CoChat
Sign Up
Cited by 0Labor market dynamics and wage inequalityEmployment and Welfare StudiesDigital Economy and Work Transformation

Large language model exposure and precarious occupations: Unpacking relationships in the Canadian labor force

Open access
Jetha, Arif, Liao, Qing, Smith, Peter, Vu, Viet, Biswas, Aviroop, Smith, Brendan, Vahid Shahidi, Faraz·2026·Scandinavian Journal of Work, Environment & Health
DOI 10.5271/sjweh.4312PMC PMC13555092
Open this paper in CoChat — free accountOpen PDF

Abstract

OBJECTIVE: The adoption of digital technologies has historically impacted the most precarious occupations and contributed to widening labor market inequities. Large language models (LLM) may reshape this relationship. This study examines the association between occupational exposure to LLM and occupational precarity. METHODS: Using Canada’s Labour Force Survey, occupational exposure to LLM and four dimensions of precarity (contractual instability, earnings inadequacy, schedule unpredictability, working-time mismatch) were examined. A multidimensional index was developed to summarize an occupation’s overall exposure to precarity. Four multivariate linear regression models with cluster-robust standard errors estimated the associations between LLM exposure and each dimension of precarity. A fifth multivariate model examined the relationship between LLM exposure and the multidimensional precarity index. Utilizing model coefficients, mean estimates of occupational LLM exposure were produced. RESULTS: Using the multidimensional precarity index, our analysis showed that occupations characterized by low exposure to precarity had a significantly higher mean LLM exposure [mean 0.386, 95% confidence interval (CI) 0.356–0.417] compared to occupations with medium (mean 0.258, 95% CI 0.221–0.295), high (mean 0.260, 95% CI 0.194–0.328) or very high precarity (mean 0.205, 95% CI 0.136–0.275). Apart from earning adequacy, LLM exposure was also lower among occupations using each separate dimension of precarity. CONCLUSION: Occupations most likely to be exposed to LLM are those where precariousness is lowest. These occupations have previously been sheltered from technological change. There is a need of examine the impacts of LLM on workers in job where the technology is prominent.

AI snapshot

Written from the paper’s full text.

TL;DR

In a pooled cross-sectional analysis of Canada’s Labour Force Survey (2021–2024), occupations with lower precarity had higher occupational exposure to large language models (LLM). The most precarious occupations were least exposed, reversing the pattern seen in some prior technological-change waves.

Key findings

  • Using the multidimensional precarity index, adjusted mean LLM exposure was 0.386 (95% CI 0.356–0.417) for low-precarity occupations, compared with 0.258 (95% CI 0.221–0.295) for medium, 0.260 (95% CI 0.194–0.328) for high, and 0.205 (95% CI 0.136–0.275) for very high precarity; medium, high, and very high did not differ significantly from each other.
  • The overall mean occupational LLM exposure value was 0.34 (minimum 0; maximum 0.84) across the four LFS survey years.
  • In adjusted models, the highest-precarity quartile had significantly lower LLM exposure for temporary employment (β=-0.080, P<0.05), irregular work hours (Q3 β=-0.150, Q4 β=-0.240, both P<0.001), and involuntary part-time work (Q4 β=-0.100, P<0.01). Low-wage Q4 was not significant after adjustment (β=0.059, P=0.395), though it was significant in the univariate model (β=-0.143, P<0.001).
  • For the multidimensional index, adjusted coefficients relative to low precarity were medium β=-0.130 (P<0.001), high β=-0.130 (P<0.01), and very high β=-0.180 (P<0.001).

Methods

  • Pooled cross-sectional analysis of Statistics Canada’s Labour Force Survey, 2021–2024; a nationally representative monthly survey of Canada’s household population aged 15+ with approximately 100,000 workers. Analyses were conducted at the occupational level (512 occupations after exclusions), excluding Canadian Territories and several groups such as non-employed, self-employed, armed forces, and institutional residents.
  • Precarity was measured across four dimensions: temporary employment, low wages (less than two-thirds of median hourly wage), irregular work hours, and involuntary part-time work (<30 hours/week while wanting ≥30 hours/week). Occupations were ranked into quartiles of exposure, and a multidimensional index counted how many dimensions placed an occupation in the highest quartile (low=0, medium=1, high=2, very high=3–4). LLM exposure used Eloundou et al.’s measure, crosswalked from O*NET to Canadian NOC codes, as a continuous 0–1 score reflecting the share of tasks exposed to LLM.
  • Descriptive analyses, correlations, univariate models, and five multivariate linear regression models with cluster-robust standard errors were estimated; models adjusted for occupational gender, age, educational attainment, provincial, and industry composition. Sensitivity analysis reclassified precarity using tertiles. Analyses used SAS version 9.4.

Limitations

  • The paper states that LLM exposure is a proxy based on task content and cannot determine realized exposure or recent technological advancements; it focuses on software/LLM and not robotics or the broader range of AI systems; it does not capture new tasks, productivity benefits or risks, or changes to employment outcomes; occupational-level exposure may miss within-occupation heterogeneity and may introduce ecological bias if individual inferences are drawn; the LFS limited the precarity dimensions available; and findings are a snapshot at one point in time.
  • Evident from the design: the pooled cross-sectional, occupational-level analysis cannot establish causal direction or temporal ordering between LLM exposure and occupational precarity.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.5271/sjweh.4312

Cites

28 of 28 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

18 of 69 cited sources checked · 6 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    The precariousness of an occupation is a critical determinant to the health and safety of workers.

    “The precariousness of an occupation is a critical determinant to the health and safety of workers (1–3).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [1] Benach J, Vives A, Amable M, Vanroelen C, Tarafa G, Muntaner C. Precarious employment: understanding an emerging social determinant of health. Annu Rev Public Health 2014; 35: 229– 53. 10.1146/annurev-publhealth-032013-182500 24641559 source
    • cites [2] Hajat A, Andrea SB, Oddo VM, Winkler MR, Ahonen EQ. Ramifications of precarious employment for health and health inequity: emerging trends from the Americas. Annu Rev Public Health 2024 May; 45(1): 235– 51. 10.1146/annurev-publhealth-071321-042437 38012123 PMC11128534 source
    • cites [3] Shahidi FV, Liao Q, Landsman V, Mustard C, Robson LS, Biswas A et al. Is precarious employment an occupational hazard? Evidence from Ontario, Canada. Occup Environ Med 2024 Aug; 81(8): 381– 7. 10.1136/oemed-2024-109535 39095092 source
  • Premisecitation

    Being employed in a precarious occupation can indirectly interrupt access to safe housing, nutritious food and other resources known to shape health.

    “Indirectly, being employed in a precarious occupation can interrupt access to safe housing, nutritious food and other resources known to shape health (2).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [2] Hajat A, Andrea SB, Oddo VM, Winkler MR, Ahonen EQ. Ramifications of precarious employment for health and health inequity: emerging trends from the Americas. Annu Rev Public Health 2024 May; 45(1): 235– 51. 10.1146/annurev-publhealth-071321-042437 38012123 PMC11128534 source
  • Premisecitation

    Higher wages are associated with greater occupational exposure to LLM.

    “Our univariate findings align with past studies showing a relationship between higher wages and greater occupational exposure to LLM (18, 22).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source
    • cites [22] Azuara Herrera O, Ripani L, Torres Ramirez E. AI and the increase of productivity and labor inequality in Latin America: potential impact of large language models on Latin American workforce. Washington (DC): Inter-American Development Bank; 2024. Contract No.: 13 April 2024. source
  • Premisecitation

    There is a small body of research which signal an inverse relationship between occupations that are precarious as well as exposure to LLM.

    “There is a small body of research which signal an inverse relationship between occupations that are precarious as well as exposure to LLM (ie, where LLM can be used to perform tasks that make up an occupation).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source
    • cites [22] Azuara Herrera O, Ripani L, Torres Ramirez E. AI and the increase of productivity and labor inequality in Latin America: potential impact of large language models on Latin American workforce. Washington (DC): Inter-American Development Bank; 2024. Contract No.: 13 April 2024. source
  • Interpretationcitation

    Over time, the uneven exposure of occupations to LLM and other emerging AI systems may contribute to labor market inequities and health disparities.

    “Over time, the uneven exposure of occupations have to LLM and other emerging AI systems may contribute to labor market inequities and health disparities (18, 42).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source
    • cites [42] Teutloff O, Einsiedler J, Kässi O, Braesemann F, Mishkin P, del Rio-Chanona RM. Winners and losers of generative AI: early evidence of shifts in freelancer demand. J Econ Behav Organ 2025; 106845: 10.1016/j.jebo.2024.106845 source

      The discussion of findings (p. 3) says AI can create labor-market winners and losers depending on workers’ skills, supporting the claim of uneven labor-market effects. The conclusion (p. 17) says long-term effects remain to be studied. The passages do not address health disparities, so that part of the claim is unsupported by this source. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • May combine two claims. 21
  • Classified as probably not a substantial claim. 33

  • May combine two claims.

    Premisecitation

    Workers in precarious occupations are more likely to face hazardous working conditions and greater injury risk compared to those working in occupations with standard work arrangements.

    “When compared to those working in occupations with standard work arrangements (ie, full-time and permanent jobs), workers in precarious occupations are more likely to face hazardous working conditions and greater injury risk (2, 9).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [2] Hajat A, Andrea SB, Oddo VM, Winkler MR, Ahonen EQ. Ramifications of precarious employment for health and health inequity: emerging trends from the Americas. Annu Rev Public Health 2024 May; 45(1): 235– 51. 10.1146/annurev-publhealth-071321-042437 38012123 PMC11128534 source
    • cites [9] Peckham T, Seixas N, de Castro AB, Hajat A. Do different patterns of employment quality contribute to gender health inequities in the US? A cross-sectional mediation analysis. Int J Environ Res Public Health 2022 Sep; 19(18): 11237. 10.3390/ijerph191811237 36141509 PMC9517248 source

      Section 1.1 says precarious employment is theorized to affect health through exposure to hazardous working conditions. It does not establish that precarious workers are more likely than workers in standard arrangements to face those conditions, and the passages found do not report greater injury risk. (LLM, conf=0.85)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Studies conducted in high income countries document that a growing number of workers are employed in occupations which are characterized as precarious.

    “Studies conducted in high income countries document that a growing number of workers are employed in occupations which are characterized as precarious (5–8).”
    Quote in the paperNot a substantial claim1 of 4 sources checked
    • cites [5] Dorn D. The rise of the machines: how computers have changed work. UBS Center Public Paper Series. 2015(4).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [6] Goos M, Manning A, Salomons A. Explaining job polarization: routine-biased technological change and offshoring. Am Econ Rev 2014; 104(8): 2509– 26. 10.1257/aer.104.8.2509 source

      Section II, “Job Polarization” (page not available), reports rising employment shares in some low-paid service and elementary occupations across 16 Western European countries. The paper classifies those occupations by pay, not as precarious, so it does not support that characterization. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [7] Muntaner C. Digital platforms, gig economy, precarious employment, and the invisible hand of social class. Int J Health Serv 2018 Oct; 48(4): 597– 600. 10.1177/0020731418801413 30213247 source
    • cites [8] Lewchuk W. Precarious jobs: where are they, and how do they affect well-being? Econ Labour Relat Rev 2017; 28(3): 402– 19. 10.1177/1035304617722943 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    A study from Japan found that the use of robots to perform job tasks within the nursing home sector alleviated demands on full-time nursing staff, but simultaneously increased the employer's preference for flexible, non-regular and part-time nurses.

    “As an example, a study from Japan found that the use of robots to perform job tasks within the nursing home sector alleviated demands on full-time nursing staff, but simultaneously increased the employer’s preference for flexible, non-regular and part-time nurses (15).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [15] Eggleston K, Lee Y, Iizuka T. Robots and labor in the service sector: evidence from nursing homes. NBER Working Paper 28322 (2021). 10.3386/w28322 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Financial insecurity and economic unpredictability faced by workers in precarious occupations are associated with greater stress and anxiety and working conditions, which are harmful to worker physical and mental health.

    “Also, research highlights that financial insecurity and economic unpredictability faced by workers in precarious occupations are associated with greater stress and anxiety and working conditions, which are harmful to worker physical and mental health (3, 10–12).”
    Quote in the paperNot a substantial claim3 of 4 sources checked
    • cites [3] Shahidi FV, Liao Q, Landsman V, Mustard C, Robson LS, Biswas A et al. Is precarious employment an occupational hazard? Evidence from Ontario, Canada. Occup Environ Med 2024 Aug; 81(8): 381– 7. 10.1136/oemed-2024-109535 39095092 source
    • cites [10] Shahidi FV, Andreacchi AT, Fuller AE, Blair A, Carnide N, Harris MA et al. Employment quality and mortality in Canada. J Epidemiol Community Health 2025 Dec; 80(1): 27– 34. 10.1136/jech-2025-224434 41125335 PMC12703285 source

      The Methods section describes precarious employment as insecure, unstable and low-paid. The Introduction links non-standard employment to poorer general and mental health, including psychological distress, and the Discussion reports higher mortality risks for workers in precarious employment. The paper does not specifically establish that financial unpredictability is associated with anxiety or explain stress as the pathway to poorer health. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [11] Bonney T, Rospenda KM, Forst L, Conroy LM, Castañeda D, Avelar S et al. Employment precarity and increased risk of hazardous occupational exposures among residents of high socioeconomic hardship neighborhoods. Ann Work Expo Health 2022 Nov; 66(9): 1122– 35. 10.1093/annweh/wxac062 36000755 PMC9746339 source

      The Methods describe income variability and uncertain hours and scheduling as measures of employment precarity. The Discussion and Conclusions link precarity to hazardous working conditions and adverse health outcomes. The passages do not establish the claimed association with greater stress and anxiety or specifically assess physical and mental health outcomes. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites [12] Oddo VM, Mabrouk S, Andrea SB, Ahonen EQ, Winkler MR, Vignola EF et al. The association between precarious employment and stress among working aged individuals in the United States. Prev Med 2024 Oct; 187: 108123. 10.1016/j.ypmed.2024.108123 39216552 PMC11700481 source

      The Discussion (page not provided) reports a positive association between precarious employment and stress, and the Introduction describes low wages and unpredictable hours as features of precarious work. The Methods (page not provided) show that the study measured perceived stress and CRP, not anxiety or workers’ physical and mental health outcomes. The paper discusses potential health harms but does not establish those additional associations in its study. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    The automation of work was most prominent among occupations with repetitive job tasks and low skill requirements, which have been amongst the most precarious.

    “The automation of work was most prominent among occupations with repetitive job tasks and low skill requirements, which have been amongst the most precarious (6).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [6] Goos M, Manning A, Salomons A. Explaining job polarization: routine-biased technological change and offshoring. Am Econ Rev 2014; 104(8): 2509– 26. 10.1257/aer.104.8.2509 source

      The paper’s introductory discussion says technological change replaces labor in routine tasks. But it identifies the resulting decline as concentrated in middling occupations relative to high- and low-skilled ones; Section II reports declines in occupations paying around the median wage and increases in some low-paid occupations. The cited passages do not characterize the affected occupations as the most precarious. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    Research conducted during past periods of technological change has indicated that the adoption of new technologies has been associated with increases in precariousness reported by workers in occupations facing the lowest quality employment conditions.

    “Research conducted during past periods of technological change have indicated that the adoption of new technologies has been associated with increases in precariousness reported by workers in occupations facing the lowest quality employment conditions (5, 6).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [5] Dorn D. The rise of the machines: how computers have changed work. UBS Center Public Paper Series. 2015(4).

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [6] Goos M, Manning A, Salomons A. Explaining job polarization: routine-biased technological change and offshoring. Am Econ Rev 2014; 104(8): 2509– 26. 10.1257/aer.104.8.2509 source

      The paper’s Conclusions (page unavailable in our copy) link routine-biased technological change to job polarization, including a rising employment share for low-paid personal service workers. But the paper analyzes occupational employment shares, not workers’ reports of increased precariousness or changes in employment conditions. Its findings therefore support the technological-change context, but not the claim’s central outcome. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    LLM can be used to summarize and synthesize vast amounts of information, manage customer inquiries, support organizational decision-making and generate technical and creative content.

    “LLM can be used to summarize and synthesize vast amounts of information (eg, legislative analysis tools), manage customer inquiries (eg, chatbot customer service applications), support organizational decision-making (eg, resume screening and job matching platforms) and generate technical and creative content (eg, content ideation tools) (17–21).”
    Quote in the paperNot a substantial claim1 of 5 sources checked
    • cites [17] Brynjolfsson E, Li D, Raymond L. Generative AI at Work, National Bureau of Economic Research working paper 31161. 2023. source

      Section 1.2 describes LLM tools that summarize legal documents and research and perform coding, writing, and graphic-design tasks. Section 2.1 describes their use in customer support. The passages do not support the organizational decision-making example of resume screening and job matching. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source
    • cites [19] Gruzd A, Mai P, Clements Haines A. The state of generative AI use in Canada 2025: exploring public attitudes and adoption trends. Figshare. 2025. source
    • cites [20] Oschinski M, Walia R. Harnessing generative AI: navigating the transformative impact on Canada’s labour market 2025 [accessed 5 Jan 2026]. Available from: https://irpp.org/wp-content/uploads/2025/05/Harnessing-Generative-AI-Navigating-the-Transformative-Impact-on-Canadas-Labour-Market.pdf

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [21] Russell S, Norvig P. Artificial intelligence: a modern approach, 4th US ed. Berkeley (CA): University of California, Berkeley; 2022 [accessed 27 March 2026]. Available from: https://aima.cs.berkeley.edu/

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A study of LLM exposure in Latin American countries found that occupational LLM exposure was highest among workers with greater educational attainment and income.

    “Similarly, a study of LLM exposure in Latin American countries found that occupational LLM exposure was highest among workers with greater educational attainment and income (22).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [22] Azuara Herrera O, Ripani L, Torres Ramirez E. AI and the increase of productivity and labor inequality in Latin America: potential impact of large language models on Latin American workforce. Washington (DC): Inter-American Development Bank; 2024. Contract No.: 13 April 2024. source

  • May combine two claims.

    Study design and conductcitation

    The concordance between the standardized occupation codes used in the O*NET database and the NOC system used in the LFS was facilitated through Codage Assisté des Professions et Secteurs d'activité - Canada, which provided an actual concordance across coding systems for 350 of 500 NOC codes (one-to-many), and Statistics Canada, which provided a theoretical crosswalk for the remaining 162 NOC codes

    “The concordance between the standardized occupation codes used in the O*NET database and the NOC system used in the LFS was facilitated through Codage Assisté des Professions et Secteurs d’activité – Canada, which provided an actual concordance across coding systems for 350 of 500 NOC codes (one-to-many) (31), and Statistics Canada, which provided a theoretical crosswalk for the remaining 162 NOC”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [31] Rémen T, Richardson L, Pilorget C, Palmer G, Siemiatycki J, Lavoué J. Development of a coding and crosswalk tool for occupations and industries. Ann Work Expo Health 2018 Aug; 62(7): 796– 807. 10.1093/annweh/wxy052 29912270 PMC6093321 source

      The 'Official' links and 'Crosswalks between classification systems' sections describe CAPS-Canada links between NOC-2011 and US-SOC-2010, including one-to-many links; Table 4 lists 500 NOC-2011 codes (page not given in our copy). The paper does not substantiate the cited figure of 350 codes with an actual concordance. It also does not establish the statement about a Statistics Canada crosswalk for the remaining 162 codes, although the citation is attached to the CAPS-Canada portion of the sentence. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
    • cites [32] Statistics Canada. Correspondence: National Occupational Classification (NOC) 2016 version 1.3 to Standard Occupational Classification (SOC) 2018. Ottawa, ON: Statistics Canada; 2022.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

    Interpretationcitation

    The use of LLM may result in increased competition among workers in high quality (least precarious) occupations and, in the long term, may contribute to skill obsolescence, dehumanization of work, decreased wages, job insecurity, and polarization in the labor market.

    “The result may be an increased competition among workers in high quality (least precarious) occupations and, in the long-term, may contribute to skill obsolescence, dehumanization of work, decreased wages, job insecurity and polarization in the labor market (36, 37).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [36] Chen WX, Srinivasan S, Zakerinia S. Displacement or complementarity? The labor market impact of generative AI. USA: Harvard Business School; 2025.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [37] Ponce Del Castillo A. Generative AI, generating precariousness for workers? AI Soc 2024; 39(5): 2601– 2. 10.1007/s00146-023-01719-9 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    The final multivariate linear regression model found that occupations with low multidimensional precarity had a significantly higher adjusted mean LLM exposure (0.386, 95% CI 0.356-0.417) than occupations with medium (0.258, 95% CI 0.221-0.295), high (0.260, 95% CI 0.194-0.328), or very high (0.205, 95% CI 0.136-0.275) precarity, with no significant variation among the latter three.

    “We found that occupations characterized by low exposure to precarity had a significantly higher adjusted mean LLM exposure value (mean 0.386, 95% CI 0.0.356–0.417) when compared to occupations characterized by medium (mean 0.258, 95% CI 0.221–0.295), high (mean 0.260, 95% CI 0.194–0.328) or very high (mean 0.205, 95% CI 0.136–0.275) precarity where there was no significantly different variation in”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Automation technologies can improve working conditions for some workers while simultaneously deepening precarity for others.

    “Yet, evidence suggests a nuanced pattern; automation technologies can improve working conditions for some workers while simultaneously deepening precarity for others (14).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [14] Beneito P, García-Vega M, Vicente-Chirivella O, Wilemme G. Robots and firms’ labour search: the role of temporary work agencies. 2024. 10.2139/ssrn.5928552 source

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationreasoning

    Unlike previous waves of technological change which put pressure on workers in lower skilled occupations, the use of LLM within the workplace is affecting occupations with higher skill requirements whose job tasks may be codifiable.

    “Unlike previous waves of technological change which put pressure on workers in lower skilled occupations, the use of LLM within the workplace are affecting occupations with higher skill requirements whose job tasks maybe codifiable.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Occupational exposure to LLM was highest among occupations that were characterized by higher wages.

    “Occupational exposure to LLM was highest among occupations that were characterized by higher wages (one indicator of whether an occupation is precarious or not) (18).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    Using sample weights produced by Statistics Canada, weighted monthly LFS cycles were combined to produce annual estimates.

    “Using sample weights produced by Statistics Canada (24, 26), weighted monthly LFS cycles were combined to produce annual estimates.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [24] Statistics Canada. Labour force survey (Survey 3701): Statistics Canada; 2024 [accessed 05 Jan 2026]. Available from: https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3701

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [26] Statistics Canada. Methodology of the Canadian Labour Force Survey: Chapter 6 Weighting and estimation Ottawa: Statistics Canada; 2017 [accessed 13 April 2026]. Available from: https://www150.statcan.gc.ca/n1/pub/71-526-x/2017001/chapt6-eng.htm

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    We used the novel LLM exposure measure by Eloundou et al to examine the impact of the technology on occupations.

    “We used the novel LLM exposure measure by Eloundou et al (18) to examine the impact of the technology on occupations.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source

  • May combine two claims.

    Study design and conductown data

    Linearity was evaluated by plotting residuals against each continuous predictor and the fitted values; the lack of apparent nonlinear patterns and the approximately symmetric distribution of residuals around zero suggested that the linearity assumption was satisfied.

    “Linearity was evaluated by plotting residuals against each continuous predictor and the fitted values; the lack of apparent nonlinear patterns and the approximately symmetric distribution of residuals around zero suggested that the linearity assumption was satisfied.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    The LLM exposure measure was developed using the US O*NET 27.2, which is an American occupational information database describing nearly 1000 occupations and occupation-specific information via standardized job-oriented descriptors and worker-oriented descriptors.

    “The measure was developed using the US O*NET 27.2, which is an American occupational information database describing nearly 1000 occupations and occupation-specific information via standardized job-oriented descriptors and worker-oriented descriptors (30).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [30] O*NET. O*NET Online 2025 [13 April 2026]. Available from: www.onetonline.org

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationreasoning

    These occupations have previously been sheltered from technological change.

    “These occupations have previously been sheltered from technological change.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    O*NET describes 19 265 core or supplemental job tasks (where supplemental tasks were weighted half as much), associated with 2087 detailed work activities (DWA) across 923 standardized occupations (where a DWA may appear in multiple occupations).

    “O*NET describes 19 265 core or supplemental job tasks (where supplemental tasks were weighted half as much), associated with 2087 detailed work activities (DWA) across 923 standardized occupations (where a DWA may appear in multiple occupations).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [30] O*NET. O*NET Online 2025 [13 April 2026]. Available from: www.onetonline.org

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The occupational LLM exposure measure and analytical approach cannot tell us how workers within each occupation are affected by the technology.

    “The occupational LLM exposure measure and analytical approach we utilized cannot tell us how workers within each occupation are affected by the technology.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    The rubric was applied to each DWA and job task by a team of human labelers, and results were aggregated at the occupational level to produce an overall score representing the proportion of job tasks within an occupation that can achieve significant time savings by utilizing an LLM.

    “The rubric was applied to each DWA and job task by a team of human labelers. Results were subsequently aggregated at the occupational level to produce an overall score representing the proportion of job tasks within an occupation that can achieve significant time savings by utilizing an LLM.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source

  • May combine two claims.

    Premisecitation

    The LFS is a nationally-representative cross-sectional monthly survey of Canada's household population (15 years of age) with an approximate sample of 100 000 workers built using probability sampling.

    “The LFS is a nationally-representative cross-sectional monthly survey of Canada’s household population (15 years of age) with an approximate sample of 100 000 workers built using probability sampling (24).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [24] Statistics Canada. Labour force survey (Survey 3701): Statistics Canada; 2024 [accessed 05 Jan 2026]. Available from: https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3701

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    There was some evidence of models showing statistically significant departures from normality; visual inspection suggested that these deviations were minor.

    “There was some evidence of models showing statistically significant departures from normality; visual inspection suggested that these deviations were minor.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A great deal of attention has been directed towards examining and addressing precarious occupations because of their established relationship with adverse health and economic outcomes in the working population.

    “A great deal of attention has been directed towards examining and addressing precarious occupations because of their established relationship with adverse health and economic outcomes in the working population (33, 34).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [33] Ahonen E, Baron S, Brosseau L et al. Health and safety issues for workers in nonstandard employment. In: McQueen D, editor. Oxford Research Encyclopedia of Global Public Health. Oxford, England: Oxford University Press; 2018. source
    • cites [34] Kalleberg AL. Precarious lives: job insecurity and well-being in rich democracies: John Wiley & Sons;2018. PMC6703155 31435077

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    Using an approach developed by Shahidi et al, we examined four dimensions of precarity using items collected in the LFS: contractual instability, earnings inadequacy, schedule unpredictability and working-time mismatch.

    “Using an approach developed by Shahidi et al (3), we examined four dimensions of precarity using items collected in the LFS: contractual instability (ie, temporary employment), earnings inadequacy (ie, low wages), schedule unpredictability (ie, irregular hours) and working-time mismatch (ie, involuntary part-time employment).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] Shahidi FV, Liao Q, Landsman V, Mustard C, Robson LS, Biswas A et al. Is precarious employment an occupational hazard? Evidence from Ontario, Canada. Occup Environ Med 2024 Aug; 81(8): 381– 7. 10.1136/oemed-2024-109535 39095092 source

  • Classified as probably not a substantial claim.

    Premisecitation

    Eloundou et al created a rubric to assess occupational LLM exposure based on the share of an occupation's job tasks where access to an LLM alone or with a simple interface would lead to 50%-time savings and the share of an occupation's tasks where additional software is needed on top of an LLM to realize the 50%-time savings.

    “Using this definition, they created a rubric to assess occupational LLM exposure based on the share of an occupation’s job tasks where access to an LLM alone or with a simple interface would lead to 50%-time savings (based specifically on manual human coding) and the share of an occupation’s tasks where additional software is needed on top of an LLM to realize the 50%-time savings.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Potential outliers and influential observations were examined using leverage values and Cook's distance; although several observations were flagged for further inspection, manual review did not identify any extreme or unduly influential values, and no observations were excluded from the analyses.

    “Potential outliers and influential observations were examined using leverage values and Cook’s distance. Although several observations were flagged for further inspection, manual review did not identify any extreme or unduly influential values, and no observations were excluded from the analyses.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    Those most exposed to LLM could see productivity benefits.

    “Alternatively, those most exposed to LLM could see productivity benefits.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    Cluster-robust standard errors were employed for univariate and multivariate linear regression models to address non-constant error variance.

    “To address non-constant error variance, cluster-robust standard errors were employed for univariate and multivariate linear regression models.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    The LFS uses the National Occupational Classification (NOC; 2021), a standardized system to classify Canadian occupations based on the type of work performed, including main duties, educational requirements, and other information.

    “The LFS uses the National Occupational Classification (NOC; 2021), a standardized system to classify Canadian occupations based on the type of work performed, including main duties, educational requirements, and other information (25).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [25] Government of Canada. Employment and Social Development Canada. National Occupational Classification Ottawa: Government of Canada; 2025 [13 April 2026]. Available from: https://noc.esdc.gc.ca

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Eloundou and colleagues recently found that just under 20% of occupations in the US consisted of job tasks in which 50% are exposed to LLM.

    “Eloundou and colleagues (18) recently found that just under 20% of occupations in the US consisted of job tasks in which 50% are exposed to LLM.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] Eloundou T, Manning S, Mishkin P, Rock D. GPTs are GPTs: labor market impact potential of LLMs. Science 2024 Jun; 384(6702): 1306– 8. 10.1126/science.adj0998 38900883 source

  • May combine two claims.

    Interpretationreasoning

    When compared to past periods of technological change, LLM impact a different portfolio of occupations which are potentially associated with higher job quality and may contribute to emerging health and socioeconomic implications.

    “When compared to past periods of technological change, we show that LLM impact a different portfolio of occupations which are potentially associated with higher job quality and may contribute to emerging health and socioeconomic implications.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    As AI systems like LLM become more advanced, there could be a net increase in the number of workers in precarious occupations who may be exposed to conditions that impact health and safety.

    “As AI systems like LLM become more advanced in their predictive capabilities, there could be a net increase in the number of workers in precarious occupations in the labor market who may be exposed to conditions that impact health and safety.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    Although the measurement tool was designed in the US, its characterization of occupations according to the extent to which they can be affected by AI has been applied to labor market data in different countries.

    “Although the measurement tool was designed in the US, it’s characterization of occupations according to the extent to which they can be affected by AI has been applied to labor market data in different countries.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    Statistics Canada manages missing data prior to the dataset being accessed by individual research teams utilizing several imputation approaches including carry-forward, deterministic and donor imputation.

    “Statistics Canada manages missing data prior to the dataset being accessed by individual research teams utilizing several imputation approaches including carry-forward, deterministic and donor imputation (24).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] Statistics Canada. Labour force survey (Survey 3701): Statistics Canada; 2024 [accessed 05 Jan 2026]. Available from: https://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3701

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Those working in precarious occupations could be excluded from the productivity benefits associated with LLM, resulting in missing out on potential financial gains.

    “At the same time, those working in precarious occupations could be excluded from the productivity benefits associated with LLM that could result in missing out on potential financial gains that can accompany the use of AI at work.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The findings provide a snapshot regarding the impact of LLM on occupations at one point in time.

    “Lastly, our findings provide a snapshot regarding the impact of LLM on occupations at one point in time.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Utilizing the LFS meant that the study was limited in the availability of precarious occupation dimensions examined.

    “Utilizing the LFS meant that we were limited in the availability precarious occupation dimensions we examined.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    In contrast to past periods of change, a different portfolio of occupations may be most affected by LLM, creating novel challenges and opportunities for different groups of workers related to their employment in precarious occupations.

    “In contrast to past periods of change, we show that a different portfolio of occupations may be most affected LLM, creating novel challenges and opportunities for different groups of workers related to the extent to which they are employed in precarious occupations.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Younger workers, who are most likely to work precariously, are more likely to start their careers in occupations that are highly exposed to generative AI, which could contribute to declining rates of employment in this age group.

    “For instance, some early-stage research shows that younger workers, who are most likely to work precariously, are more likely to start their careers in occupations that are highly exposed generative AI which could contribute to declining rates of employment in this age group (35, 44).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [35] Gimbel M, Kinder M, Kendall J, Lee M. Evaluating the impact of AI on the labor market: Current state of affairs. The Budget Lab at Yale University. 2025.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [44] Brynjolfsson E, Chandar B, Chen R. Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford Digital Economy Lab. 2025. Available from: https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    The LLM exposure measure focused specifically on software (e.g., LLM-powered chatbot or application programming interfaces) and does not examine other technologies such as robotics.

    “Additionally, the LLM exposure measure focused specifically on software (eg, LLM -powered chatbot or application programming interfaces) and does not examine other technologies such as robotics where there is uncertainty around reliability and adoption and where there may be difficulty in creating a measurement rubric.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Findings may not account for the broader range of AI systems that can exist within different industries.

    “Moreover, findings may not account for the broader range or AI systems that can exist within different industries.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Sensitivity analysis exhibited the same gradient pattern, supporting the robustness of the quartile-based analysis.

    “Sensitivity analysis exhibited the same gradient pattern, supporting the robustness of our quartile-based analysis (see supplement 3).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The study used a representative national labor force survey to describe occupational LLM exposure and multiple precarious occupation dimensions, enabling population-level estimates.

    “Study strengths include the use of a representative national labor force survey to describe occupational LLM exposure and multiple precarious occupation dimensions enabled us to develop population-level estimates.”
    Quote in the paperSubstantial claim
45 of 45

  • The paper relies on this without support.

    Working in a precarious occupation can be directly and indirectly associated with the health and safety of workers.

    “Working in a precarious occupation can be directly and indirectly associated with the health and safety of workers.”
1 of 1

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Q1 and Q4 precarity labels are reversed in the results

    Major

    The methods define Q1 as the lowest exposure to precarity and Q4 as the highest, but the results call Q1 the highest and Q4 the lowest. This reverses the interpretation of the reported comparison.

  • Crosswalk code counts total 512, not 500

    Minor

    The 350 codes with an actual concordance and the remaining 162 codes total 512, rather than the stated base of 500 NOC codes.

    Reported 500 · recomputed 512

2 of 2

Cites

Flagged first, then by citations

14 of the 42 references couldn’t be matched to a known paper.

Research works printed but not linked: 2

  1. [26]Statistics Canada. Methodology of the Canadian Labour Force Survey: Chapter 6 Weighting and estimation Ottawa: Statistics Canada; 2017 [accessed 13 April 2026]. Available from: https://www150.statcan.gc.ca/n1/pub/71-526-x/2017001/chapt6-eng.htm
  2. [28]Panel of Research Ethics. Tri-Council Policy Statement: Ethical Conduct for Research Involving Humans – TCPS 2 (2022), Chapter 5: Privacy and Confidentiality: Government of Canada; 2023 [accessed 13 April 2026]. Available from: https://ethics.gc.ca/eng/tcps2-eptc2_2022_chapter5-chapitre5.html#d

Authors
Jetha, A., Liao, Q., Smith, P., Vu, V., Biswas, A., Smith, B., & Vahid Shahidi, F.
Affiliations
  • Institute for Work & Health (CA)
  • University of Toronto (CA)
  • Public Health Ontario (CA)
ORCID
Venue
Scandinavian Journal of Work, Environment & Health
Publisher
Scandinavian Journal of Work, Environment and Health
Published
2026-06-03
Volume
52
Issue
5
Pages
615-624
ISSN
  • 0355-3140 (ISSN-L)
  • 1795-990X
DOI
10.5271/sjweh.4312
PMC
PMC13555092
OpenAlex
W7163351214
Semantic Scholar
df0d89d949aa44ad806daec18e296101ca202b80
Cited by
0
References
42
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Scandinavian Journal of Work Environment & Health (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (submittedVersion, cc-by)
Funders
Social Sciences and Humanities Research Council of Canada
Topics
  • Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Labor market dynamics and wage inequality
  • Health Sciences › Health Professions › General Health Professions › Employment and Welfare Studies
  • Social Sciences › Sociology and Political Science › Digital Economy and Work Transformation
Keywords
large language models, precarious employment, digital technology adoption
MeSH terms
Job Security, Large Language Models, Adult, Canada, Employment, Female, Humans, Occupations, Surveys and Questionnaires

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-12.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-12.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review