When a colleague sends you AI-generated work that's unhelpful, who ends up paying for it, and can you even tell?

This explores whether managers and individual contributors run into low-quality AI work differently, and who ends up absorbing its cost. The collection doesn't answer this directly. None of these studies splits its findings by role or seniority. What it does show is something that may matter more than rank: the cost of bad AI work lands on whoever receives it, and the signals that would let anyone see it coming tend to be hidden.

The receiving end is where the damage shows up. In a survey of about 1,000 U.S. desk workers, respondents estimated that 15.4% of the work they receive is AI-generated but unhelpful. Each instance took nearly two hours to deal with, which was longer than if the sender had simply done the work themselves How much work that employees receive is actually unhelpful AI content?. The cost is social as well as practical. About half of recipients rated the sender as less capable and reliable, and nearly a third said they'd be less willing to work with that person again Does receiving AI-written work change how we judge the sender?. Anyone whose job is mostly reviewing, integrating, or approving other people's output is structurally on the receiving end, and that includes managers. Peers who depend on a colleague's handoff are on it too.

The less obvious part is that people expect this judgment and adjust for it. Across four experiments with more than 4,000 participants, AI users expected to be seen as less competent and diligent. They were less willing to tell managers and colleagues that they had used AI Do people fear judgment when they use AI at work?. So the person best placed to catch workslop, the one reviewing it, often doesn't know AI was involved. Hiding AI use doesn't stop low-quality work from moving up the chain. It removes the label that would tell a reviewer to look more closely. Research on disclosure suggests that being open helps calibrate trust only when people also see repeated outcomes Does revealing AI identity help or hurt user trust?. Concealment cuts off that feedback loop.

The filtering that would normally stop bad output before it's handed off is also weak. Writers edited AI-generated paragraphs only 23% of the time, and their edits left the text about 96% unchanged Do writers actually edit AI-generated text before publishing?. Senders may not see a problem in the first place. The 'LLM fallacy' describes people crediting AI output to their own skill How does AI-assisted work reshape how people see their own abilities?. Writers who had more control over the text felt more ownership of it, while personalizing the AI made no difference Does user control over AI text shape feelings of ownership?. Together these suggest one possible pattern, though the collection doesn't test it: producers may feel more confident in their work than its quality justifies, while reviewers carry the hidden cost without being told AI was used.

If you're looking for evidence that compares managers with individual contributors directly, this collection doesn't have it yet. The more useful idea it offers is to stop asking which job title and start asking who sits downstream of whom. Workslop flows toward reviewers, and the social pressure to hide AI use keeps it hard to spot along the way.