What Is Workslop? The AI Buzzword Costing Companies Millions

Workslop is the name researchers at BetterUp Labs and Stanford's Social Media Lab gave to a very specific kind of bad work: AI-generated reports, emails, code, or slide decks that look complete and professional on the surface but don't actually contain enough substance to move a task forward — forcing whoever receives them to figure out what's missing, fix it, or redo it from scratch. The term, introduced in a Harvard Business Review article in September 2025, has quickly become shorthand for one of the most frustrating side effects of workplace AI adoption: not that AI makes bad work obviously bad, but that it makes bad work look good enough to pass the first glance.

What does "workslop" actually mean?

The term comes directly from the researchers who coined it: Kate Niederhoffer and Gabriella Rosen Kellerman of BetterUp, Angela Lee, Alex Liebscher and Kristina Rapuano, and Jeffrey T. Hancock, founding director of Stanford's Social Media Lab. Writing in Harvard Business Review, they defined workslop as "AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task."

The key word is "masquerades." Workslop isn't sloppy in the traditional sense — a typo-ridden, half-finished memo is easy to spot and easy to dismiss. Workslop is the opposite problem: it's formatted correctly, reads fluently, uses the right jargon, and sometimes even cites plausible-sounding sources, but when you actually try to use it, the substance isn't there. A report might summarize the wrong data. A block of code might run but miss the edge cases that mattered. A memo might restate the brief back at you without adding any actual analysis.

The researchers describe the real damage this does as shifting "the burden of the work downstream" — instead of the person who used the AI tool doing the thinking, the person who receives the output has to do it, usually without knowing in advance that the thinking was never done in the first place.

Where the term came from

Workslop was introduced in a Harvard Business Review article published in September 2025, based on original survey research from BetterUp Labs — the research arm of the workplace coaching company BetterUp — conducted in partnership with Stanford's Social Media Lab, the research group founded and directed by communication professor Jeffrey Hancock.

The authors frame workslop as a symptom of a larger mismatch. Generative AI adoption inside companies has climbed fast — Gallup has tracked AI use at work roughly doubling since 2023, and Accenture has reported that the number of companies running fully AI-assisted processes nearly doubled in a single year. But that adoption hasn't translated into obvious productivity gains: the researchers point to an MIT Media Lab finding that roughly 95% of organizations that have invested in generative AI report seeing no measurable return. Workslop, they argue, is one plausible reason why — not because the AI tools don't work, but because of how people are using them: to generate the appearance of finished work rather than to actually do the work.

BetterUp followed up with its own blog post, "The hidden cost of AI 'workslop' — and how leaders can fix it," which lays out the survey data behind the HBR piece in more detail.

How common is workslop, really?

BetterUp's research is based on a September 2025 survey of 1,004 full-time U.S. desk workers. The headline numbers:

• 40% of employees said they'd received workslop from a colleague in the past month.
• On average, workers estimated that 15.4% of the work they receive counts as workslop.
• 54% of managers said they'd received workslop, compared with 38.5% of individual contributors — middle managers appear to absorb a disproportionate share of it.
• 53% of respondents admitted that at least some of the work they themselves send may qualify as workslop.
• Workslop doesn't just flow downward from leadership. The survey found it moves sideways between peers most often (40% of reported instances), followed by upward from employees to managers (19%) and downward from leaders to teams (16%).

That sideways flow is one of the more striking findings: workslop isn't mainly a story about executives dumping AI-generated busywork on junior staff. It's just as often one coworker handing another coworker something that looks finished but isn't.

What workslop looks like in practice

Because the whole point of workslop is that it looks fine at first glance, the examples tend to be mundane rather than dramatic:

• Reports and research summaries that are well-formatted and confidently written but misrepresent or oversimplify the underlying data, or cite sources that don't say what the summary claims they say.
• Code that compiles and appears to do what was asked, but that a reviewer later finds doesn't handle real edge cases, lacks tests, or quietly reimplements something that already existed elsewhere in the codebase. Developers evaluating tools like Claude Code or other AI coding agents will recognize this as the gap between code that "runs" and code that's actually production-ready.
• Memos and strategy documents that restate the original prompt or brief in more polished language without adding new analysis, risk assessment, or a real recommendation.
• Slide decks that are visually polished — on-brand, well-structured, full of bullet points — but whose content doesn't actually support the conclusion on the title slide.
• Emails that read as thoughtful and complete but turn out, on a closer read, to not actually answer the question that was asked.

In every case, the tell isn't how the output looks. It's whether someone, at some point, actually engaged with the problem — or just asked a model to produce something that resembled a solution to it.

The real cost: hours, dollars, and trust

BetterUp's survey tried to put a number on what workslop actually costs. Respondents who'd received workslop said it took them an average of 1 hour and 51 minutes to figure out what was wrong and fix it — about 20 minutes longer, they estimated, than it would have taken the original sender to simply do the work properly themselves. BetterUp translates that into what it calls an "invisible tax" of roughly $186 per affected employee per month, and extrapolates that to more than $9 million a year in lost productivity for a 10,000-person company. Those are BetterUp's own modeled estimates based on self-reported survey data, not independently audited financial figures, so it's worth treating them as directional rather than precise.

The harder-to-quantify cost may be the damage to working relationships. Recipients of workslop reported feeling annoyed (54%), frustrated (46%), confused (38%), and even offended (22%). Roughly half said the experience made them view the sender as less creative, less capable, and less reliable; 42% said it made the sender seem less trustworthy; more than a third said it made the sender seem less intelligent. Nearly one in three said they'd be less willing to work with that person again. In other words, workslop doesn't just cost time — survey respondents said it quietly erodes the sender's credibility, every time it happens.

Workslop vs. "AI slop": what's the difference?

Workslop is often confused with the older and broader term "AI slop," but the two describe different problems. According to Wikipedia's entry on the term, AI slop is "digital content made with generative artificial intelligence that is perceived as lacking in effort, quality, or meaning," usually mass-produced "to gain advantage, to earn money, or to deceive people" — think low-effort AI images and videos clogging social feeds, or spammy AI-written articles chasing search traffic.

Workslop is narrower and more internal. It isn't aimed at the public or at an algorithm; it's AI-generated content passed between coworkers inside an organization, under the normal expectation that it represents real, finished work. AI slop is a problem for platforms, readers, and the broader internet. Workslop is a problem for the person sitting next to you in a meeting who now has to quietly redo what you sent them — and, per BetterUp's survey, may now also think a little less of you for having sent it.

Why workslop keeps happening

Part of the explanation is simple mismatch between incentive and effort. Many employees are under real or perceived pressure to show they're "using AI" — whether that's a formal company mandate or just the general expectation that AI tools should be making everyone faster. Generating a polished-looking document with an AI assistant takes minutes; doing the underlying thinking that would make that document actually useful still takes real time. When speed is rewarded and quality isn't checked, the incentive tilts toward quantity over substance.

It also helps explain why workslop can be hard to catch in the moment: AI tools are, by design, good at producing fluent, well-organized, confident-sounding text and code. That's exactly the surface signal people normally use to judge whether something is finished. Buzzwords like "vibe coding" — writing software by describing what you want rather than reviewing the code line by line — capture a related dynamic: it's now trivially easy to produce something that looks like a finished product without anyone having verified that it actually is one.

How to avoid producing workslop

The BetterUp/Stanford researchers and the follow-up coverage of their work converge on a few practical habits:

• Treat AI output as a draft, not a deliverable. If you wouldn't be comfortable explaining and defending every claim or line yourself, it isn't ready to send.
• Keep a human accountable for the final product. Someone — ideally the sender — needs to have actually read, checked, and understood the output before it goes out, not just skimmed it for formatting.
• Give the tool real context, not just a one-line prompt. Thin prompts tend to produce thin, generic output that looks plausible but misses what the task actually needed.
• Be honest about what AI did and didn't do. Flagging "I used AI to draft this, here's what I verified" is more useful to a colleague than letting them discover the gaps themselves.
• Match the tool to the task. Teams comparing options like GitHub Copilot and Claude Code for engineering work, for instance, still need a review step regardless of which assistant produced the first draft.

How to spot workslop before it wastes your time

From the receiving end, a few quick checks can save the 1-hour-51-minute average cleanup time BetterUp's respondents reported:

• Ask a specific follow-up question. Real work usually survives a pointed question about a detail; workslop often falls apart the moment you ask "why did you choose this number" or "what does this function do if the input is empty."
• Check for genuine specificity. Polished generalities — language that could apply to almost any project — are a classic workslop tell. Real analysis engages with your actual data, your actual codebase, your actual constraints.
• Verify sources and numbers independently before repeating them further up the chain, especially in reports and research summaries.
• Give direct, specific feedback rather than just quietly redoing the work yourself — the researchers note that silently absorbing the extra effort just lets the pattern continue.

Can companies actually fix this?

The researchers' own answer, in follow-up commentary, is that workslop is a management and culture problem more than a technology one — and that it isn't inevitable. Their recommendations for organizations include setting clear, explicit guidelines for what "acceptable" AI-assisted work looks like, rather than leaving it to individual judgment; encouraging what BetterUp Labs calls a "pilot mindset" — approaching AI tools with both optimism and a sense of personal responsibility for the outcome, rather than treating the tool as a replacement for judgment; and rebuilding basic collaboration habits, like clear prompting, context-sharing, and feedback, that tend to erode when people start treating AI output as a shortcut around them rather than an input to them.

What to watch next

Workslop is likely to stay part of the workplace AI vocabulary for a while, for the same reason the HBR researchers flagged it in the first place: adoption of generative AI at work keeps climbing, but the return on that adoption remains uneven, and workslop offers one concrete, measurable explanation for the gap. Expect more companies to start formalizing AI-use guidelines explicitly aimed at this problem — spelling out when AI drafts need a named human reviewer before they go out — rather than leaving appropriate use entirely up to individual employees. It's also likely that "workslop" gets applied more loosely over time, the way many workplace buzzwords do, so it's worth holding onto the original, narrower definition: AI-generated work that looks finished but isn't, specifically inside a workplace context, as distinct from the broader flood of low-quality AI content washing over the public internet.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage