Skip to content

Reviewed edition · qasb report 10/2026 · Topic report · Work

AI and work: signs of weaker entry opportunities

For young people and for hiring in AI-exposed occupations, mostly descriptive studies from Germany, France and Denmark find less favourable trends. October’s evidence shows no broad employment collapse.

Published
October 27, 2026 · as of October 26
Reading time
12 min
Publisher
qasb Research
Method
qasb methodology v1

Key findings

  1. −3.8%Full-time employment of 20–29-year-olds in highly AI-exposed occupations, Germany 2025 versus 2022 (less exposed −2.6%).8Evidence: observed
  2. about −28%Permanent hires of 21–29-year-olds in France’s most AI-exposed occupations, March 2026 versus November 2022.3Evidence: observed
  3. −25%Employment in AI-exposed occupations at Danish firms that first used AI in 2023, against their own trend by late 2025.19Evidence: observed
  4. 41%of firms in Germany with the highest occupational AI exposure used AI in the second half of 2025.9Evidence: observed

1.0Summary

Among young people in highly AI-exposed occupations, full-time employment in Germany and permanent hiring in France developed less favourably than in less exposed ones. Both findings are descriptive. US data on recent graduates and UK business reports show little effect.

  • Companies adopting AI shifted their workforce towards senior positions. Danish firms that first used AI in 2023 hired below trend without laying off more.
  • Experiments, some run or co-funded by AI vendors, show performance gains for IT apprentices and patent lawyers. Teachers in Lima saved no measurable time.
  • Surveys link AI use with better jobs, more responsibility and more frequent loneliness without establishing cause and effect.
  • Causal evidence on wages and health is missing, as are October studies on social partners, health care, the creative industries and customer service.

2.0Starting point: use and measurement

Ahead of the labour market data, this chapter asks who actually uses AI at work and how reliable occupational measures built on model judgements are. It draws on an EU-wide survey of workers, a survey of firms in Germany and an audit of language-model judges. All three describe distributions or measurement quality, not an effect of AI.

Figure 1

AI use at work by job quality

Share of workers in the EU using AI at work, AIM-WORK survey 2024–2025

Cross-sectional AIM-WORK data: they show how use is distributed, not an effect of AI on job quality. They cover conscious use of AI tools, mainly chatbots. The source does not state the sample size.Evidence: observed · Sources 13

According to the AIM-WORK survey 2024–2025, 29.9% of workers in the EU used AI at work; in better-quality jobs the figure was 35.3%, in jobs with lower earnings and security 16.9%. Weekly use was 12.7% versus 7.2%.13A1

In Germany, 41% of firms with the highest occupational AI exposure used AI in the second half of 2025, compared with 13% of firms with the lowest exposure; yet among firms in the most exposed group that neither used nor planned to use AI, 85% saw no possible application, barely fewer than among the least exposed (91%).9A2

Further findings

  • Depending on the judge model, 3.0% to 97.9% of the same AI responses were rated acceptable, versus 61.1% by workers in the occupation.14A3

3.0Entry: signs of weaker opportunities

Studies approach the question of whether AI makes entry into work harder in different ways: with employment and hiring data on young people by the AI exposure of their occupation in Germany and France, with online profiles at foreign affiliates in 41 countries, and with Danish register data on firms that use AI.

Figure 2

Full-time employment by AI exposure in Germany

Change 2025 versus 2022, 20–29-year-olds and all age groups

Unadjusted changes without uncertainty measures. For all age groups the source gives approximations (about 1% higher, a little over 3% lower). Adjusted for working-from-home suitability and cyclical vulnerability, the association for 20–29-year-olds is not significant.Evidence: observed · Sources 8

In Germany, full-time employment of 20–29-year-olds in 2025 was 3.8% below its 2022 level in highly AI-exposed occupations and 2.6% below in less exposed occupations; across all age groups, however, employment in the most exposed occupational groups was about 1% above, and in the less exposed groups a little over 3% below, its 2022 level.8A4

In France, permanent-contract (CDI) hires of 21–29-year-olds in the occupations most exposed to generative AI were about 28% below their November 2022 level in March 2026, compared with about 8% in the least exposed occupations; contract terminations also fell more there (about 23% versus 6%).3A5

In Danish firms that first used AI in 2023, the employment shortfall by late 2025 arose through less hiring rather than layoffs and was concentrated in AI-exposed occupations (25% below trend versus 7% in other occupations); wages did not respond.19A6

Further findings

  • The junior share at foreign affiliates of AI-adopting firms was 1.9 percentage points below control affiliates, mainly via senior growth.18A7

4.0Labour market data: no broad collapse visible so far

Further data from the US, the UK and five East Asia and Pacific economies cover unemployment, business reports and job postings, although the East Asian postings run only to mid-2025. They are complemented by qualitative reports to the US Federal Reserve from contacts not selected at random.

Figure 3

Unemployment among young US bachelor’s graduates in summer

Aged 22–25, no longer enrolled, June–August average, 2022 to 2026

US only. Relative to older graduates and young people without a degree, the rate did not rise significantly in 2026. An expanded measure including people outside the labour force who want a job was the highest of the five summers in 2026 at 10.4%, also without a significant increase.Evidence: observed · Sources 4

The unemployment rate of young US bachelor's graduates (aged 22–25, no longer enrolled) was 7.3% in summer 2026, within the range of summers 2022–2025; relative to older graduates and young people without a degree it did not rise significantly.4A8

Of UK businesses with 10 or more employees that used AI in September 2026, 6% said they had fewer staff as a result of adopting AI and fewer than 1% said they had more; 38% of them named administrative and clerical roles as most affected.5A9

In the October 2026 Beige Book, five of the twelve Federal Reserve Districts reported that some firms had reduced staffing through attrition, delayed hiring or expected less job creation after productivity-raising investment in AI; overall employment was little changed in most Districts.10A10

Further findings

  • In five East Asia and Pacific economies, postings for more automatable occupations have not developed discernibly differently since ChatGPT.6A11
  • US manufacturing postings requiring AI skills advertised wages about 70% higher than all manufacturing postings, an unadjusted gap.17A12

5.0Tasks: better performance, time savings unresolved

Three randomised experiments measure the effect of AI: on IT apprentices in test tasks, on patent lawyers over three months and on teachers in Lima over eight months. Two observational studies, at Meta and at a consumer goods multinational, add to the picture. Possible conflicts of interest: Google ran and funded the patent experiment, Microsoft co-funded the Lima study and the Meta study is in-house.

Figure 4

Additional points achieved with ChatGPT in a field experiment

Gain over a no-internet control group, 673 final-year IT apprentices

Rounded values. The no-internet control group achieved 42.3% of attainable points on the second task beyond the curriculum. The authors note the effect may be overstated relative to normal working conditions. Test tasks rather than on-the-job work.Evidence: observed · Sources 1

In a preregistered field experiment with 673 final-year IT apprentices, access to ChatGPT, compared with a no-internet control group, raised the share of points achieved on two tasks beyond the apprenticeship curriculum by about 26 and 31 percentage points, and on the within-curriculum task by about 44 percentage points.1A13

In a preregistered three-month field experiment with 133 patent lawyers from eleven US firms, an AI drafting assistant raised the blind-rated quality of patent drafts by 0.34 standard deviations after 10 days and 0.38 after 90 days; in a subsequent review task without AI, only experienced lawyers (seven or more years) outperformed the control group, while junior lawyers showed no average gain.16A14

A programme of training, an AI licence and follow-up saved teachers in Lima no measurable time on any of seven core tasks.

In a randomized trial with 1,269 sixth-grade teachers in Lima, a programme of one-day training, an AI licence and eight months of follow-up saved no measurable time on any of seven core tasks; however, teachers wrote better prompts in blind grading and were less likely to rate an inadequate AI output as good.2A15

Further findings

  • At Meta, after accounting for context, more AI code was not linked to more incidents but landed disproportionately in less central code.7A16
  • At a multinational, Copilot users’ collaboration time rose by about 1.4 hours a month relative to non-users in the first three months.12A17

6.0Working conditions: rules, algorithms, well-being

The chapter combines the legal position, the AI Act’s deadlines for employment systems, with surveys of workers in the US and the EU and of platform workers in Phnom Penh. These capture prevalence, self-assessments and associations, not measured effects. Algorithmic management also covers systems without AI, and the Phnom Penh survey does not measure AI use.

Figure 5

AI Act: deadlines for employment systems

Amending Regulation (EU) 2026/1744 and the start of the high-risk obligations

  1. Jul 8, 2026Amending Regulation 2026/1744Regulation (EU) 2026/1744 of 8 July 2026 postpones the high-risk obligations for employment systems. On AI literacy (Article 4) it requires measures, not a specific level.
  2. Aug 2, 2026Original datePreviously planned start of the high-risk obligations for employment systems, postponed by the amendment.
  3. Dec 2, 2027High-risk obligations applyObligations for employment systems under Annex III, such as candidate selection or performance monitoring, including informing workers’ representatives and affected workers before use.
Legal position, not a measured effect. The information duty follows Union and national rules on informing workers, and national co-determination rules were not examined. Older than the reporting period and included as context only.Evidence: observed · Sources 21–22

Older legal context (July 2026): Since Regulation (EU) 2026/1744, the AI Act’s high-risk obligations for employment systems under Annex III, such as candidate selection, task allocation or performance monitoring, including the duty of employers as deployers to inform workers’ representatives and affected workers before use, apply only from 2 December 2027 instead of 2 August 2026. The AI literacy obligation (Article 4) now requires measures to support it, but no specific level of literacy.21,22A18

Of US employees who had used AI in their job, 63% said it helped them complete tasks faster and 46% that it improved the quality of their work; 31% said their AI use had led their employer to ask them to take on more responsibilities. Of all employees, 52% said they had less influence over the adoption of new technology than they would like.11A19

According to the AIM-WORK survey, about 40% of workers in the EU in all three job-quality clusters received work instructions from algorithms; algorithmic evaluation, such as automatic rewarding, benchmarking or task assignment based on customer ratings, affected 28.2% on EU average and 31.9% in jobs with atypical and longer working times.13A20

Further findings

  • Among employed US adults, more AI use was associated with slightly higher life satisfaction and more frequent loneliness.20A21
  • Young platform drivers in Phnom Penh averaged 11 to 11.6 hours a day including waiting time, and the ILO survey does not measure AI use.15A22

7.0Evidence

Not every statement in this report is equally well supported. The table shows the key statements with their evidence type, AI relationship and sources.

StatementEvidenceAI linkSources
41% of firms in Germany with the highest occupational AI exposure used AI in the second half of 2025observednot stated9
−3.8%: Full-time employment of 20–29-year-olds in highly AI-exposed occupations, Germany 2025 versus 2022 (less exposed −2.6%)observednot stated8
about −28%: Permanent hires of 21–29-year-olds in France’s most AI-exposed occupations, March 2026 versus November 2022observednot stated3
−25%: Employment in AI-exposed occupations at Danish firms that first used AI in 2023, against their own trend by late 2025observednot stated19
Young US bachelor's graduates in summer 2026: 7.3% unemployed, within the earlier rangeobservednot stated4
IT apprentices with ChatGPT: about 26 and 31 percentage points more beyond the curriculumobservednot stated1
AI Act: obligations for employment systems apply only from 2 December 2027observednot stated21,22
Algorithmic evaluation affected 28.2% of workers in the EUobservednot stated13

The 22 reviewed statements from 22 original sources fall into chapters 2.0 (3), 3.0 (4), 4.0 (5), 5.0 (5) and 6.0 (5); all are classed as observed. “Not stated”: the edition does not assess the AI relationship of these statements separately. The scope and limitations of every statement are listed in the appendix (A1 to A22).

Original sources checked on October 27, 2026.

8.0What organisations can do now

  1. Keep an eye on entry: analyse hiring and employment of newcomers in highly AI-exposed roles separately. In the Danish data, the shortfall arose through less hiring, not layoffs.
  2. Monitor how newcomers build skills: whether they build lasting expertise with AI support is open. In the patent experiment, the small junior group showed no average advantage over the control group in a review task without AI.
  3. Measure the effect in your organisation rather than assume it: the experiments cover single occupations, test tasks or a few months. Occupational measures based on language-model judgements depend strongly on the rating instrument.
  4. Inform and enable: support AI literacy (Article 4 of the AI Act). From December 2027, workers’ representatives and affected workers must be informed before Annex III employment systems are used. National co-determination rules were not examined.

9.0Method and limitations

Reporting period: October 1, 2026 up to the data cutoff of October 26, 2026 at 1:00 PM GMT+1 (Berlin time). This topic report is published before the end of the month; sources published after the cutoff follow in the next edition of the series. Compiled: October 27, 2026 at 12:23 AM GMT+1. This is the first baseline. The sources come from different years; their publication dates are not events in the reporting month. No change from an earlier edition was measured.

The report uses only reviewed public original sources. Customer conversations, private dossiers and personal customer assessments are not inputs. Observed, model estimate, scenario and open describe the evidence for each finding. Source values retain their original units and baselines; they are not combined into an overall score.

Editorial approval: qasb editorial team · Approved: October 27, 2026 at 12:25 AM GMT+1. The editorial presentation arranges the reviewed statements and contains no figure that is not in them. Presentation approved by the qasb editorial team · October 27, 2026 at 2:27 PM GMT+1.

Limits

  • This topic report covers one topic area: “Work”. It is not an exhaustive survey of all studies or incidents.
  • A narrow study finding is not a forecast for a particular customer, occupation or security system.
  • No qualified finding does not prove no effect; source-access gaps are disclosed separately.
  • This report contains no illustrative measurement series or invented confidence rating.

10.0Sources

  1. 1Task Expansion with Generative AI: Experimental Evidence on Middle-Skilled, Early-Career WorkersStanford Digital Economy Lab (Working Paper, zugleich IZA Discussion Paper 18999); Hanna Brosch, Christina Langer, Philipp Lergetporer, Sandra Mueller, Harald Pfeifer & Nele TschöpeOctober 7, 2026digitaleconomy.stanford.edu/publication/task-expansion-with-generative-aiexperimental-evidence-on-middle-skilled-early-career-workers/
  2. 2Not Faster, but Better: Generative AI and Teacher Skill in Peru (Policy Research Working Paper 11476)Weltbank (Development Research Group / Education and Skills Global Department); Carla Z. Glave, Carolina Lopez, Ezequiel Molina & Román Andrés ZárateOctober 1, 2026documents1.worldbank.org/curated/en/099309210012625906/pdf/IDU-c261aadb-6c32-48af-a07c-9cff5e86c716.pdf
  3. 3Exposition des métiers à l’IA générative, adoption par les entreprises et dynamiques du marché du travail en France (LIFT Working Paper 2026-02)HEC Paris, Laboratoire LIFT; Marco Guido Palladino (Banque de France), Antonin Bergeaud, Antoine Bertheau (Banque de France), Simon Bunel & Oscar DegosOctober 8, 2026heclift.github.io/website/files/papers/LIFT_WP_2026-02.pdf
  4. 4The Early Impacts of AI on Employment among Recent College Graduates (NBER Working Paper 35796)National Bureau of Economic Research; Robert W. Fairlie & Jane Wu (University of California, Los Angeles)September 28, 2026www.nber.org/papers/w35796
  5. 5Business insights and impact on the UK economy: 8 October 2026Office for National Statistics (ONS), Business Insights and Conditions Survey (BICS) Welle 165October 8, 2026www.ons.gov.uk/businessindustryandtrade/business/businessservices/bulletins/businessinsightsandimpactontheukeconomy/8october2026
  6. 6East Asia and Pacific Economic Update, October 2026: Riding the AI WaveWeltbank (World Bank Group)October 6, 2026documents1.worldbank.org/curated/en/099839510062630534/pdf/IDU-eaea19eb-317c-4a27-8332-d217283ebd7a.pdf
  7. 7A Preliminary Analysis of the Impact of AI Assisted/Generated Code on Quality, Centrality and Review Time at ScaleAudris Mockus, Peter C. Rigby, Don Stewart, Chandra Maddila & Nachiappan Nagappan (Meta Platforms; University of Tennessee; Concordia University); ESEM 2026, LIPIcs Bd. 394, Schloss Dagstuhl – Leibniz-Zentrum für InformatikOctober 5, 2026drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ESEM.2026.93
  8. 8KI als Jobkiller für Berufseinsteiger? Entwarnung mit VorbehaltInstitut für Arbeitsmarkt- und Berufsforschung (IAB), IAB-Forum; Melanie Arntz, Michele Dinelli & Oliver SchlenkerOctober 13, 2026iab-forum.de/ki-als-jobkiller-fuer-berufseinsteiger-entwarnung-mit-vorbehalt/
  9. 9Weshalb die meisten Betriebe Künstliche Intelligenz (noch) nicht nutzenInstitut für Arbeitsmarkt- und Berufsforschung (IAB), IAB-Forum; Christian KagerlOctober 20, 2026iab-forum.de/weshalb-die-meisten-betriebe-kuenstliche-intelligenz-noch-nicht-nutzen/
  10. 10The Beige Book: Summary of Commentary on Current Economic Conditions by Federal Reserve District, October 2026Federal Reserve System (Board of Governors); erstellt bei der Federal Reserve Bank of PhiladelphiaOctober 14, 2026www.federalreserve.gov/monetarypolicy/files/BeigeBook_20261014.pdf
  11. 11AI Benefits at Work Unevenly DistributedGallup, Inc.; Jeffrey M. Jones (American Job Quality Study mit Jobs for the Future und The Families & Workers Fund)October 6, 2026news.gallup.com/poll/714602/benefits-work-unevenly-distributed.aspx
  12. 12The Effects of Generative AI on Collaboration and Social Networks Among Knowledge WorkersGiuseppe Soda, Alessandro Iorio & Leonardo Rizzo (Università Bocconi, SDA Bocconi); arXiv-Preprint 2610.32461v1October 23, 2026arxiv.org/abs/2610.32461v1
  13. 13European employment and social outlook: annual review 2026, Chapter 2Europäische Kommission, Generaldirektion Beschäftigung, Soziales und Integration (DG EMPL); Daten der AIM-WORK-Befragung von DG EMPL und Gemeinsamer ForschungsstelleOctober 13, 2026ec.europa.eu/employment_social/employment_analysis/esde/2026/Chapter%202.html
  14. 14Right Order, Wrong Scale: Auditing LLM Judges for Occupational AI MeasurementHarry Lyu & Neil Thompson (Massachusetts Institute of Technology); arXiv-Preprint 2610.02492v1October 1, 2026arxiv.org/abs/2610.02492v1
  15. 15Young digital platform workers in Cambodia: Evidence and insightsInternationale Arbeitsorganisation (ILO), Research Brief; Jong Eun Oh & Phu HuynhOctober 8, 2026www.ilo.org/sites/default/files/2026-10/26-001_Research%20Brief%20-%20Cambodia%20Youth%20DLP%20-%20Final.pdf
  16. 16Does AI Assistance Enhance or Erode Expertise? Evidence from a Three-Month Field Experiment in Patent Drafting (NBER Working Paper 35720)National Bureau of Economic Research; David Autor, Tanya Rodchenko, Josh Martin, Zanna Iscenko, Scott Strand, David Pearl & Melissa Ferere (Studie von Google durchgeführt und finanziert)September 7, 2026www.nber.org/papers/w35720
  17. 17AI on the Factory Floor: Evidence from Manufacturing Job PostingsBoard of Governors of the Federal Reserve System, FEDS Notes; Maria D. TitoSeptember 30, 2026www.federalreserve.gov/econres/notes/feds-notes/ai-on-the-factory-floor-evidence-from-manufacturing-job-postings-20260930.html
  18. 18How Does AI Change Labor Demand? Evidence from 41 CountriesStanford Digital Economy Lab (Working Paper); Bharat Chandar (Stanford University) & Bouke Klein Teeselink (King’s College London)September 21, 2026digitaleconomy.stanford.edu/publication/how-does-ai-change-labor-demand/
  19. 19AI Adoption and Firm Size: Employment Effects in Monthly Administrative Data (Working Paper No. 223)Danmarks Nationalbank; Simone Maria Bonin, Saman Darougheh & Andreas KuchlerSeptember 16, 2026www.nationalbanken.dk/en/news-and-knowledge/publications-and-speeches/working-paper/2026/ai-adoption-and-firm-size-employment-effects-in-monthly-administrative-data
  20. 20Associations of artificial intelligence use at work with psychological distress, loneliness, and life satisfaction among employed US adultsJemar R. Bather, Adolfo G. Cuevas, Melody S. Goodman & José A. Pagán (NYU School of Global Public Health); Preventive Medicine Reports (Elsevier)September 18, 2026doi.org/10.1016/j.pmedr.2026.103640
  21. 21Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI)Europäisches Parlament und Rat der Europäischen Union; Amtsblatt der EU, Reihe L, 24.7.2026July 24, 2026eur-lex.europa.eu/eli/reg/2026/1744/oj
  22. 22Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act)Europäisches Parlament und Rat der Europäischen Union; Amtsblatt der EU, Reihe L, 12.7.2024July 12, 2024eur-lex.europa.eu/eli/reg/2024/1689/oj
Authors, reuse basis and rights records
  1. 1Stanford Digital Economy Lab (Working Paper, zugleich IZA Discussion Paper 18999); Hanna Brosch, Christina Langer, Philipp Lergetporer, Sandra Mueller, Harald Pfeifer & Nele Tschöpe · Reuse basis: All rights reserved (© Stanford University; terms allow downloads for personal, non-commercial use only); our own factual paraphrase and link only, no figures · Rights and license record
  2. 2Weltbank (Development Research Group / Education and Skills Global Department); Carla Z. Glave, Carolina Lopez, Ezequiel Molina & Román Andrés Zárate · Reuse basis: World Bank working paper with no Creative Commons statement in the document (licence field in the Open Knowledge Repository unreachable at retrieval); use under the World Bank terms with attribution; our own factual paraphrase and link only, no tables or figures · Rights and license record
  3. 3HEC Paris, Laboratoire LIFT; Marco Guido Palladino (Banque de France), Antonin Bergeaud, Antoine Bertheau (Banque de France), Simon Bunel & Oscar Degos · Reuse basis: CC BY 4.0 (per the document: reproduction permitted with attribution); our own paraphrase and link only, no figures · Rights and license record
  4. 4National Bureau of Economic Research; Robert W. Fairlie & Jane Wu (University of California, Los Angeles) · Reuse basis: All rights reserved (© 2026 Robert W. Fairlie & Jane Wu); short excerpts of up to two paragraphs permitted with full credit; here our own factual paraphrase and link only · Rights and license record
  5. 5Office for National Statistics (ONS), Business Insights and Conditions Survey (BICS) Welle 165 · Reuse basis: Open Government Licence v3.0; our own paraphrase with attribution and link · Rights and license record
  6. 6Weltbank (World Bank Group) · Reuse basis: CC BY 3.0 IGO; our own paraphrase with attribution, no figures; this interpretation has not been reviewed by the World Bank · Rights and license record
  7. 7Audris Mockus, Peter C. Rigby, Don Stewart, Chandra Maddila & Nachiappan Nagappan (Meta Platforms; University of Tennessee; Concordia University); ESEM 2026, LIPIcs Bd. 394, Schloss Dagstuhl – Leibniz-Zentrum für Informatik · Reuse basis: CC BY 4.0; our own paraphrase and link only, no figures · Rights and license record
  8. 8Institut für Arbeitsmarkt- und Berufsforschung (IAB), IAB-Forum; Melanie Arntz, Michele Dinelli & Oliver Schlenker · Reuse basis: CC BY-SA 4.0 (as stated on the article page); our own factual paraphrase with attribution and link, no figures · Rights and license record
  9. 9Institut für Arbeitsmarkt- und Berufsforschung (IAB), IAB-Forum; Christian Kagerl · Reuse basis: CC BY-SA 4.0 (as stated on the article page); our own factual paraphrase with attribution and link, no figures · Rights and license record
  10. 10Federal Reserve System (Board of Governors); erstellt bei der Federal Reserve Bank of Philadelphia · Reuse basis: Public domain according to the Federal Reserve Board website policy (citation requested); our own paraphrase with attribution and link, no figures or third-party data reproduced · Rights and license record
  11. 11Gallup, Inc.; Jeffrey M. Jones (American Job Quality Study mit Jobs for the Future und The Families & Workers Fund) · Reuse basis: All rights reserved (© 2026 Gallup, Inc.); citing individual findings with attribution and a link to the original article is provided for in Gallup Permissions; no continuous reproduction of content, no graphics; our own factual paraphrase and link only · Rights and license record
  12. 12Giuseppe Soda, Alessandro Iorio & Leonardo Rizzo (Università Bocconi, SDA Bocconi); arXiv-Preprint 2610.32461v1 · Reuse basis: CC BY-NC-ND 4.0; our own paraphrase and link only, no figures, no adaptation of the original · Rights and license record
  13. 13Europäische Kommission, Generaldirektion Beschäftigung, Soziales und Integration (DG EMPL); Daten der AIM-WORK-Befragung von DG EMPL und Gemeinsamer Forschungsstelle · Reuse basis: Reuse under Commission Decision 2011/833/EU, content under CC BY 4.0 unless otherwise stated; our own paraphrase with attribution, no charts; the Commission is not responsible for this interpretation · Rights and license record
  14. 14Harry Lyu & Neil Thompson (Massachusetts Institute of Technology); arXiv-Preprint 2610.02492v1 · Reuse basis: CC BY 4.0; our own paraphrase and link only, no figures · Rights and license record
  15. 15Internationale Arbeitsorganisation (ILO), Research Brief; Jong Eun Oh & Phu Huynh · Reuse basis: CC BY 4.0 (© ILO 2026, as stated in the document); our own paraphrase with attribution and link, no figures · Rights and license record
  16. 16National Bureau of Economic Research; David Autor, Tanya Rodchenko, Josh Martin, Zanna Iscenko, Scott Strand, David Pearl & Melissa Ferere (Studie von Google durchgeführt und finanziert) · Reuse basis: All rights reserved (© 2026 the authors); short excerpts of up to two paragraphs permitted with full credit; here our own factual paraphrase and link only · Rights and license record
  17. 17Board of Governors of the Federal Reserve System, FEDS Notes; Maria D. Tito · Reuse basis: Public domain according to the Federal Reserve Board website policy (citation requested); our own paraphrase with attribution and link, no figures or third-party data reproduced · Rights and license record
  18. 18Stanford Digital Economy Lab (Working Paper); Bharat Chandar (Stanford University) & Bouke Klein Teeselink (King’s College London) · Reuse basis: All rights reserved (© Stanford University; terms allow downloads for personal, non-commercial use only); our own factual paraphrase and link only, no figures · Rights and license record
  19. 19Danmarks Nationalbank; Simone Maria Bonin, Saman Darougheh & Andreas Kuchler · Reuse basis: No licence statement found in the document or on the publication page; copyright with Danmarks Nationalbank or the authors; our own brief factual paraphrase and link only, no figures or tables · Rights and license record
  20. 20Jemar R. Bather, Adolfo G. Cuevas, Melody S. Goodman & José A. Pagán (NYU School of Global Public Health); Preventive Medicine Reports (Elsevier) · Reuse basis: CC BY 4.0; our own paraphrase and link only, no figures · Rights and license record
  21. 21Europäisches Parlament und Rat der Europäischen Union; Amtsblatt der EU, Reihe L, 24.7.2026 · Reuse basis: Official EU legal act; reuse permitted with attribution under the EUR-Lex legal notice (Decision 2011/833/EU); our own paraphrase and link; only the version published in the Official Journal is authentic · Rights and license record
  22. 22Europäisches Parlament und Rat der Europäischen Union; Amtsblatt der EU, Reihe L, 12.7.2024 · Reuse basis: Official EU legal act; reuse permitted with attribution under the EUR-Lex legal notice (Decision 2011/833/EU); our own paraphrase and link; only the version published in the Official Journal is authentic · Rights and license record

How to cite

qasb Research (2026). AI and work: signs of weaker entry opportunities. Topic report 10/2026, data cutoff October 26, 2026. https://www.qasb.ai/en/reports/2026-10-arbeit

Appendix

For every statement: evidence type, change status, scope and limitations, and the observation period, source value, comparison and uncertainty where the source gives them. The texts are the approved entries of the edition, as is its summary.

Scope and limitations of all 22 statements (A1–A22)

Edition summary

October’s evidence shows no broad employment collapse in AI-exposed occupations, but several mostly descriptive signs of weaker entry opportunities: in Germany full-time employment, and in France permanent hiring, of young people in highly exposed occupations developed less favourably, companies adopting AI shifted their workforce towards senior positions, and Danish AI users with fewer than 100 full-time equivalents hired less, while US data on recent graduates and UK business reports show little effect. Experiments show performance gains with AI support for IT apprentices and patent lawyers but no measurable time savings for teachers in Lima; without AI, only experienced patent lawyers benefited; surveys link AI use with better jobs, more responsibility and more frequent loneliness without establishing cause and effect. Causal evidence on wages and health is missing, as are October studies on social partners, health care, the creative industries and customer service; after an amendment in July 2026, the AI Act’s high-risk obligations for employment systems apply only from December 2027.

  1. A1Back to chapter 2.0

    Evidence: observed · baseline · Source 13

    Scope
    Representative survey of workers in all EU Member States (AIM-WORK, the Commission’s DG Employment and Joint Research Centre, 2024–2025); Commission analysis by three job-quality clusters; covers conscious use of AI tools, mainly chatbots.
    Limitations
    Cross-sectional data showing the distribution of use, not an effect of AI on job quality. Possible strains (cognitive offloading, fewer learning opportunities for entry-level workers) are discussed in the same chapter mainly on the basis of other studies; these were not adopted here as findings. Sample size not stated in the chapter.
    Observation period
    Fieldwork 2024 to 2025
    Source value
    35.3 Percent of workers in better-quality jobs using AI at work
    Baseline
    EU average: 29.9%
    Comparison
    Lower earnings and security: 16.9%; atypical and longer working times: 19.3%
    Uncertainty
    No uncertainty reported in the source

  2. A2Back to chapter 2.0

    Evidence: observed · baseline · Source 9

    Scope
    IAB Establishment Panel, survey in the second half of 2025, firm responses; occupational AI exposure derived from firms’ occupational mix in mid-2022 following Eloundou et al. (2024), grouped from lowest to highest exposure; about two thirds of all firms neither used nor planned to use AI.
    Limitations
    Firm self-reports; exposure describes the technical potential of the occupational mix, not suitability in the individual firm. According to the author, agentic AI systems have broadened possible uses since the survey, so assessments may change. Sample size not stated in the article; IAB-Forum article without peer review.
    Observation period
    Second half of 2025
    Source value
    41 Percent of firms with the highest occupational AI exposure that used AI
    Baseline
    Firms with the lowest exposure: 13%
    Comparison
    Seeing no possible application among firms neither using nor planning AI: 85% at highest and 91% at lowest exposure
    Uncertainty
    No uncertainty reported in the source

  3. A3Back to chapter 2.0

    Evidence: observed · baseline · Source 14

    Scope
    Audit of 33 existing judge configurations from six model families on 4,501 test ratings; the reference is a US survey of 45,796 ratings by participants with experience in the chosen occupation (recruited via Prolific); acceptance threshold 7 on a nine-point scale.
    Limitations
    English-language text responses for US occupations only; the reference comes from online respondents and is not representative of all workers. It measures the validity of model judgements, not actual automation or employment effects. Preprint under review.
    Observation period
    Survey data snapshot of 18 August 2026
    Source value
    61.1 Percent of AI responses rated acceptable by workers in the occupation
    Baseline
    Worker judgement (threshold 7 of 9)
    Comparison
    Language-model judges: 3.0% to 97.9% (median 43.9%)
    Uncertainty
    No uncertainty reported for these shares in the source

  4. A4Back to chapter 3.0

    Evidence: observed · baseline · Source 8

    Scope
    Monthly employment data from the Federal Employment Agency, January 2021 to December 2025; on average about 22.3 million full-time employees subject to social insurance aged 20–64 (excluding civil servants, the self-employed, those only in marginal jobs and apprentices); AI exposure following Eloundou et al. (2024), mapped to KldB 2010 occupational groups; Poisson regressions also accounting for suitability for working from home and cyclical vulnerability.
    Limitations
    The authors regard the findings as purely descriptive. Accounting for working-from-home suitability and cyclical vulnerability, the negative association for 20–29-year-olds is not significant (about 1.8% less favourable per standard deviation of exposure in 2025), while the association with the economic downturn is more than twice as strong; it is significantly negative only in manufacturing and non-academic occupations. Exposure measures potential, not actual AI use; IAB-Forum article without peer review.
    Observation period
    January 2021 to December 2025 (2025 annual average versus 2022)
    Source value
    -3.8 Percent change in full-time employment of 20–29-year-olds in highly AI-exposed occupations, 2025 versus 2022
    Baseline
    2022 annual average
    Comparison
    Less exposed occupations: −2.6%
    Uncertainty
    No uncertainty given for the unadjusted changes; according to the source the adjusted association is not statistically significant

  5. A5Back to chapter 3.0

    Evidence: observed · baseline · Source 3

    Scope
    Administrative labour-movement data (Dares, MMO) for the French private sector excluding agriculture, temporary agency work and private households, January 2019 to March 2026; exposure index built from more than 1,500 France Travail occupation descriptions rated by a language model; plus about 6,000 firms reporting their AI use in early 2026.
    Limitations
    Descriptive analysis without causal identification: monetary tightening, the post-pandemic correction in IT and other factors affect the same occupations. The index measures technical substitution or assistance potential, not actual use. Firms with intensive AI use hired relatively fewer young people on permanent contracts, but their total employment did not diverge clearly. Working paper without peer review; the PDF states only “October 2026”, and the date follows the companion Banque de France note.
    Observation period
    January 2019 to March 2026 (reference month November 2022)
    Source value
    -28 Percent relative to November 2022 (CDI hires of 21–29-year-olds in the most exposed occupation quintile, seasonally adjusted and smoothed)
    Baseline
    November 2022 level
    Comparison
    Least exposed quintile: about −8%
    Uncertainty
    No uncertainty reported in the source for these descriptive series

  6. A6Back to chapter 3.0

    Evidence: observed · baseline · Source 19

    Scope
    Statistics Denmark IT usage survey (about 4,000 firms with at least 10 employees) linked to monthly employer–employee register data; each firm compared with its own pre-2023 employment trend; overall, adopters fell about 11% behind comparable non-adopters by late 2025, driven entirely by firms with fewer than 100 full-time equivalents; Danmarks Nationalbank working paper.
    Limitations
    The authors explicitly do not claim causal effects, as AI use and employment are likely jointly determined by firm strategy; they also note that AI-exposed occupations began deteriorating before ChatGPT. So far hardly visible economy-wide; with annual balance-sheet data the effects would not be detectable (confidence band roughly −8% to +5%). Working paper without peer review.
    Observation period
    2022 to late 2025 (2023 adoption cohort)
    Source value
    25 Percent below trend in employment in AI-exposed occupations at AI adopters (late 2025)
    Baseline
    Own pre-2023 employment trend
    Comparison
    Other occupations: 7% below trend; all occupations combined: about 11% behind comparable non-adopters
    Uncertainty
    Significant according to the source; confidence intervals shown only in figures

  7. A7Back to chapter 3.0

    Evidence: observed · baseline · Source 18

    Scope
    Revelio Labs data with 1.25 billion job postings and 154 million employment records, monthly panel January 2021 to March 2026; AI adoption inferred from job postings involving generative AI; matched control affiliates in the same country and sector, instrumented with AI adoption by other parent companies in the headquarters’ metropolitan area.
    Limitations
    Based on online profiles and postings from one data provider, with coverage varying by country; identification assumes the instrument works only through AI adoption. Total employment rose with only marginal significance (3.3%); in country-by-country estimates the junior-share decline was significant in only seven of 31 countries. Working paper without peer review.
    Observation period
    January 2021 to March 2026
    Source value
    -1.9 Percentage points of junior positions in employment (March 2026)
    Baseline
    Comparable control affiliates without AI adoption
    Comparison
    Senior employment: +6.7%; junior employment: −2.5% (not significant)
    Uncertainty
    Confidence intervals shown only in the source’s figures; not reproduced here

  8. A8Back to chapter 4.0

    Evidence: observed · baseline · Source 4

    Scope
    Current Population Survey (CPS) microdata, June–August 2026 compared with summers 2022–2025; difference-in-differences and event-study models with older college graduates (aged 30–49) and young non-graduates as comparison groups; US only.
    Limitations
    Measures unemployment, not hiring or AI use; short period. An expanded measure including people outside the labour force who want a job was the highest of the five summers in 2026 at 10.4%, but did not show a significant increase either. Working paper without peer review; first published on 28 September 2026 according to NBER metadata.
    Observation period
    June–August 2026 (comparison: summers 2022–2025)
    Source value
    7.3 Percent unemployment rate (June–August 2026 average)
    Baseline
    Summer 2022: 7.1%; 2023: 6.3%; 2024: 7.8%; 2025: 7.2%
    Comparison
    Older college graduates (aged 30–49): 2.8% (summer 2025: 2.7%)
    Uncertainty
    Summer 2026 coefficients in the DiD and event-study models not statistically significant

  9. A9Back to chapter 4.0

    Evidence: observed · baseline · Source 5

    Scope
    Business Insights and Conditions Survey (BICS), wave 165, fieldwork 21 September to 4 October 2026, 9,933 responses (25.7% response rate); weighted estimates; base: AI-using businesses with 10 or more employees.
    Limitations
    Business self-reports, not measured employment change and no counterfactual without AI; official statistics in development from a voluntary survey; shares apply only to AI-using businesses. Naming a role as “most affected” does not mean jobs were cut there.
    Observation period
    September 2026 (fieldwork 21 September to 4 October 2026)
    Source value
    6 Percent of AI-using businesses with 10 or more employees
    Baseline
    All AI-using businesses with 10 or more employees
    Comparison
    Fewer than 1% reported more staff as a result of adopting AI
    Uncertainty
    Weighted estimate; no confidence intervals in the bulletin text

  10. A10Back to chapter 4.0

    Evidence: observed · baseline · Source 10

    Scope
    Qualitative reports from businesses and other contacts of the twelve Federal Reserve Banks, collected up to 5 October 2026; District reports include banking contacts with lower staffing needs due to automation and AI, IT staffing contacts reporting less job creation, and one accounting contact saying that some large accounting firms were delaying contracts for entry-level hires.
    Limitations
    Contacts are not selected at random; qualitative assessments with no count of affected jobs. The summary does not name the five Districts; IT staffing contacts explicitly said AI was not causing layoffs. The Beige Book does not represent the views of Federal Reserve officials.
    Observation period
    Information collected up to 5 October 2026 (changes since the previous report)
    Source value
    5 Federal Reserve Districts (of 12) reporting AI-related lower staffing, delayed hiring or less expected job creation
    Baseline
    All 12 Districts
    Comparison
    Overall employment: 3 Districts modest, 5 slight increases, 4 no significant change
    Uncertainty
    Qualitative contact reports; no uncertainty measure

  11. A11Back to chapter 4.0

    Evidence: observed · baseline · Source 6

    Scope
    Cambodia, Malaysia, Mongolia, the Philippines and Thailand; Lightcast job postings from the first quarter of 2018 to the second quarter of 2025; event study relative to the quarter before ChatGPT's release (third quarter of 2022), occupations above versus below the median of an AI automatability index; method following Huang et al. (2026).
    Limitations
    The World Bank stresses that the estimates cannot be interpreted as causal. Online postings cover only part of hiring, especially where the informal sector is large. The comparison figure for high-income countries (−5.8%) comes from a cited study and was not checked independently here.
    Observation period
    Q1 2018 to Q2 2025

  12. A12Back to chapter 4.0

    Evidence: observed · baseline · Source 17

    Scope
    Lightcast online job-posting data for US manufacturing with advertised wages, through July 2026; AI skills defined following Acemoglu et al. (2022); most recently 11% of manufacturing postings required AI-related skills (all sectors: 8%); production occupations (SOC 51) shown separately.
    Limitations
    Descriptive, unadjusted wage differences; according to the source they may reflect differences in position type, location and employer, and the small number of AI postings is volatile. Postings show required rather than actual skills and only positions advertised online; FEDS Notes present the author’s own views.
    Observation period
    Posting sample period through July 2026
    Source value
    70 Percent higher advertised wages in manufacturing postings requiring AI skills than in all manufacturing postings (period average)
    Baseline
    All US manufacturing online job postings
    Comparison
    Production occupations: about 30% since 2023
    Uncertainty
    No uncertainty reported; according to the source, substantial volatility due to small numbers

  13. A13Back to chapter 5.0

    Evidence: observed · baseline · Source 1

    Scope
    Twelve vocational schools in six German states; IT specialists in application development and system integration; individual random assignment to ChatGPT Pro with internet access or an exam-like condition without internet; occupation-specific tasks during school hours.
    Limitations
    The comparison is with a no-internet condition, not with internet search alone; the authors note the effect may be overstated relative to normal working conditions. Test tasks rather than on-the-job work; open answers were graded by a language model against a fixed rubric. Nothing on employment, wages or long-term skill development; working paper without peer review.
    Observation period
    October 2025 to January 2026
    Source value
    30.6 Percentage points (share of points achieved, task “Beyond 2” beyond the curriculum)
    Baseline
    No-internet control group: 42.3% of attainable points
    Comparison
    Group with ChatGPT access (intention-to-treat, with controls)
    Uncertainty
    Standard error 2.0 percentage points (heteroskedasticity-robust); p < 0.01

  14. A14Back to chapter 5.0

    Evidence: observed · baseline · Source 16

    Scope
    Random assignment of access to a custom AI assistant that was not released as a standalone product; benchmark drafting tasks after 10 and 90 days, then redlining a patent application without AI; scoring by blinded expert patent attorneys; study period May 2025 to February 2026.
    Limitations
    Small sample, and the junior subgroup is smaller still, so its estimates are less precise; the firms work for a sophisticated client and are not representative. No baseline skills assessment; only three months. The study was conducted and funded by Google (possible conflict of interest); working paper without peer review.
    Observation period
    May 2025 to February 2026
    Source value
    0.38 Standard deviations higher quality on benchmark drafting tasks after 90 days
    Baseline
    Control group without the AI assistant
    Comparison
    After 10 days: 0.34; review task without AI: 0.32 overall, 0.45 among experienced lawyers
    Uncertainty
    p = 0.01 (90 days), p = 0.03 (10 days)

  15. A15Back to chapter 5.0

    Evidence: observed · baseline · Source 2

    Scope
    390 public schools in Metropolitan Lima (196 treatment, 194 control), school year 2025; time use from an endline survey (607 respondents), prompt quality graded with two rubrics (562 observations); tool: Microsoft Copilot.
    Limitations
    Time data are self-reported (duration of the most recent instance) and available for just under half of the teachers; pupil outcomes were not measured. The study was funded in part by the Microsoft AI Economy Institute and examined Microsoft Copilot (possible conflict of interest). Working paper without peer review.
    Observation period
    School year 2025 (programme period of about 28 weeks; endline survey in December 2025)
    Source value
    0.375 Points on a 5-point prompt-quality rubric (designed independently of the training)
    Baseline
    Control group: 3.374 points
    Comparison
    Treatment group (intention-to-treat)
    Uncertainty
    Standard error 0.086; p < 0.01

  16. A16Back to chapter 5.0

    Evidence: observed · baseline · Source 7

    Scope
    More than 10,000 developers and about 400,000 code changes at Meta; character-level provenance tracking of suggestions from the code-completion tool CodeCompose to landed code; about two months of data from spring 2025; peer-reviewed conference paper (ESEM 2026).
    Limitations
    Observational data without random assignment; the authors warn that naive comparisons attribute the context of AI use to AI itself and do not treat the associations as evidence of a causal effect. Provenance data were available for only about half of the changes in the period; completion-based AI only, before agentic tools; one company. The authors are Meta employees studying an in-house tool (possible conflict of interest).
    Observation period
    Spring 2025 (about two months)

  17. A17Back to chapter 5.0

    Evidence: observed · baseline · Source 12

    Scope
    Passively collected behavioural data (Microsoft Viva Insights) on 11,868 knowledge workers in one business unit, including 1,992 sustained users above a minimum threshold, September 2024 to September 2025; staggered difference-in-differences and instrumental-variable models.
    Limitations
    One business unit of one company; averaged over all event times and cohorts, the effect on collaboration time is not significant (0.95 hours; 95% confidence interval −0.19 to 2.08). Conversation content is not observed, so explanations such as checking AI outputs, discussing the tool or reinvesting saved time remain open; only the first three months; use was not randomly assigned. Preprint without peer review.
    Observation period
    September 2024 to September 2025
    Source value
    1.43 Hours of self-initiated collaboration time per month (average of the first three months after adoption)
    Baseline
    Non-users in the same business unit (9,876 employees)
    Comparison
    Users after adoption (staggered difference-in-differences estimate)
    Uncertainty
    95% confidence interval 0.45 to 2.41 hours

  18. A18Back to chapter 6.0

    Evidence: observed · baseline · Sources 21, 22

    Scope
    Official Journal texts of Regulation (EU) 2024/1689 (Article 26(7), Annex III point 4, Chapter III Section 3) and the amending Regulation (EU) 2026/1744 of 8 July 2026 (new Article 4, amended Article 113); directly applicable in all EU Member States.
    Limitations
    Legal position, not a measured effect; whether and how employers support AI literacy and inform workers is not documented. Under the Regulation, the information duty follows Union and national rules on informing workers; national co-determination rules were not examined here. Older than the reporting period and included as context only.
    Observation period
    Legal position as of October 2026 (amending Regulation published in the Official Journal on 24 July 2026)

  19. A19Back to chapter 6.0

    Evidence: observed · baseline · Source 11

    Scope
    Year two of the American Job Quality Study by Gallup with Jobs for the Future and The Families & Workers Fund; web and mail survey from 26 January to 24 March 2026; 15,482 employees aged 18–75 from the Gallup Panel and an address-based random sample, weighted; effects reported by 7,845 employees who had used AI in their job at least a few times in the past year.
    Limitations
    Self-assessments, not measured productivity; the source notes employers might judge the effects differently. More responsibility can mean upgrading or work intensification, and the survey does not distinguish them. Frequent users more often hold quality jobs (52% versus 32% of non-users); cause and effect remain open. The article now links to the full year-two report, which was not checked here.
    Observation period
    26 January to 24 March 2026
    Source value
    63 Percent of AI-using US employees reporting faster task completion
    Baseline
    All employees who used AI in the past year
    Comparison
    Regular users: 79%; irregular users: 38%
    Uncertainty
    Margin of sampling error ±2 percentage points (95% confidence) for AI users

  20. A20Back to chapter 6.0

    Evidence: observed · baseline · Source 13

    Scope
    AIM-WORK survey 2024–2025 in all EU Member States, Commission analysis by three job-quality clusters; in the cluster with atypical working times (including drivers and sales workers), algorithmic evaluation affected 48.3% in Spain and 47.4% in Portugal.
    Limitations
    Cross-sectional; algorithmic management also covers systems without AI. The Commission refers to other studies for possible links with stress, work pace and reduced autonomy, which are not measured here. Platform workers are not shown separately.
    Observation period
    Fieldwork 2024 to 2025
    Source value
    28.2 Percent of workers subject to algorithmic evaluation (EU average)
    Baseline
    EU average
    Comparison
    Jobs with atypical and longer working times: 31.9%
    Uncertainty
    No uncertainty reported in the source

  21. A21Back to chapter 6.0

    Evidence: observed · baseline · Source 20

    Scope
    US Census Bureau Household Trends and Outlook Pulse Survey, 13–30 March 2026; 6,238 employed adults, weighted to 152 million; AI use measured as the number of eleven possible assisted work tasks, standardised; models adjusted for sociodemographics, region, occupation and telework.
    Limitations
    Cross-sectional, no causal conclusion; reverse causation is possible, for example if lonely workers turn to AI more often. Self-reports; workplace social interaction and work intensity were not measured.
    Observation period
    13–30 March 2026
    Source value
    1.13 Adjusted prevalence ratio for feeling lonely per standard deviation more AI use
    Baseline
    Workers with one standard deviation less AI use (model comparison)
    Comparison
    Life satisfaction: +0.11 standard deviations (95% confidence interval 0.06 to 0.16)
    Uncertainty
    95% confidence interval 1.02 to 1.25

  22. A22Back to chapter 6.0

    Evidence: observed · baseline · Source 15

    Scope
    Survey in the second half of 2025 as part of an ILO diagnostic study on working conditions and social security in platform work; young workers made up 42.9% of respondents and 75.7% of delivery riders.
    Limitations
    According to the ILO, the sample is not nationally representative; self-reports. The study measures neither algorithmic management nor the platforms’ use of AI; it describes working conditions in app-based platform work; no comparison group outside platform work.
    Observation period
    Second half of 2025
    Source value
    27.1 Percent of young platform workers reporting a work-related injury
    Baseline
    Older platform workers (aged 31 and over): 19.3%
    Comparison
    Young delivery riders: 36.3%
    Uncertainty
    No uncertainty reported in the source

And in your case? Check it with sources.

Start an analysisFor organizations
All reports