AI tools are now common enough at work that their weak outputs have become a workplace issue of their own. A study from BetterUp Labs and the Stanford Social Media Lab describes a pattern it calls "workslop": AI-generated material that looks like work but does not add enough value for the people who receive it.
The problem is not only that poor AI output exists. The study suggests that when it is passed between employees, managers, and teams, it shifts the burden onto someone else. That person must interpret, repair, question, or redo the content before useful work can continue.
What the study calls workslop
The researchers define workslop as low-value, meaningless AI-generated content shared inside organizations. In a survey of 1,150 full-time US workers, respondents estimated that 15.4 percent of all work content they receive falls into that category.
That figure matters because work content is not neutral. A memo, summary, plan, analysis, or message often creates follow-up work for someone else. If the content is unclear or shallow, the recipient has to spend time figuring out what is missing.
According to the study, workslop does not move in only one direction. Most of it comes from coworkers, at 40 percent. It also moves from staff to managers, at 18 percent, and from managers to staff, at 16 percent.
Respondents in professional services and tech said they encounter it even more frequently. That detail points to a practical risk for knowledge-heavy workplaces: the more a job depends on written communication and shared analysis, the more damaging weak AI output can become.
The hidden cost of low-value AI content
The researchers say each workslop incident takes survey participants nearly two hours to handle. That time can include reading, checking, clarifying, correcting, or replacing material that should have made the work easier in the first place.
Using those responses, the authors calculated what they call an "invisible tax" of $186 per employee every month. For a company with 10,000 workers, the study estimates more than $9 million in lost productivity each year.
The cost is easy to miss because it is not always recorded as a failed project or a direct expense. It can appear as a slower decision, a meeting that needs more preparation, a manager who has to untangle a vague update, or a teammate who spends extra time translating an AI-generated draft into something usable.
That is why workslop is different from an ordinary mistake. A weak human draft may still show clear intent or domain knowledge. AI-generated workslop can appear polished while still lacking useful judgment, which may delay the moment when recipients realize the work is not ready.
Morale and trust also take damage
The study reports that the emotional effect is significant. More than half of respondents, 53 percent, said workslop annoys them. Another 38 percent said it leaves them confused, while 22 percent said they felt offended.
Those reactions are not only about personal preference. In a workplace, sending low-value material can signal that the recipient’s time is less important than the sender’s convenience. That makes AI misuse a relationship problem as well as a productivity problem.
About half of respondents rated colleagues who send workslop as less creative, capable, or reliable. The study also found that 42 percent said they trust those coworkers less.
The workplace consequences continue from there. One in three survey participants said they want to work less with colleagues who send workslop. Another 34 percent said they reported these incidents to teammates or supervisors.
In other words, poor AI use can become visible socially before it becomes visible in a budget. A team may first notice frustration, extra checking, and reduced confidence in colleagues. The financial cost follows because those frictions make everyday collaboration slower.
Not all AI users behave the same way
The researchers identify two broad types of AI users at work: "pilots" and "passengers". The distinction is important because it suggests the issue is not AI use by itself, but how people choose to use it.
According to the study, pilots are proactive and optimistic. They use AI 75 percent more often at work, mostly to boost creativity. Nearly all pilots, 95 percent, also use AI outside of work.
Passengers use AI differently. The study says they mainly use AI to avoid work. That creates a clear management challenge: the same tool can either support better thinking or become a way to pass unfinished thinking to someone else.
This difference helps explain why blanket AI mandates can backfire. If employees are told simply to use AI more, without standards for quality, review, and accountability, the organization may increase the amount of AI-generated content without increasing the amount of useful work.
What managers can take from the findings
The authors advise against broad AI mandates. Instead, they recommend clear guidelines and a collaborative view of AI, with AI-generated and human work held to the same standards.
That recommendation follows directly from the problem. If workslop is costly because it shifts effort onto recipients, then the sender should remain responsible for quality before sharing AI-assisted material.
For managers, the study points to several practical expectations:
- AI-generated work should be reviewed before it is sent to colleagues.
- Teams need clear standards for what counts as useful work content.
- AI should support judgment, not replace responsibility for the final output.
- Human and AI-assisted work should face the same bar for clarity and usefulness.
The broader AI labor picture remains mixed in the source material. A Danish study of 25,000 employees found that AI hasn't meaningfully changed wages or hours. Anthropic's first AI Labor Market Index found that 36 percent of all jobs already use AI for at least a quarter of their tasks, with 57 percent of usage focused on support roles and 43 percent on automation.
A separate Stanford study found that employment for 22- to 25-year-olds in AI-exposed jobs dropped by 13 percent since late 2022, while employment for more experienced workers stayed flat.
Against that backdrop, workslop is a more immediate workplace signal. AI may change jobs in large and uneven ways, but inside teams, the daily question is simpler: does the output help the next person do better work, or does it hand them a mess to clean up?