Half the participants in an experiment were told three campaign slogans came from AI software, the other half that a marketing professional wrote them. The slogans were the same either way, and the label was assigned at random. Those who thought AI wrote them found the task less meaningful, rated the slogans less creative and less persuasive, and were less likely to offer one of their own.
The result comes from a preregistered experiment reported by Brookings, run as nationally representative surveys in the United States (1,511 respondents) and the Netherlands (2,117). Participants were told that too few people drink enough water and invited to help write a slogan for a campaign about it. They rated how meaningful the task felt before and after reading the three slogans, and the closing offer to submit one of their own served as the measure of effort.
About 26% took it up, and the AI label cut that share by 3.4 percentage points — roughly an eighth. The fall in perceived task meaning was about 0.07 standard deviations, a change the authors call small beside an effect on effort they call sizeable. The study is a working paper and has not been through peer review.
The authors suggest this may be part of why AI’s measured productivity gains have not shown up across whole economies — output rises while the people involved contribute less of their own — though they add that their experiment does not test that idea directly. They advise employers to build workflows that keep each person’s contribution visible and give them real responsibility for the result.
They set out the limits themselves: a short creative task rather than a full work process, and participants who only judged output labeled AI-made, never working with an AI system. The finding is not that AI drains meaning from work overall, they say — only that people may think a task matters less, and ease off, when they believe the creative part is already done.