A new survey out of Quebec puts real numbers on something a lot of managers have been quietly wondering: is AI actually making work better, or just different?
Researchers at the International Observatory on the Societal Impacts of AI and Digital Technology, working with 11 labor unions that together represent 1.4 million members, surveyed 4,595 union workers in 2025. The results do not support the simple story that AI frees people up to do better work. They support a messier one.
Among people who use AI every day, only 44 percent report a real productivity gain. A full 26 percent say it actually slowed them down, mostly because of low quality AI output that had to be checked, corrected, or redone. That matches a much larger pattern found elsewhere: a 2025 study out of MIT looked at 300 corporate AI projects and found that 95 percent of them delivered no measurable financial return. Different countries, different methods, same conclusion: dropping an AI tool into a workflow does not automatically make that workflow better.
The Quebec data goes further than most surveys by showing who wins and who does not. People with a postgraduate degree are more than twice as likely to report a productivity gain from AI as people with only a high school education, 55 percent versus 22 percent. Professionals tend to choose when and how to use AI. Technical, industrial, and service workers are more often told to use it, and they report more monitoring through the software, not less.
This lines up with separate global research from PwC, which found that workers who use generative AI daily report higher pay and stronger job security than occasional users. The pattern across every one of these studies is the same: casual, unsupported use of AI barely moves the needle, and sometimes hurts. Deep, trained, well governed use pays off. Most companies are stuck in the first category without realizing it.
The trust numbers explain why. Only 24 percent of workers say their organization has any actual policy on how AI should be used. Barely 12 percent were consulted before it was introduced. Just 35 percent say they trust their employer's approach to AI at all. Nearly one in five fear losing their job to it, and that fear is highest among older workers and those in manufacturing, not the office workers most forecasts warn about.
That last point matters. Most predictive models point to office and knowledge work as most exposed to AI. But the people most afraid are the ones with the least say in how it gets used and the least education to fall back on. Fear tracks powerlessness more than it tracks actual exposure.
The lesson for anyone rolling out AI at work is not that the technology fails. It is that handing someone a tool without training, rules, or a say in how it is used produces exactly the mixed results this survey found. The businesses seeing real gains are the ones treating AI adoption as a change to manage with people, not a license to distribute and forget.