Published on November 18, 2026 | 6 min

"Use AI," they said. "Your work will be perfect," they said.

Enterprise IT leaders wasted no time integrating AI into their workflows, lured by the promise of polished work in a fraction of the time. But the checking and correcting that follows is far less sexy. How is that work measured, if it's measured at all?

Ideally, any work product—like a status report, vendor summary, or policy draft—is read, fact-checked, corrected, and reworded by a human before anyone relies on it. When these steps are skipped, the result can be what researchers call “AI workslop”: AI-generated work that looks fine at first glance but is incomplete, inaccurate, or missing context, creating more work for the next person in the workflow.

Until leaders measure the time teams spend verifying AI-assisted work, reported productivity gains are incomplete. Here’s why that cost is easy to miss, who ends up it, and what leaders can measure instead.

The hidden cost and how far to trust the numbers

AI workslop may seem like a classic case of garbage in, garbage out, but the truth is more complicated. Generative AI tools can fill gaps with plausible-sounding details that were never in the prompt, so fixing the output takes more than a quick proofread.

That scrutiny adds up: in a September 2025 survey of 1,150 U.S. desk workers by BetterUp, a coaching and workforce-development company, and Stanford University's Social Media Lab, each instance of AI workslop took an average of nearly two hours to resolve. And these instances were far from rare, with 40% of respondents saying they had received workslop the previous month. Based on self-reported salaries and time estimates, fixing workslop cost roughly $186 a month per affected employee—more than $9 million a year for a 10,000-person company.

Admittedly, these numbers warrant a closer look. The survey is more than a year old, covers only U.S. desk workers, and relies on respondents' memory and estimates, not their colleagues' actual output. BetterUp also markets its coaching as part of the remedy for workslop, which is a reason to weigh the data and the remedy separately. Still, the core finding is hard to dismiss: workslop is a mainstream workplace issue, and a costly one at that.

Where the burden lands and why it's a leadership issue

AI workslop shifts effort from the person who produces it to the person who receives it. In the BetterUp and Stanford data, workslop passed most often between peers, but it also moved up and down the org chart: 18% of respondents received it from direct reports, and 16% from managers or higher-ups.

This is where the costs go beyond time and money to the interpersonal. About half of respondents viewed colleagues who sent workslop as less creative, capable, and reliable. Forty-two percent saw them as less trustworthy, and 37% considered them less intelligent. BetterUp and Stanford trace part of the problem to leadership: blanket mandates to use AI everywhere model a lack of discernment and invite thoughtless copy-and-paste use. They also found that workers with high agency and optimism (“pilots”) used AI more purposefully, while “passengers” were more likely to use it to avoid doing work.

Leaders can’t fix what they can’t see, and workslop is hard to spot, partly because workers still tend to keep their AI use private, according to Kate Niederhoffer, chief scientist at BetterUp. Recipients may not be able to tell what was AI-generated and what, if anything, what was checked. Setting a norm for disclosure, and easing the pressure to produce faster, starts with leadership.

What to measure and how to set standards

Thomas DeFabrizio, Americas CFO at workforce solutions company Vertage, said in an email that if he were starting an AI deployment from scratch, he would spend more time measuring how the work gets done today before turning anything on. “Without that baseline, 12 months later you end up arguing about whether AI created value because nobody can remember what ‘before’ actually looked like.” He suggests starting with basics such as how long the work takes, how many people touch it, the rework and error rates, and the unit of output the team is trying to improve.

Translated into workslop equivalents, leaders could start with how often employees received it in the past month, then track the hours spent fixing each instance as well as the share of AI-assisted deliverables needing rework. Finally, they could count who absorbs the cost, including reviewer and recipient hours as well as the sender’s hours. Several of these metrics will rely on self-reports, so they're better for tracking direction over time than for exact accounting.  

Measuring the problem is only the first step, though. The researchers from BetterUp and Stanford urge leaders to set clear norms and hold AI-assisted work to the same standard as human work. In practice, that could include:

  • A definition of "done": work that's been read, fact-checked, corrected, and revised by a human before anyone relies on it.
  • A disclosure norm so colleagues know when AI was used to produce work and what a human checked.
  • Explicit ownership, meaning whoever sends the work is accountable for it. 

Make the invisible work visible

Niederhoffer notes that when employees keep their AI use private, their firms miss out on the innovation too, because nothing that works gets shared. The same disclosure that helps catch bad work early also lets quality work spread.

DeFabrizio shares his own way of measuring this invisible work, and it’s a useful note to end on: “If I cannot explain what the AI replaced, what people still have to do, and what changed economically, I am not ready to call it ROI.” Until leaders can say who checks AI-assisted work and how long it takes, the productivity gains they report are incomplete.

Frequently asked questions

What is AI workslop?

AI workslop is AI-generated work that looks polished but lacks the substance, accuracy, or context to be useful, so a colleague has to fix it or redo it. It shifts the cost of the work onto the person who receives it.

What's the difference between workslop and AI slop?

AI slop refers to low-quality AI-generated content flooding public platforms such as social media. Workslop is the workplace version: output that passes between colleagues looking like finished work, with the cleanup left to the recipient.

Who coined the term "workslop"?

Researchers at BetterUp Labs and the Stanford Social Media Lab, including BetterUp chief scientist Kate Niederhoffer and Stanford Social Media Lab founding director Jeff Hancock, introduced the term in a September 2025 Harvard Business Review article.

How much does workslop cost a company?

Researchers estimated that each instance of workslop took nearly two hours to resolve, which came to roughly $186 a month per affected employee. Assuming about 41% of a 10,000-person company’s workforce is affected, that’s more than $9 million a year.

What percentage of employees say they've received workslop?

In BetterUp and Stanford's September 2025 survey of 1,150 U.S. desk workers, 40% said they received workslop in the previous month.

How can you tell if AI-generated work is workslop?

The clearest indicator of workslop is how much work it creates for the person who receives it. Common giveaways include polished formatting around thin content, claims without sources, generic details where specifics belong, and missing context. If using it means asking clarifying questions, rechecking facts, or redoing it, it's likely workslop.

Does workslop mean AI actually reduces productivity?

Not necessarily. The research measures the cost of low-quality output, not AI's overall effect on productivity, and BetterUp's researchers say workslop spreads when AI becomes a crutch instead of a tool for collaboration. The takeaway for leaders is that time saved is only part of the equation, because the checking and correcting that follows has to be counted too.

What can CIOs and IT leaders do to prevent workslop?

Start by measuring it: how often employees receive it, how long it takes to fix, how often AI-assisted deliverables need reworking, and who absorbs the cost. From there, standards help. Define what "done" means for AI-assisted work, make disclosure of AI use routine, hold AI-assisted and human-written work to the same bar, and make the sender accountable for what they hand off.

Related reading

Lauren Spiller

Lauren Spiller

Enterprise Analyst, ManageEngine

Want to read<br>this article on the go?

Want to read
this article on the go?

Do it on the ManageEngine
Insights app.

Thanks for subscribing to ManageEngine Insights. If you do not receive our welcome email, please check your SPAM folder.