The workslop crisis

A word for it

Researchers at Stanford and BetterUp have given us a word for something most knowledge workers already recognize: workslop. AI-generated content that looks polished but lacks substance. Slick slides. Lengthy reports. Overly condensed summaries. Code without context. Work that creates the appearance of productivity without the thinking behind it.

The study surveyed 1,150 full-time U.S. desk workers last September. Forty percent reported receiving workslop from colleagues in the previous month. Each incident took an average of two hours to resolve. That scales to about $186 per person per month in invisible rework, or roughly $9 million per year for a company with 10,000 employees.

The emotional data is worse. Fifty-three percent of respondents said they were annoyed to receive AI-generated work. Twenty-two percent were offended. Close to half said it made them view the sender as less creative and less reliable. One project manager described it plainly: "Receiving this poor quality work created a huge time waste. I had to take on effort to do something that should have been her responsibility."

Workslop doesn't just waste time. It erodes trust between the people doing the work.

What it looks like from the receiving end

Joe Depa has seen enough AI-generated presentations to know the signs. As EY's global chief innovation officer, he's on the receiving end of a lot of polished slides. The tells, he says, are consistent.

The writing is too smooth. No shifts in pattern, structure, or flow. It leans on buzzwords and corporate language. Topics get addressed too broadly, with little consideration for the specific audience in the room. And there's a particular kind of hedging that AI does by design: steering away from clear recommendations, presenting vague alternatives instead. "Vagueness or general statements that don't really tell you anything," Depa told Business Insider.

His advice to teams is simple. Write your own content first, then use AI to refine it. "If you write it yourself first and then ask for the enhancement using AI, I feel like that's much more productive." The order matters. Most people start with AI and try to add their thinking afterward. Depa is saying it should be the opposite.

The mirror

The most important data point in the workslop conversation isn't about the content itself. It's about the people creating it.

Anthropic's Economic Index, published earlier this month, measured something deceptively simple: the years of education needed to understand a user's prompt and the years of education needed to understand the AI's response. Across 117 countries, the correlation between the two was 0.925. Across U.S. states, 0.928. Near-perfect.

In plain terms: the AI calibrates to you. Give it a sophisticated prompt and you get a sophisticated response. Give it a shallow prompt and you get polished shallow output. The model doesn't add depth you didn't bring. It mirrors the quality of your input with unsettling precision.

This reframes the entire workslop problem. The issue isn't that AI produces bad content. It's that AI faithfully reproduces the quality of thinking behind the request. People generating workslop aren't using a broken tool. They're getting exactly what they asked for.

Even the makers notice

The tool makers are seeing the same thing from a different angle.

When OpenAI released GPT-5.2 in December, users described the writing as "unwieldy" and "hard to read." Within a day, forums filled with complaints about flatter tone, worse conversational quality, and a noticeable regression from earlier models. OpenAI's own system card acknowledged the dip, noting quality issues particularly in the model's instant mode.

The regression wasn't accidental. OpenAI chose to prioritize coding, reasoning, and engineering capabilities. GPT-5.2 became the first model to exceed 90% on ARC-AGI-1 and hit a perfect score on AIME 2025 mathematics. Writing got worse because the company decided to make other things better.

That tradeoff tells you something about where the pressure is. Benchmarks reward technical capability. Organizations need writing that sounds human. The models are optimizing for the former while most workplace AI use demands the latter, and the gap between those two is where workslop breeds.

The institutional failure

Here's the part nobody wants to hear. Workslop is a leadership problem.

Most organizations rolled out AI the same way. Someone in the C-suite got excited. A mandate went out: use AI, be more productive. Licenses were purchased. Maybe there was a lunch-and-learn. Then everyone was left to figure it out on their own.

What didn't happen? Three things.

No culture. The Stanford researchers specifically recommend that leaders model thoughtful AI use. Not just use AI, but use it well, visibly, and talk about what works and what doesn't. Most organizations have no feedback norms around AI output quality. If a colleague hands you AI-generated slop, there's no established way to flag it without making it personal. So people absorb the rework silently, and the problem stays invisible to everyone except the person cleaning it up.

No process. There are no review standards for AI-assisted work in most companies. No equivalent of code review for AI-generated reports. No shared understanding of what "good enough" means when AI is involved. The Stanford team calls for cultivating a "pilot" mindset, where AI is a tool that requires active supervision rather than a replacement for thinking. Most organizations haven't defined that standard, so each person invents their own.

No accountability. When nobody owns the quality of AI-generated output, nobody is responsible when it's bad. The tool gets blamed. "AI wrote it" becomes a deflection, not an explanation. But as the Anthropic data shows, the output is a direct function of the input. If the work is shallow, someone submitted a shallow prompt and didn't review the result. That's a skills gap, a standards gap, or both. Either way, it's a management problem.

The uncomfortable math

Forty percent of knowledge workers have received workslop in the past month. The correlation between user sophistication and output quality is near-perfect. Put those two findings together and the picture gets hard to look away from.

A significant portion of the workforce is using AI to generate content they couldn't meaningfully evaluate. Not because they're lazy. Because they're doing what they were told to do, with tools they were given and training they weren't.

The $9 million annual cost isn't a technology failure. It's the price of mandating a tool without building the institutional infrastructure to use it well. Companies that figure this out will treat AI proficiency the same way they treat any other core competency. They'll build review processes. They'll create feedback loops that catch workslop before it reaches someone else's inbox. They'll hold the humans accountable for the quality of what the AI produces, because the humans are the ones determining that quality.