It is not the restructuring that makes people ill — it is the unpredictability

You have been at that all-hands. Leadership says «exciting journey», «we need to become more robust», «nothing has been decided yet». Someone raises a hand and asks what happens to their department. The answer is that they will come back to that. The meeting lasts forty minutes, and afterwards you know exactly as much as you did walking in.

It is not the meeting that keeps you awake that night. It is the sentence that was not said.

I spent a PhD on this question, and I still find it odd how systematically we treat it as a detail. Information about the process is filed under communications — something you do nicely if there is time. The research says it is the variable itself.

Two downsizings, same company

In June 2025, Eirik Degerud, Andrea Rørvik Marti and Tom Sterud at the Norwegian National Institute of Occupational Health published a study in PLOS Mental Health. They followed 3,319 Norwegian employees across three waves — 2013, 2016 and 2019. All were free of mental distress at the start, and all stayed with the same employer throughout.

They looked at what happened to those who experienced downsizing in their own department. Then they split the group in two: those who felt they had received sufficient involvement, information and support during the process, and those who did not.

5.13 Without information

1.28 With information

An odds ratio of 5.13 means the odds of crossing the clinical threshold for mental distress were about five times higher than for those who experienced no downsizing. An odds ratio of 1.28 with a confidence interval crossing 1 means, in practice: we could not detect any risk.

Same event. Same workplace. Same uncertainty about the job. The only thing that differed was whether people were told what was going on.

What I found myself, and what I did not

This fits my own material, but not quite the way I expected.

With Stein Knardahl and Jan Olav Christensen I followed employees in 66 Norwegian companies over two years. In one paper — 7,985 people at baseline — we looked at who crossed the clinical threshold for mental distress. Reorganisation gave an odds ratio of 1.29. Downsizing 1.51. Layoffs 1.46. All moderate, all as expected.

The interesting part was the stacking. Two or more changes at once: 1.75. Changes at both measurement points, meaning two years running: 1.84. It is not the individual change that is heavy. It is that they arrive together, and that they do not stop.

In a second paper from the same material — 12,652 people at baseline — we looked at what actually changed in the work environment after a restructuring. Demands went up and stayed up. Job control went down. And two things fell both in the short and the long run: how predictable the job felt, and how employable people believed themselves to be.

In the third paper we tested whether leadership could dampen this. We found that supportive, empowering and fair leadership was associated with lower risk — leadership quality gave an odds ratio of 0.69. But we found no significant interaction. Leadership acted as an independent factor, not as a shock absorber against unpredictability.

That is an uncomfortably precise distinction, and it is worth taking seriously: my study did not find that good leadership protects against not knowing. Degerud’s study found that concrete information about the process made almost the whole difference. The advice that follows is not «be a kinder manager». It is «say what is happening, and when».

What we genuinely do not know

What is well established, across many studies and several countries: a systematic review of 39 longitudinal studies found that restructuring harms people’s well-being even when nobody loses their job. The state itself — not knowing what the work looks like in six months — carries a cost.

A meta-analysis of prospective cohort studies found perceived job insecurity gave an odds ratio of 1.29 for depressive symptoms, while actual unemployment gave 1.19. The confidence intervals overlap heavily, so I will not claim the fear is worse than the loss. But it is at least as bad — and that is bad enough that eight months of «we are reviewing our structure» is not free.

Why this concerns the AI rollout at your workplace

Here is the point I am actually writing this for.

Nearly every Norwegian organisation is now introducing AI tools. It is presented as technology: new systems, training, licences, a pilot in one department. It is not treated as a restructuring.

But look at what an AI rollout actually does to the work environment, measured in the variables we have numbers for. Demands change, and rarely downward. Control over one’s own work changes. Predictability falls — because nobody knows which tasks will still be theirs in a year. Perceived employability falls for the same reason.

That is the same exposure profile the restructuring literature has been measuring for twenty years. We do not need to wait for AI-specific health studies to know what determines the outcome.

And according to EU-OSHA’s 2025 survey, Norway is among the countries where most employees report using AI-driven software at work — 32% against 18% in the EU. We are far ahead in adoption and hold some of the best data in the world on what restructuring does to people. As far as I can tell, nobody has put those two facts in the same sentence in Norwegian.

One more thing, since it is circulating: in February this year a proposal from September 2025 for a clinical syndrome called AI Replacement Dysfunction received wide international coverage. It is not a study. It is a position paper with no data, published in a low-threshold journal. Do not build anything on it.

Three things that cost almost nothing

1. Give a date, not a mood.

   «We do not know yet, but on 15 October we will, and you will hear from us then» is information. «Exciting journey» is not. The first is also easier to say, because it does not require pretending to know something you do not.

2. Say what will not change.

   Almost every restructuring plan describes what becomes new. Almost none says what stays. It is the cheapest predictability available, and it is nearly always available.

3. Count the changes, not just this one.

   The effects stack. Before you start the AI pilot: how many other transitions are people in right now? If the answer is three, you are not starting one change. You are starting the fourth.

None of this is about making people comfortable. It is about uncertainty being a measurable occupational exposure, one that can be reduced without changing the decision — and that we have known this long enough that it is becoming embarrassing that it is still filed under communications.