How much do desk workers actually suffer from AI workslop?

A representative survey of 962 American full-time desk-workers, run by BetterUp Labs and Stanford's Social Media Lab (Liebscher, Lee, Rapuano, Kellerman, Niederhoffer, Hancock; PsyArXiv, 2026-02-05), finds that 52.7% of respondents "reported sending some amount of AI-generated workslop to others," which the authors define as "predominantly AI-generated content that appears to fulfill a given task, but lacks the substance to meaningfully advance the work at hand." A further 38% reported receiving workslop from colleagues, and said the effort of "processing, discussing, and revising - or even redoing" that work cost them "around 3.4 hours per month on average." Receiving workslop also "negatively affected their mood and harmed their perceptions of their colleagues as trustworthy, reliable collaborators."

The paper reports, from the same self-report survey, that sending workslop "was more common among people who had high trust in AI, low agency over AI use, organizational encouragement to use AI, and low psychological safety at work." It frames workslop as "a failure mode of AI-mediated communication" (building on Hancock et al.'s AI-mediated-communication framework): senders "offload cognitive work to AI systems that lack full context, creating work with manipulated quality signals that impose disproportionate burdens on the recipient." The authors also stress that workslop is "an inherently perceptual phenomenon" — whether an artifact counts as workslop depends on the recipient judging it low-quality, low-effort, or AI-generated, a judgment made with "limited visibility into the steps their colleagues take," so that "even work produced with substantial prompting and thoughtful revision may ... still be perceived as workslop if that effort is not legible to the recipient."

That perceptual framing extends Which workplace cues survive AI mediation and which disappear?: where that paper traces the asymmetry to an "output-centered organization" of work in which the finished deliverable stands as proof labor occurred, this survey supplies the downstream cost of the same blind spot — colleagues who cannot see hidden effort instead judge the visible, legible signal and penalize the sender's trustworthiness when it reads as a shortcut. It also sits opposite Can generative AI replace the benefits of having a human teammate?, which shows AI sustaining collaborative benefit under a structured task; workslop describes what happens outside that boundary condition, when AI use is low-agency or low-oversight. And it names a second, social cost alongside the one in Does generative AI actually save workers time or intensify it?: that ethnography found AI adding burden through expanded scope and dissolved pauses, while this survey finds AI adding burden through the corrective labor it pushes onto colleagues.

The excerpt does not establish causation: trust in AI, agency, organizational encouragement, and psychological safety are correlational predictors drawn from the same self-report instrument as the sending behavior they predict, so the direction of influence among them is not tested. The sample is American full-time desk-workers recruited through CloudResearch's Connect platform, skewed white (76.6%) and bachelor's-degree-or-higher (78.6%), and does not generalize to other countries, occupations, or work arrangements. The 3.4-hours-per-month cost and the mood and trust effects are recipients' self-reported perceptions of harm, not independently measured productivity or relationship outcomes. At the strength this evidence allows, the finding supports treating workslop as a widespread, named cost of low-oversight AI use rather than a rare failure, strong enough to justify managing the four named predictors but not strong enough to establish which one, if any, causally drives the behavior.