The AI jobs panic vs the data

A breakdown of the actual data that contradicts the fear of mass unemployment.

ChatGPT has been around for four years, and the mass layoffs that were promised never came. However, it's highly likely your feed says otherwise.

Doom may feel like the default, but employment numbers tell a different story. Here's what the data actually shows.

Your news feed is optimised for fear

Before we look at the statistics we first need to look at where we're getting our information from, and for most people that's a social media feed.

Social media runs on an attention economy, and good news doesn't catch the eye quite like bad news. In a study by Robertson et al. (2023), researchers ran 105,000 headline variants across ~370 million impressions and found that each additional negative word in a headline raised click-through rates (CTR) by 2.3%. Positive words, on the other hand, actually reduced CTR by 1.0%.

Negativity bias doesn't just drive higher engagement rates, but also helps bad news spread faster. Brady et al. (2017) found every moral-outrage word in a post lifted sharing by approximately 20%.

With AI being so new and unknown, naturally opinions will be split on the subject. But with negativity bias shaping algorithms and driving engagement, the headlines that portend the worst are the ones that stand out.

Unemployment fears are as yet unfounded

When ChatGPT first arrived on the scene, many feared that AI would replace jobs and eventually cause an employment crisis. Four years on, the evidence shows the contrary.

In June 2026, US unemployment rates were at a historical low, with just 4.2% of the labour force out of work, and only a 1.1% layoff or discharge rate. These figures indicate that AI hasn't trigger mass layoffs, and an employment crisis has not yet come to pass.

Yale's Budget Lab backs this, as a study examining data through June 2026 found no clear relationship between AI exposure and unemployment, and that if anything, there was a slight employment increase in high-exposure roles.

Graduates are the exception

There is one case where the 2020s has seen a shift in employment opportunities, and it's affecting the graduate market. From 1990 through the late 2010s, recent graduates enjoyed a lower unemployment rate than the general population, dubbed the "graduate advantage". But as of late, this has inverted.

Graduate unemployment rates sit at approximately 5.7%, higher than the national 4.2%. While the 5.7% figure itself is unremarkable by historical standards, it indicates a shift in hiring patterns. This is echoed in big tech: new-graduates hires fell from 15% in 2019 to 7% in 2024, and entry-level hiring is down 50% compared to pre-pandemic levels. On the contrary, job postings demanding 5+ years' experience rose from 37% to 42%.

Cycle or AI? It's not obvious yet

As the bottom rung of the employment ladder narrows, it's easy to assume that entry-level roles are being offloaded to AI agents. But how can we be sure it's not just an ordinary business cycle?

An argument for this all being part of a cycle is the fact that young-worker employment is the most cyclical segment of the labour market, often being the last hired and the first fired. This entry-level role reduction isn't just coinciding with the rise of AI, but is following the sharpest interest-rate tightening in four decades.

There is some evidence for an AI effect on entry-level employment rates in Stanford's 'Canaries in the Coal Mine' study, where within the same AI-exposed occupations and firms, workers aged 22-25 fell approximately 19% relative to older colleagues. If this were a cycle, you'd expect to see a hit across all ages, but instead we're seeing younger workers disproportionately affected.

However, the Stanford publication also noted that AI-exposed declines aren't statistically significant until 2024, with all declines before likely non-AI factors, and the study only measures correlation, not causation. With these factors in mind, it's unlikely that AI shoulders all the blame for a drop in entry-level hires.

Historically, innovation leads to growth

AI isn't the first time automation has been the driver of a doom narrative, but there's a reason technology has never produced permanent mass unemployment.

When a new technology arrives, productivity rises and production cost falls, resulting in a drop in price and a rise in output. As a result, real incomes and business returns improve, firms invest and demand increases, and more people are hired to handle the increase in workload. Instead of causing unemployment, automation and the innovation loop create more jobs and higher living standards, with examples of this pattern evident throughout history.

Take the introduction of ATMs in the late 1960s. Between 1988 and 2004, tellers per branch dropped from 20 to 13, but cheaper branches meant more could open, and the number of urban banks rose 43%. As the number of branches increased, so did total teller employment.

We're now starting to see the same thing happen with AI. In a study of 5,179 customer-support agents, AI assistance raised issues resolved per hour by 14% on average, but 34% for the least skilled workers and little more than nothing for the top performers. In this setting, AI didn't replace the junior worker, but rather helped the junior worker perform on a senior level.

The macro estimates point to GDP gain

Scaled across the economy, the macro estimates are large. Goldman Sachs projects generative AI could raise global GDP by ~7% (~$7 trillion) and lift productivity growth by 1.5 points a year over a decade. Even the field's most cited skeptic, MIT's Daron Acemoglu, models a positive ~0.7% increase in productivity and ~1.1% GDP gain. The question isn't whether there'll be growth, but rather how much growth there will be.

There is one caveat: while productivity may increase, it doesn't automatically mean pay increases alongside it. This was the case from WWII to 1973, where US pay tracked productivity almost exactly, but since then they've diverged. Between 1973 and 2014, net productivity rose 72%, but median compensation rose by just 8.7%. Technology grows the pie, but it's policy and bargaining power that determines how it's sliced.

Don't believe the doom, check the data

Since AI appeared on the scene four years ago, none of the doom that was predicted has come to pass: people are still employed and jobs are still being created.

This isn't to dismiss the squeeze felt by current graduates, impacted by the decline in entry-level opportunities. However, it would be short-sighted to ignore other factors, such as the business cycle, and attribute the drop in graduate opportunities to AI alone.

Here's my position: AI moves work up a level rather than eliminating it, and the productivity gains feed into a loop that's good for workers and for growth.

Doom is loud because fear is a growth hack, while data is quiet because reassurance doesn't trend. So the next time you see a post prophesising the collapse of the job market on your feed, seek out the statistics that paint the full picture.

Sources

1. Robertson, C.E., et al. "Negativity drives online news consumption." Nature Human Behaviour (2023). nature.com · free: PMC10202797
2. Vosoughi, S., Roy, D., Aral, S. "The spread of true and false news online." Science (2018). pubmed 29590045
3. Brady, W.J., et al. "Emotion shapes the diffusion of moralized content in social networks." PNAS (2017). pnas.org
4. Baumeister, R., et al. "Bad Is Stronger Than Good." Review of General Psychology (2001). PDF
5. US BLS Employment Situation, June 2026 (unemployment 4.2%; +57k payrolls), as reported. CNBC · BLS
6. US BLS JOLTS, June 2026 (layoffs/discharges rate 1.1%). bls.gov/jolts
7. The Budget Lab at Yale, "AI Is Probably Not (Yet) the Reason for Labor Market Weakening" (2025). budgetlab.yale.edu
8. Federal Reserve Bank of New York, "The Labor Market for Recent College Graduates" (Q1 2026, ~5.7%). newyorkfed.org
9. Federal Reserve Bank of St. Louis, "It's (Still) the Business Cycle: Young Adult Workers in a Low-Hire, Low-Fire Labor Market" (2026). stlouisfed.org

Compiled from public data, August 2026. Analysis and any errors are the author's. Not investment or economic advice.