Do AI-exposed graduates' early penalties fade or persist beyond the first two years?

Research question

For graduates who entered the labour market between 2022 and 2025, do the employment, earnings and occupational-match penalties of AI-exposed fields of study persist, fade or widen from the second to the fifth year after graduation, do graduates catch up by moving into exposed occupations later, and how do the size and decay of the penalty compare, in the same data and model, with the scarring of graduates who entered during an earlier recession?

Why it matters

The entry penalties found so far are measured at or near the first job. Whether they fade over time or compound, because missed first jobs can mean missed on-the-job learning, decides whether this is a temporary hiring pause or a lasting cohort disadvantage. Students choosing fields and departments planning programmes need that distinction, and none of the current evidence can provide it yet.

The debate lists this among what we still do not know: What happens to displaced or downshifted entrants over five to ten years?

Evidence that motivates it

1. US graduates from the most AI-exposed tenth of college majors experienced a 5 percentage-point decline in initial employment and a 13% decline in initial earnings relative to less-exposed majors after ChatGPT. (LD02-HQ-C06 v1.0)
   Graduates from the most AI-exposed tenth of US majors saw a 5-point fall in initial employment and a 13% fall in initial earnings; the claim explicitly does not license long-run scarring, which is exactly what this design would measure.
   Cites: Orr et al. (2026), “Graduating into Disruption”
2. In US Census administrative data, regression-adjusted employment of workers aged 22–24 in the most AI-exposed industry-state cells fell 12% over 10 quarters after ChatGPT, with reduced hiring the primary mechanism. (LD02-HQ-C05 v1.0)
   US Census data show a 12% relative employment fall for 22–24-year-olds in the most exposed cells, with reduced hiring the primary mechanism, so the group to follow is mainly would-be entrants rather than laid-off workers; the claim does not attribute all of the fall to AI.
   Cites: Tucker (2026), “You’re (not) Hired”

Claims are shown as they read on the day this pack was built; claim pages keep every earlier version.

A possible design

Graduate-cohort event study. Treatment intensity is the pre-2022 AI exposure of each field of study (the share of a field's graduates working in exposed occupations, times occupation exposure). Compare cohorts graduating 2016–2021 with 2022–2025 cohorts, following each for every available year after graduation: employment, earnings, whether the job is in an occupation typical for the field, and later moves into exposed occupations (catch-up). Benchmark the size and decay of any penalty against the same register's estimates for graduates who entered during an earlier downturn, using the same model. US Post-Secondary Employment Outcomes (PSEO) public tables may provide a parallel, coarser US check if recent graduation cohorts are released. The US Census paper already measures early-career penalties (press coverage reports a two-year horizon; unconfirmed in the paper), so the new contribution lies in years three to five, occupational match and catch-up, the same-model recession benchmark and a European register; a US analysis would extend that paper rather than re-estimate first-year penalties.

Data needed

A population register that links graduation records (field, year, institution) to monthly or yearly employment and earnings: Norway, Denmark, Sweden, Finland or the Netherlands all have such linkages via remote access. US PSEO public tables (outcomes by degree level and major; institutional coverage varies by state, and which cohorts and years after graduation are published must be checked) for the US comparison. A field-to-occupation exposure crosswalk built from pre-2022 data.

Feasibility

A PhD project. Low in money (register remote access for an affiliated researcher), high in time: each extra year of follow-up only arrives with the data. Can start as a first paper with one to two years of follow-up and grow into a thesis as later years arrive. Pre-registering the benchmark comparison before outcomes for 2024–2025 cohorts are observed would make the design more credible.

Has this been done?

Partly done already: related work covers some of this question (first-pass screen of 12 October 2026, checked again on 13 October 2026).

US Census work finds a 13% initial earnings drop and a 5-point drop in initial employment for the most AI-exposed majors, says the effects attenuate further from entry and compares the loss with "a large recession"; one press report (unconfirmed in the paper) puts the gap at about 5% two years after graduation; Texas records show first-year penalties; LinkedIn evidence shows exposed-job entry fell for 2021-onward cohorts. No study was found that follows post-2022 graduate cohorts beyond two years, measures occupational match and later catch-up into exposed occupations, or benchmarks the decay of the penalty against recession-graduate scarring with the same model and data. No register study outside the US was found that follows graduates of AI-exposed fields after 2022.

Literature check. The Census CES-WP-26-56 paper already measures initial penalties for the most AI-exposed majors, reports that they attenuate further from entry and compares the loss with "a large recession", so early-career penalties and a rough recession comparison are no longer open for the US. Only press coverage (Chronicle of Higher Education) reports a two-year figure (the earnings gap narrowing from 13% to about 5%); the paper's abstract gives no horizon and that figure is unconfirmed in the paper (it may conflate the 5-point employment drop). The abstract compares the loss with "a large recession" without naming the Great Recession. Years three to five, occupational match and catch-up, a same-model benchmark against recession cohorts, and any non-US register evidence remain open, so the question should be narrowed to those. Checked by Claude (SocioLens Project Lead) with AI research agents, by web search; not a person and not a domain expert.

Where the search could not reach. Screened by web search on the date shown; Google Scholar, SSRN and paywalled journal pages were not systematically searched, the CES working paper PDF was blocked to automated fetching, and two items were read at abstract level only (for the CES paper, the abstract page). Absence of a hit is not proof that cohort follow-up studies are not under way. Literature check of 2026-10-13: The CES paper PDF is blocked to automated fetching (robots), so the two-year result and recession comparison come from press coverage only and must be checked against the paper before publication. Google Scholar and paywalled journals were not searched. The shared WebSearch budget for this session (200 calls across all agents) ran out partway through this round. Absence of a hit is not proof that longer-horizon cohort studies are not under way.

Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.

Nearest work found

• Orr, Tucker & Warren, "Graduating into Disruption: Labor Market Outcomes for AI-Exposed College Majors," US Census Bureau Center for Economic Studies Working Paper CES-26-56 (September 2026) Judged from an abstract, summary or search listing only Some details not confirmed
  Abstract page only read (the PDF could not be fetched): US administrative graduate records; most exposed decile of majors shows a 5-point drop in initial employment and 13% lower initial earnings after ChatGPT, with about half from moving into lower-paying sectors such as restaurants and retail; effects weaken with time since entry but remain substantial for the most exposed majors, and the abstract page gives no horizon. Press coverage (Chronicle of Higher Education, listed below) reports that the gap for the most exposed majors narrows from 13% to about 5% two years after graduation; not confirmed in the paper. The Chronicle lists the authors as Warren, Tucker and Orr; the order on the title page was not checked. Initial outcomes, not multi-year trajectories.
• Frank, Javadian Sabet, Simon, Bana & Yu, "AI-exposed jobs deteriorated before ChatGPT," arXiv:2601.02554 (5 January 2026) Judged from an abstract, summary or search listing only
  Abstract only read: monthly US unemployment-insurance records show rising risk in exposed occupations from early 2022; LinkedIn profiles show graduates from 2021 onward entering exposed jobs less often, before ChatGPT; syllabi data suggest more AI-exposed coursework was associated with better first jobs after ChatGPT. Raises the timing problem this design must handle; does not follow cohorts for scarring or recovery.
• Friebel, Huang, Li, Shukla & Zhang, "Pyramids, Diamonds, and Oscillations: AI and the Structure of Internal Labor Markets," NBER conference paper (10 April 2026)
  Theory predicts a possible 'lost cohort' of juniors entering during a hiring freeze, with the next cohort benefiting from a later senior shortage; the empirical section uses aggregated Burning Glass Institute job postings, is descriptive and does not track individuals. Supplies a testable prediction, not a test.
• Dodini & Smith, "AI plays a role in weak labor market for college graduates," Federal Reserve Bank of Dallas, Dallas Fed Economics (22 September 2026)
  Texas public-university administrative education and earnings records, major-level exposure from postings and an Anthropic task measure: graduates of more exposed majors saw lower employment and about 5% lower first-year earnings for 2021–2024 cohorts, and more went on to graduate school. First four quarters after graduation only; does not follow cohorts for recovery or scarring.
• Chronicle of Higher Education, "The Most AI-Exposed College Graduates Are Entering Another Great Recession" (September 2026), reporting on Census CES-WP-26-56 Judged from an abstract, summary or search listing only Some details not confirmed
  Opened; secondary press report, not the paper. Says the CES paper uses college-supplied student and major records linked to state unemployment-insurance earnings, finds a 13% first-quarter earnings gap for the top exposure decile that narrows to about 5% two years after graduation, and quotes the authors comparing the effect with Great Recession entry effects (the paper's abstract says only "a large recession"), that recession scarring lasts five to ten years, and that it is too early to tell whether AI exposure will do the same. The two-year figure and the recession comparison could not be confirmed in the paper itself (PDF blocked), and a second press report (Staffing Industry Analysts) gives no two-year trajectory.
• Kostøl et al., kiindeksen.no AI index for Norway, education pages (GitHub repository andreaskostol/AI-Norway, pull request 1, 2026) Judged from an abstract, summary or search listing only Some details not confirmed
  Opened (pull request description only). Adds field-of-study pages that map Norwegian higher-education programmes (DBH data, 35 institutions) to occupational AI exposure. Descriptive mapping of programmes to exposure; the description shows no linkage to post-graduation employment or earnings and no cohort follow-up. Relevant as a possible Norwegian field-to-exposure crosswalk (our inference). Authorship beyond the repository owner not confirmed.

References

Sources behind the motivating claims:

• Orr, Tucker & Warren, “Graduating into Disruption,” US Census CES-WP-26-56, September 2026 (SRC-LD02-ORR-TUCKER-WARREN-2026, original)
• Tucker, “You’re (not) Hired,” US Census CES-WP-26-27, April 2026 (SRC-LD02-TUCKER-CES-26-27, original)

BibTeX

The first entry cites this opportunity; then its nearest work; then the sources above. Nearest work and sources are recorded as reference strings, so each goes in whole as the title; add authors and year from the original before you cite it.

@misc{sociolens_ro_ld02_03_v2,
  author = {{SocioLens project}},
  title = {Do AI-exposed graduates' early penalties fade or persist beyond the first two years?},
  year = {2026},
  date = {2026-10-13},
  howpublished = {\url{https://sociolens-web.pages.dev/opportunities/ro-ld02-03/v2/}},
  note = {Research opportunity RO-LD02-03, version v2}
}

@misc{ro_ld02_03_nearest_1,
  title = {Orr, Tucker \& Warren, "Graduating into Disruption: Labor Market Outcomes for AI-Exposed College Majors," US Census Bureau Center for Economic Studies Working Paper CES-26-56 (September 2026)},
  url = {https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-56.html},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only; Some details not confirmed}
}

@misc{ro_ld02_03_nearest_2,
  title = {Frank, Javadian Sabet, Simon, Bana \& Yu, "AI-exposed jobs deteriorated before ChatGPT," arXiv:2601.02554 (5 January 2026)},
  url = {https://arxiv.org/abs/2601.02554},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only}
}

@misc{ro_ld02_03_nearest_3,
  title = {Friebel, Huang, Li, Shukla \& Zhang, "Pyramids, Diamonds, and Oscillations: AI and the Structure of Internal Labor Markets," NBER conference paper (10 April 2026)},
  url = {https://conference.nber.org/conf_papers/f232578.pdf},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{ro_ld02_03_nearest_4,
  title = {Dodini \& Smith, "AI plays a role in weak labor market for college graduates," Federal Reserve Bank of Dallas, Dallas Fed Economics (22 September 2026)},
  url = {https://dallasfed.org/research/economics/2026/0922},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{ro_ld02_03_nearest_5,
  title = {Chronicle of Higher Education, "The Most AI-Exposed College Graduates Are Entering Another Great Recession" (September 2026), reporting on Census CES-WP-26-56},
  url = {https://www.chronicle.com/article/the-most-ai-exposed-college-graduates-are-entering-another-great-recession},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only; Some details not confirmed}
}

@misc{ro_ld02_03_nearest_6,
  title = {Kostøl et al., kiindeksen.no AI index for Norway, education pages (GitHub repository andreaskostol/AI-Norway, pull request 1, 2026)},
  url = {https://github.com/andreaskostol/AI-Norway/pull/1},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only; Some details not confirmed}
}

@misc{src_ld02_orr_tucker_warren_2026,
  title = {Orr, Tucker \& Warren, “Graduating into Disruption,” US Census CES-WP-26-56, September 2026},
  url = {https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-56.html},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{src_ld02_tucker_ces_26_27,
  title = {Tucker, “You’re (not) Hired,” US Census CES-WP-26-27, April 2026},
  url = {https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-27.html},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}