AI job displacement shows up in slow rehiring, not in layoffs

AI job displacement is real, but it is not showing up where most people look. US employers are laying off workers at close to the lowest rate since the official survey began in 2000. The damage is in how long it takes to find the next job. A new Goldman Sachs study of more than 15,000 workers who lost their jobs between 2016 and 2025 found that those most likely displaced by AI in 2025 took about a month longer to find work than other laid-off workers, Axios reported on 3 November.

The official figures point the same way. In September 27.3% of the unemployed had been out of work for 27 weeks or more, up from 23.9% a year earlier and 23.3% in September 2019. Over the same year the number of unemployed fell. Fewer Americans are losing their jobs, and those who do are staying out longer.

Share of US unemployed out of work for 27 weeks or more, September of each year, %. Source: US Bureau of Labor Statistics via FRED (UEMP27OV and UNEMPLOY), seasonally adjusted; our calculation.

2019	23.3
2025	23.9
2026	27.3

Source: US Bureau of Labor Statistics via FRED (UEMP27OV and UNEMPLOY), seasonally adjusted; our calculation.The Macro Notebook

What the Goldman study found

There is no count of how many workers have lost jobs to AI. The study, by Goldman economist Pierfrancesco Mei, instead estimates which layoffs were most likely caused by it, from the worker’s occupation, how much their industry uses AI and whether their position was cut. Three findings stand out:

• A month longer to find work. Workers most likely displaced by AI in 2025 took about a month longer than other laid-off workers to get a new job. They were about 10 percentage points more likely to be still unemployed a year later.
• Older workers wait longer. Under 35, the gap was about three weeks; for those aged 35 and over, five weeks.
• Back into the line of fire. Of the workers from AI-exposed jobs who found new work in 2025, about three-quarters went into jobs with similar or greater exposure to AI, up from just over half in 2023.

The last point is the quiet one. Most displaced workers are moving into jobs that AI can reach just as easily, so the next round of cuts may find many of the same people.

Why layoffs and payrolls miss AI job displacement

The measures most people watch are built to count firings. The layoffs and discharges rate in the Labor Department’s job openings survey was 1.0% of jobs in September, below its 2019 average of 1.2%. Weekly jobless claims were 194,000 in the week to 24 October, and they ran under 200,000 for four weeks in a row to 3 October, a run last seen in 1969, as we reported on 9 October.

Hiring is the weak side. The hires rate was 3.4% in September and has been 3.5% or lower every month since March 2024. Before that, outside March and April 2020, a rate as low as 3.4% was last seen in February 2014; in 2019 it averaged 3.9%. Payroll growth has slowed too, as we reported on 2 October. This is the low-hire, low-fire market: a worker who keeps a job is safe, and a worker who loses one waits.

The waiting shows in the household survey. The number out of work for 27 weeks or more rose to 1.94 million in September from 1.82 million a year earlier, while total unemployment fell to 7.11 million from 7.61 million. The median spell of unemployment lengthened to 11.5 weeks from 10.1. The unemployment rate fell to 4.2% from 4.4%, so the headline figure looks better even as the wait gets longer.

Recent graduates face the hardest start

The workers with the least experience have it worst. Recent college graduates, aged 22 to 27, had an unemployment rate of 5.4% on average in the third quarter, against 4.0% for all workers, the New York Fed’s figures released on 2 November show. In 2019 the gap averaged 0.3 points; now it is 1.4 points. The rate did improve during the quarter, to 5.2% in September, the lowest in more than a year. But 42.5% of employed recent graduates held jobs that do not need a degree, the highest share since October 2020 on the bank’s figures.

Axios notes that AI may be taking some of the entry-level work that used to start graduates’ careers. The data cannot prove that AI is the cause. They do show that the young and the displaced, the two groups most exposed to it, are the ones who wait.

What to watch next

The October jobs report on 6 November will give new figures for long-term unemployment and the median spell, the numbers to watch rather than payrolls alone. Weekly claims come on 5 November. The New York Fed’s next update on graduates is due in February.