "Occupations and demand for workers in them.",
      "setting": "United States.",
      "practices_supported": [
        "map-tasks-before-automating"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "explainable-ai-visual-inspection",
      "url": "https://www.ai-first.company/evidence/studies/explainable-ai-visual-inspection/",
      "title": "Experts decided better when the AI showed where it had looked",
      "headline": "With the picture, factory workers were 7.7 percentage points more accurate and radiologists 4.7 points more accurate than colleagues who saw only the score. Most of the factory workers with the picture did better than the AI alone.",
      "authors": "Senoner, Schallmoser, Kratzwald, Feuerriegel and Netland",
      "year": 2024,
      "paper": "Explainable AI improves task performance in human–AI collaboration",
      "source": "Scientific Reports, 14, 31150",
      "doi": "10.1038/s41598-024-82501-9",
      "source_url": "https://doi.org/10.1038/s41598-024-82501-9",
      "design": "Randomised experiment",
      "design_note": "Two experiments registered in advance. Participants were assigned by chance to see the AI’s score alone, or the score with a picture marking the areas the AI had reacted to. The AI was the same in both groups.",
      "peer_reviewed": true,
      "sample": "48 factory workers who made 9,600 assessments of electronic products, and 113 radiologists who made 5,650 assessments of chest X-rays.",
      "setting": "A Siemens electronics factory, and an online experiment with practising radiologists.",
      "practices_supported": [
        "make-the-ai-show-what-its-answer-is-based-on"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "digital-readiness-model-for-smes",
      "url": "https://www.ai-first.company/evidence/studies/digital-readiness-model-for-smes/",
      "title": "Whose view does a digital maturity score reflect?",
      "headline": "Many digital maturity models mainly record what senior managers think about progress, not measurable results. Most are built for large firms. The authors propose a model that asks general employees about day-to-day behaviour instead.",
      "authors": "Silva, São Mamede and Santos",
      "year": 2024,
      "paper": "A new proposed model to assess the digital organizational readiness to maximize the results of the digital transformation in SMEs",
      "source": "Journal of Innovation & Knowledge, 10(1), 100644",
      "doi": "10.1016/j.jik.2024.100644",
      "source_url": "https://doi.org/10.1016/j.jik.2024.100644",
      "design": "Literature review",
      "design_note": "The authors reviewed the gaps in existing maturity models and designed a new readiness model for small and medium-sized firms that asks ordinary employees, not only managers.",
      "peer_reviewed": true,
      "sample": "Existing maturity models from the research literature.",
      "setting": "Portugal (research team); the model is aimed at small and medium-sized companies.",
      "practices_supported": [
        "do-not-treat-a-maturity-score-as-a-forecast",
        "assess-readiness-from-several-sources"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "freelance-demand-after-chatgpt",
      "url": "https://www.ai-first.company/evidence/studies/freelance-demand-after-chatgpt/",
      "title": "Freelance demand for writing and translation fell 20–50% after ChatGPT",
      "headline": "After ChatGPT launched, demand for freelance skills it can replace, such as writing and translation, fell 20–50% below the trend expected without it. Short-term jobs fell most. Demand for building AI chatbots nearly tripled, and demand for novice freelancers fell.",
      "authors": "Teutloff, Einsiedler, Kässi, Braesemann, Mishkin and del Rio-Chanona",
      "year": 2025,
      "paper": "Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand",
      "source": "Journal of Economic Behavior & Organization, 235, 106845",
      "doi": "10.1016/j.jebo.2024.106845",
      "source_url": "https://doi.org/10.1016/j.jebo.2024.106845",
      "design": "Data analysis",
      "design_note": "The authors grouped job postings into 116 skill clusters and had GPT-4o label each one as replaceable by, helped by or unaffected by AI chatbots. They then compared demand in each cluster after ChatGPT's launch with the trend expected without it.",
      "peer_reviewed": true,
      "sample": "Over three million job postings from January 2021 to September 2023, according to the University of Oxford's summary of the paper, grouped into 116 skill clusters.",
      "setting": "One leading global online freelance platform, which the paper does not name. Freelancers are sole traders; many of the buyers are small firms.",
      "practices_supported": [
        "keep-entry-level-learning-tasks"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "decade-of-digital-maturity-models",
      "url": "https://www.ai-first.company/evidence/studies/decade-of-digital-maturity-models/",
      "title": "Ten years of digital maturity models: do they predict results?",
      "headline": "It is widely assumed that a higher digital maturity level means better performance. The review found that researchers question this: the models lack objective scoring criteria, and data linking maturity to performance is missing.",
      "authors": "Thordsen and Bick",
      "year": 2023,
      "paper": "A decade of digital maturity models: much ado about nothing?",
      "source": "Information Systems and e-Business Management, 21(4), 947–976",
      "doi": "10.1007/s10257-023-00656-w",
      "source_url": "https://doi.org/10.1007/s10257-023-00656-w",
      "design": "Systematic review",
      "design_note": "The authors reviewed eleven years of research on digital maturity models and set out the main points of dispute among researchers.",
      "peer_reviewed": true,
      "sample": "Published research on digital maturity models.",
      "setting": "International.",
      "practices_supported": [
        "do-not-treat-a-maturity-score-as-a-forecast"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "small-business-ai-adoption-and-hiring",
      "url": "https://www.ai-first.company/evidence/studies/small-business-ai-adoption-and-hiring/",
      "title": "Small firms that adopted AI hired faster: payroll records from Gusto",
      "headline": "Four quarters after owners said they adopted AI, their firms' headcount was 6.8% higher than at comparable firms that had not adopted it. Firms with fewer than 10 employees grew about 10% more. Firms with 10–24 employees showed no measurable change.",
      "authors": "Tremper and Bowen (Gusto)",
      "year": 2026,
      "paper": "Small Businesses that Adopted AI Are Hiring Faster",
      "source": "Gusto Insights, September 2026",
      "doi": "",
      "source_url": "https://gusto.com/resources/gusto-insights/ai-and-smb-hiring-2026",
      "design": "Data analysis",
      "design_note": "Owners said in an April 2025 survey whether and when they had started using AI. The authors then tracked each firm's headcount in its payroll records and compared adopters with firms that knew about AI but had not adopted it.",
      "peer_reviewed": false,
      "sample": "2,262 US small businesses that use Gusto payroll and answered its survey: 1,593 that used generative AI and 669 that knew about AI but had not adopted it.",
      "setting": "United States; small employers in many sectors, with separate results for firms with fewer than 10 employees and with 10–24.",
      "practices_supported": [
        "redesign-the-workflow"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "human-ai-combinations-meta-analysis",
      "url": "https://www.ai-first.company/evidence/studies/human-ai-combinations-meta-analysis/",
      "title": "When a human plus AI beats either alone: 106 experiments",
      "headline": "On average, a human working with AI did worse than the better of the two working alone. The combination lost on decision tasks and gained on tasks that involve creating content.",
      "authors": "Vaccaro, Almaatouq and Malone",
      "year": 2024,
      "paper": "When combinations of humans and AI are useful: A systematic review and meta-analysis",
      "source": "Nature Human Behaviour, 8(12), 2293–2303",
      "doi": "10.1038/s41562-024-02024-1",
      "source_url": "https://doi.org/10.1038/s41562-024-02024-1",
      "design": "Meta-analysis",
      "design_note": "The authors searched three research databases for experiments published from January 2020 to June 2023. Each experiment had to measure humans alone, AI alone, and the two together.",
      "peer_reviewed": true,
      "sample": "106 experiments, giving 370 separate comparisons.",
      "setting": "Experiments from many countries and many kinds of task.",
      "practices_supported": [
        "check-human-plus-ai-by-task-type"
      ],
      "practices_limited": [
        "keep-a-person-checking-what-goes-out",
        "make-the-ai-show-what-its-answer-is-based-on"
      ],
      "reviewed": "2026-10-13"
    },
    {
      "id": "ai-barriers-and-ai-maturity",
      "url": "https://www.ai-first.company/evidence/studies/ai-barriers-and-ai-maturity/",
      "title": "Which barriers hold back AI maturity? A survey of 108 specialists",
      "headline": "Barriers to do with skilled employees had the strongest effect on AI maturity. They held it back directly, and they made infrastructure and data problems look worse. Infrastructure and data barriers had no direct effect larger than chance.",
      "authors": "van Veen, Zhu and Ravesteyn",
      "year": 2026,
      "paper": "How AI Implementation Barriers Shape AI Maturity? The Interacting Roles of Skilled Employees, Technological Infrastructure and Data Quality",
      "source": "CONF-IRM 2026 Proceedings, 17",
      "doi": "",
      "source_url": "https://aisel.aisnet.org/confirm2026/17",
      "design": "Survey",
      "design_note": "A survey analysed with a statistical model that tests several links at once.",
      "peer_reviewed": true,
      "sample": "108 specialists from different sectors. The sample is 108 people, not 108 organizations.",
      "setting": "The Netherlands.",
      "practices_supported": [
        "budget-for-people-and-process"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "small-business-ai-payments-bank-data",
      "url": "https://www.ai-first.company/evidence/studies/small-business-ai-payments-bank-data/",
      "title": "Bank records: 17.7% of small US firms had paid for AI by end-2025",
      "headline": "By December 2025, 17.7% of small firms had paid for an AI service at least once, up from 1.7% in January 2019. The median paying firm spent about $28 a month on AI in 2025. Firms with employees paid for AI more often than sole traders.",
      "authors": "Wheat, Mac and Passalacqua (JPMorganChase Institute)",
      "year": 2026,
      "paper": "Understanding the use of AI among small businesses",
      "source": "JPMorganChase Institute, published 14 April 2026",
      "doi": "",
      "source_url": "https://www.jpmorganchase.com/institute/all-topics/business-growth-and-entrepreneurship/understanding-ai-use-by-small-businesses",
      "design": "Data analysis",
      "design_note": "The authors looked at small firms’ business-account payments at Chase from 2019 to 2025. A firm counted as an adopter once it had paid any of more than 500 AI services.",
      "peer_reviewed": false,
      "sample": "More than 4.6 million small firms with Chase business accounts, in 15 industries, January 2019 to December 2025.",
      "setting": "United States.",
      "practices_supported": [
        "start-with-ai-in-software-you-already-pay-for"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "individual-adoption-vs-organisational-maturity",
      "url": "https://www.ai-first.company/evidence/studies/individual-adoption-vs-organisational-maturity/",
      "title": "Leaders use AI daily while their organizations stay in pilots",
      "headline": "83% of the leaders surveyed use AI every day, but roughly 70% of their organizations have not progressed beyond pilots. Leaders who knew more about AI judged their own organizations more harshly.",
      "authors": "Wojewodzic",
      "year": 2026,
      "paper": "AI Transformation in Polish Organisations: The Gap Between Individual Adoption and Organisational Maturity",
      "source": "e-mentor, 2(114), 75–83",
      "doi": "10.15219/em114.1752",
      "source_url": "https://doi.org/10.15219/em114.1752",
      "design": "Survey",
      "design_note": "An exploratory study combining a survey of leaders with a review of the AI projects they described. It compares graduates of an AI leadership course (102) with a comparison group (66).",
      "peer_reviewed": true,
      "sample": "179 leaders and 539 AI projects.",
      "setting": "Poland.",
      "practices_supported": [
        "budget-for-people-and-process",
        "name-one-accountable-ai-owner",
        "assess-readiness-from-several-sources"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-13"
    },
    {
      "id": "firm-data-on-ai-four-countries",
      "url": "https://www.ai-first.company/evidence/studies/firm-data-on-ai-four-countries/",
      "title": "6,000 executives in four countries: 69% of firms use AI, nine in ten see no effect yet",
      "headline": "69% of firms actively used AI. Nine in ten executives reported no effect of AI on their firm's employment or productivity over the past three years. For the next three years they expect productivity to rise 1.4%, output 0.8% and employment to fall 0.7%.",
      "authors": "Yotzov, Barrero, Bloom, Bunn, Davis, Foster, Jalca, Meyer, Mizen, Navarrete, Smietanka, Thwaites and Wang",
      "year": 2026,
      "paper": "Firm Data on AI",
      "source": "NBER Working Paper 34836 (February 2026, revised March 2026)",
      "doi": "10.3386/w34836",
      "source_url": "https://www.nber.org/papers/w34836",
      "design": "Survey",
      "design_note": "The authors put the same questions to senior executives in four standing business surveys run with central banks and a university, between November 2025 and January 2026.",
      "peer_reviewed": false,
      "sample": "Nearly 6,000 senior executives, such as chief executives and finance directors, in the United States, United Kingdom, Germany and Australia.",
      "setting": "Four countries. The surveys were run with the Federal Reserve Bank of Atlanta, the Bank of England, the Deutsche Bundesbank and Macquarie University.",
      "practices_supported": [
        "measure-time-saved-against-a-baseline"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    },
    {
      "id": "ai-in-the-mittelstand-kfw",
      "url": "https://www.ai-first.company/evidence/studies/ai-in-the-mittelstand-kfw/",
      "title": "German small and medium-sized firms: skills and innovation predict AI use better than size",
      "headline": "20% of German small and medium-sized firms used AI in 2022–2024, five times the share in 2016–2018. Use rose from 19% in firms with fewer than 5 employees to 36% in firms with 50 or more. Once graduates, innovation, research and digital strategy were counted, size and sector mattered little.",
      "authors": "Zimmermann (KfW Research)",
      "year": 2026,
      "paper": "Einsatz von Künstlicher Intelligenz vor allem in Unternehmen mit hohen Innovations- und Digitalisierungsaktivitäten",
      "source": "KfW Research, Fokus Volkswirtschaft No. 533 (11 February 2026), with No. 554 (29 July 2026)",
      "doi": "",
      "source_url": "https://www.kfw.de/PDF/Download-Center/Konzernthemen/Research/PDF-Dokumente-Fokus-Volkswirtschaft/Fokus-2026/Fokus-Nr.-533-Februar-2026-KI-Mittelstand.pdf",
      "design": "Survey",
      "design_note": "Firms in a large standing panel answered whether they had used AI in 2022–2024. The author then compared firms that differed in one trait at a time, such as size or research, with their other traits held the same.",
      "peer_reviewed": false,
      "sample": "About 6,800 firm responses to the 2025 wave of the KfW SME Panel, scaled up to all small and medium-sized firms in Germany.",
      "setting": "Germany; small and medium-sized firms in all sectors, from fewer than 5 employees upwards. KfW is Germany's state-owned development bank.",
      "practices_supported": [
        "assess-readiness-from-several-sources"
      ],
      "practices_limited": [],
      "reviewed": "2026-10-17"
    }
  ]
}