AI attribution alone cuts discretionary effort by 13%, and your engagement survey won’t catch it

A preregistered study published by Brookings this month asked 3,628 people in the US and the Netherlands to help write a public health slogan about drinking more water. Nobody’s slogan changed. What changed was a single label: whether the three example slogans in front of them were credited to a marketing professional or to AI software. That one detail was enough to move discretionary effort by 13%, while how meaningful people said the task felt barely shifted at all.

So why do people contribute less if they don’t feel any different about the work?

The study comes from economists Milena Nikolova, Viliana Milanova and Feicheng Wang, published as a working paper through the Global Labor Organization and summarised on Brookings. The design is clean. Participants rated how meaningful the task felt, then saw three identical slogans randomly credited to a marketing professional or to AI software, rating them for creativity and persuasiveness. They rated task meaning again, then were offered the chance to submit a slogan of their own, the voluntary step where the researchers measured discretionary effort.

The discretionary effort drop your engagement survey will miss

Here’s the mechanism problem for anyone running people analytics. Most organisations track AI sentiment through the same instrument they use for everything else: an annual or quarterly survey asking how people feel about their work. This study suggests that instrument is pointed at the wrong variable. The meaning score barely moved. Discretionary effort, the extra mile people choose to give beyond the minimum, dropped by a relative 13%. A survey built to detect feelings will report that nothing much happened. Behaviour says otherwise.

Meaningful work remains one of the strongest levers organisations have for engagement, which is exactly why a wobble this small is easy to wave away, and exactly why the effort number sitting next to it shouldn’t be.

Global engagement has just hit its lowest point since 2020, and most diagnostic tools pointed at the problem are surveys asking people how they feel. If AI attribution erodes contribution while leaving stated sentiment intact, the standard dashboard won’t flag it, not until the effect compounds across enough tasks to surface somewhere else: in output, in retention, or in the unscripted effort that never gets logged.

Nobody in this study used AI

The detail that sharpens the result is what didn’t happen. Nobody in either group used an AI tool. They didn’t collaborate with one, prompt one, or check its output for accuracy. They were simply told a machine had made something, and that label alone changed how they rated the work’s creativity and persuasiveness, how much they trusted AI, and how much of their own effort they were willing to add. The slogans were identical throughout.

So this is an attribution effect, not a tool effect.

What moved discretionary effort here was credit rather than capability, so the live question for most HR teams runs wider than which AI workflow to roll out. It extends to how contribution gets attributed once that workflow exists, and whether the person doing the reviewing, checking and improving still sees their own name on the result.

That sits close to something this magazine has flagged before: 42% of employees are already using AI in secret, often specifically to keep the credit for themselves rather than share it with a tool nobody asked them to name. The Brookings finding suggests that instinct is rational. If flagging AI involvement costs perceived credit and invites harsher judgement of otherwise identical work, hiding it is the cheaper option. But how long will we all be able to do that….

---