Almost one in five workers AI replaced were asked back

Almost one in five American workers who say AI replaced their job have since been asked to come back. That number comes from a survey run by SHL in June and July 2026, and it is the most interesting labour statistic of the month — not because AI failed, but because the accounting that justified the cuts was wrong.

The detail is in what the returning workers asked for before they agreed to return: 51% wanted a substantial pay rise, 42% wanted written job-security guarantees, and 40% wanted the employer to say plainly where AI is used. Those are not the demands of people who are grateful. They are the terms of people who have already been through one round.

What the survey actually found

SHL commissioned Propeller Insights to survey more than 1,000 workers in the US and more than 1,000 in the UK between June and July 2026. In the US, 19.1% of respondents said they had been replaced by AI and were afterwards asked to come back. Across both countries the figure is 16%.

Two other numbers are worth keeping:

• 24% of workers aged 18 to 24 were replaced and later asked back. The youngest cohort has the least protection and the most exposure to entry-level automation, and it is also being recalled fastest.
• 32% of US hiring managers, in a separate Robert Half poll, said the same or similar roles were reinstated after AI-driven cuts.

Managers and workers are describing the same reversal from opposite sides of the table.

Why the reversal keeps happening

Three patterns explain it, and none of them is about the model being bad.

The workflow was never fully costed. A tool can draft a support reply, but somebody has to check it, handle the exceptions, and take the blame when it is wrong. Remove the checker and the error rate goes up where customers notice. Rehiring the person who used to catch those errors is cheaper than losing the account.

Hiring back is slower than letting go. A laid-off worker knows the systems, the customers and the unwritten rules. A replacement needs months. When a company works out in week six that the work is not getting done, the fast fix is a phone call, not a job advert.

The cut was a signal, and reversing it partly is damage control. Once a company announces layoffs "because of AI", it has told the market its headcount was inflated. Calling some people back keeps the productivity story while quietly admitting the arithmetic was off.

Why this matters legally, not just emotionally

Here is the part most AI coverage skips. A company that calls workers back after an AI redundancy is admitting, in payroll records, that the automation did not remove the role.

That admission matters because the rules around AI dismissal are starting to have teeth.

In the United States, the WARN Act never asks why the jobs went. It asks how many and where: 60 days' written notice if an employer with 100 or more employees cuts at least 33% of the active workforce and at least 50 people at a single site inside 30 days, or 500 or more regardless of percentage. Most AI restructurings fall below both floors, which is why they often owe nothing beyond contractual notice.

California is the first state to change the paperwork. Senate Bill 951, effective 1 January 2027, requires an employer to disclose when a covered layoff is caused in whole or in substantial part by AI or automation. It does not change the 60-day notice or the 75-employee threshold for Cal-WARN. It changes what has to be said out loud.

And this week the Philippines went further. Senate Bill 2563, filed on 8 November 2026 by Senator Erwin Tulfo, would require six months' notice to the Department of Labor and Employment before AI-driven dismissals, reskilling and redeployment first, fair selection criteria, and separation pay of at least two months' salary or two months' salary per year of service, whichever is higher. It is a bill, not a law, but it is the first time a national legislature has published a price list for replacing people with software.

What to do with this if you are the one being replaced

Three things follow from the data.

First, keep a record of what the job involved. The recall conversation is about specific work, not job titles.

Second, if your employer says AI will do the job, ask what percentage it does today. Not next year. Today. A tool that drafts 60% of the tickets still needs someone for the other 40%.

Third, if a former employer calls, remember that the recall rate is high enough that you can negotiate. The people in this survey did, and half of them got a pay rise out of it.

For employers, the lesson is narrower and more expensive. An AI cut that gets reversed costs the redundancy payment, the rehiring cost, the lost knowledge, and now a premium on top. The survey does not say AI does not work. It says the business case usually leaves out the part where a person checks the output.

I wrote the full breakdown of what each jurisdiction requires, with the thresholds and the notice periods, here: AI layoff laws in the US, UK and Philippines, compared. And if your question is specifically whether the WARN Act applies to an AI layoff, the short answer is that it depends on headcount: Does the WARN Act cover AI layoffs?.

Originally published at voxahq.in.