AI-Exposed Jobs Were Already Declining Before ChatGPT Launched

The story most of us have been telling about AI and jobs goes something like this: ChatGPT launched in November 2022, everything changed, and now workers in AI-exposed occupations are in trouble. It's a clean narrative. It's also incomplete.

A new study published in January 2026 by researchers from the University of Pittsburgh, RAND Corporation, and other institutions analyzed 10.5 million LinkedIn profiles, millions of unemployment insurance records, and 3 million university course syllabi to trace exactly when AI-exposed jobs started showing signs of stress [Fact]. Their finding challenges the dominant narrative: the deterioration began months before ChatGPT ever existed.

The Trouble Started in Early 2022

The researchers found that unemployment risk for workers in AI-exposed occupations started climbing in early 2022 — a full 8 to 10 months before ChatGPT's public launch in November of that year [Fact]. This wasn't a subtle shift. Computer and mathematical occupations (SOC group 15), which include software developers, computer programmers, and web developers, saw the largest increase in unemployment during the 2022-2024 period compared to other occupation groups [Fact].

Even more telling, the share of recent college graduates entering AI-exposed occupations began declining as early as 2021 [Fact]. New graduates were already steering away from these fields before the ChatGPT moment that supposedly triggered the shift.

So what was actually happening? The tech industry was going through a structural transformation that had little to do with a chatbot. The post-pandemic correction in tech hiring, rising interest rates squeezing venture capital, and a broader recalibration of inflated tech valuations all contributed [Claim]. AI was part of the picture — companies had been quietly integrating machine learning tools into workflows for years — but the dramatic "ChatGPT killed jobs" framing misses these deeper currents.

The Counterintuitive Finding: LLM Skills Pay More

Here's where the research gets genuinely surprising. You might expect that workers most exposed to AI would be uniformly worse off. But the study found something that cuts against that assumption.

Recent graduates whose university coursework included substantial LLM-related skills training — not just basic computer science, but specific training in the tools and techniques behind large language models — actually earned higher starting salaries and found jobs faster after ChatGPT's release compared to peers without that training [Fact].

The researchers used a clever methodology here. They analyzed 3 million course syllabi across hundreds of universities to measure how much LLM-relevant content each program offered. Graduates from programs with deeper AI/LLM integration didn't just survive the post-ChatGPT labor market — they thrived in it [Fact].

This creates an important nuance. AI isn't simply destroying jobs in a blanket fashion. It's creating a wedge between workers who understand and can work with these tools and those who can't. The dividing line isn't between "tech workers" and "everyone else" — it's between AI-fluent workers and AI-unfamiliar ones, even within the same occupation [Claim].

How This Fits With What We Already Know

This research doesn't exist in isolation. It connects to several other findings that paint a more complex picture of AI's labor market impact.

The Dallas Federal Reserve has separately documented a decline in youth employment in tech-adjacent fields, noting that the trend predates the generative AI boom [Fact]. The Stanford HAI 2025 AI Index, drawing on ADP payroll data covering 16 million workers, found that AI exposure correlates with slower hiring growth but not outright job losses — at least not yet [Fact].

Meanwhile, the Economic Innovation Group (EIG) has pushed back on AI-centric explanations entirely, arguing that tech sector layoffs in 2022-2023 were primarily driven by over-hiring during the pandemic boom and subsequent corrections, not by AI displacement [Claim]. Their analysis suggests that attributing these job losses to AI gives the technology too much credit — or blame — for what were fundamentally business cycle adjustments.

Our own data at AI Changing Work shows that software developers have an AI exposure score of 83/100, while accountants sit at 73/100 [Fact]. These are high numbers, but exposure and actual displacement are very different things. The Brookings Institution's 2026 research reinforces this gap, finding that despite high theoretical exposure, actual AI-driven unemployment remains modest compared to what exposure scores might predict [Fact].

What This Means for Workers in Exposed Occupations

If you're a software developer, programmer, web developer, or accountant reading this, the research offers both caution and a clear path forward.

The caution: Your occupation is genuinely changing. The pre-ChatGPT decline in hiring and the post-ChatGPT acceleration of that trend are real. Pretending AI isn't reshaping your field would be a mistake.

The path forward: The same research shows that AI skills are a powerful differentiator, not just a defensive measure. Workers who invest in understanding LLM tools — how to prompt them effectively, how to integrate them into workflows, how to evaluate their outputs critically — are positioning themselves on the right side of the wedge this technology is creating.

This isn't about learning to code if you're an accountant, or about becoming an AI researcher if you're a web developer. It's about understanding how AI tools apply to your specific work and becoming the person in your organization who can bridge the gap between what AI can do and what your team needs done [Claim].

The study's most important lesson may be its simplest: the timeline matters less than the direction. Whether AI's full impact on your occupation arrives in 2027 or 2032, the trajectory is clear. The workers who start adapting now — while their existing skills still carry premium value — will have a significant advantage over those who wait for the disruption to become undeniable.

For a detailed look at how AI affects your specific occupation, including task-level automation scores and trend data, explore our analysis pages for software developers, computer programmers, web developers, and accountants.

Sources

• Frank, M.R., Javadian Sabet, A., Simon, L., Bana, S.H., Yu, R. (2026). "AI-exposed jobs deteriorated before ChatGPT." arXiv:2601.02554. arXiv
• Dallas Federal Reserve (2025). Youth employment trends in technology sectors.
• Stanford HAI (2025). AI Index Report — ADP labor market analysis.
• Economic Innovation Group (2025). Tech sector employment analysis.
• Brookings Institution (2026). AI exposure and labor market stability reports.

Update History

• 2026-03-22: Initial publication based on Frank et al. (2026) analysis of 10.5M LinkedIn profiles and unemployment records.

---

This article was generated with AI assistance using data from the cited sources. All factual claims are attributed and tagged with confidence indicators ([Fact], [Claim], [Estimate]). For detailed occupation-level data, visit the individual occupation pages linked above. Learn more about our AI-assisted content process.