You open a report twenty pages long. It reads clean and confident, exactly what you asked for. Five pages in, the argument still has not gone anywhere. An hour later you close it having learned one thing: it took someone five minutes to generate, and it took you an hour to find that out.

Researchers at BetterUp and the Stanford Social Media Lab call that workslop, AI generated work that looks finished but is not, and quietly shifts the effort onto whoever receives it.

The Numbers Behind It

Kate Niederhoffer, Alexi Robichaux, and Jeffrey Hancock surveyed 1,150 full time employees across industries in the United States. The results were published in Harvard Business Review in January 2026. 53% admitted sending a colleague AI generated work they knew was unhelpful, low effort, or low quality, at least some of the time. 41% recalled a specific instance of workslop that affected their own work.

What It Costs Is Talent, Not Just Time

The researchers documented real cases, and one is worth sitting with. At one company, low quality AI generated code created a wave of critical bugs in the shared codebase. One engineer grew so resentful of cleaning up after teammates that they quit, giving only two days notice. Losing that engineer cost far more than any AI subscription saved. That is what makes workslop dangerous. It does not just waste an afternoon, it drives good people out the door.

Why Blaming Employees Misses The Point

It would be easy to blame the people producing workslop. The research points the other way. 41% of employees said leadership pushed AI use without clear instructions or the context needed to apply it to real work, at the same time budgets tightened and roles consolidated. An employee told to use AI everywhere, but never shown what good AI assisted work looks like, produces something that satisfies the mandate and rarely satisfies the work.

Four Things That Actually Reduce It

Employees who feel competent and in control of their AI tools are half as likely to produce workslop. Teams with a strong culture of trust, where people can admit AI did not help without fear of judgment, cut workslop by 61%. Four practices build on that.

Set department specific guardrails, not one company wide mandate. A sales team drafting outreach needs different guidance than a finance team summarizing reports. One blanket instruction to use AI everywhere ignores that difference, and it is the exact pattern the research ties to workslop.

Add a review step before AI assisted work goes out. A short gut check helps: did a person add something the model could not, and would the reasoning hold up to a direct question. That one step would have caught the case above before it reached anyone’s inbox.

Train by department, not by slide deck. A legal team’s prompts and risk tolerance look nothing like a marketing team’s. For one sales team, we built training around their actual calls and deals, prompts and projects designed to challenge account strategy before a big meeting, not a generic AI overview handed to every department the same way.

Make it safe to say a tool did not work. Of everything the researchers measured, trust moved the number the most. That starts with a manager who asks how AI use is really going, and means it.

AI Adoption Is Not AI Capability

Giving employees access to ChatGPT or Claude is easy. Building an organization that knows when to use it, how to use it, and how to review what it produces is harder. Workslop is the gap between the two.

Counting how many employees use AI tells leaders very little on its own. The better question is simpler: is the work actually better because they are using it?