AI 'Workslop' Is Costing Companies $9 Million a Year. The Employees Catching It Are the Only Real Producers Left.

Priya opens the report her teammate sent the night before and reads the first paragraph twice, because something in it refuses to resolve into meaning. The formatting is immaculate. The tone is confident. Three bullet points cite a client conversation from the previous week, quoted almost word for word. She pulls up the actual call transcript to grab the full quote for the summary slide. The conversation in the report never happened.

She spends the next hour and a half rebuilding the section from a real source, quietly, without flagging it to anyone, because flagging it feels smaller than simply fixing it and moving on. This is not the first time this month. It will not be the last.

She has a name for what just happened to her now, even if she has never said it out loud in a meeting. Researchers coined it workslop: AI-generated content that looks buttoned-up on the surface but has nothing real underneath, quietly transferring the actual thinking to whoever opens it next. Almost a year after the term entered the workplace vocabulary, the argument about it online has split into two camps. One says AI is hollowing out whatever discipline knowledge work had left. The other says people just need to prompt it better. Both camps keep missing the group of people who already solved this months ago, without posting about it, because solving it was never about the tool.

How big is the workslop problem, actually?

The number comes from a specific study, not a vibe. BetterUp Labs and the Stanford Social Media Lab surveyed 1,150 full-time US desk workers in September 2025 and found that 40 percent had received workslop, AI-generated content that looked finished but wasn't, in the previous month alone. Resolving each incident took an average of nearly two hours: tracking down the real source, rewriting the section, having the awkward conversation about what actually happened in that client call. Multiplied across a workforce, the study put the hidden cost at $186 per employee per month, or roughly $9 million a year for a company of 10,000 people. Not in software licenses. Not in training. In the invisible tax of people quietly redoing work that already appeared, on the surface, to be done.

HBR's coverage of the same research put the figure at 41 percent and added a detail that reframes the entire story: AI-led work processes have nearly doubled inside organizations since 2023, and usage has climbed right alongside the hype. Yet a widely cited MIT study found that 95 percent of organizations report no measurable return on their AI investment. Adoption went up. Output, the kind you can actually point to, did not follow. Workslop is one of the more concrete explanations for the gap: a tool marketed as a productivity multiplier is generating a second, hidden job, and that job is cleanup.

The word itself captures the specific, sinking feeling of receiving it. Coverage of the trend describes the sensation precisely: confusion first, when the content doesn't parse the way finished work should, then frustration, when you realize the sender used AI to generate volume and handed you the actual thinking to do. If that sequence sounds familiar, you have been workslopped, and so has almost half the desk-working population in any given month.

Why does everyone keep doing it if it clearly does not work?

Because it looks like productivity from the outside, and looking productive has quietly become its own reward.

The same research found sharp negative sentiment: more than half of recipients reported feeling annoyed by workslop, 38 percent confused, and 22 percent outright offended, and half viewed the senders as less creative, less capable, and less reliable going forward. That reputational cost is steep and largely invisible to the sender, who sees a document produced in four minutes instead of forty and reads that speed as discipline. It is not. Speed at producing volume and discipline at producing substance are different systems entirely, the same way multitasking and focus are different systems, and the workplace keeps rewarding the wrong one because the wrong one is easier to see.

Here is the detail that makes it a genuine enemy rather than a simple mistake: 18 percent of workers who use AI regularly admitted, in the same survey, to sending unhelpful, low-effort AI content themselves. The people most annoyed by workslop are, in meaningful numbers, also producing it. This is what Discipline as Aesthetic looks like in a knowledge-work costume. Posting about waking up early without actually training is the fitness version. Sending a beautifully formatted, four-minute report without reading it back is the exact same move, performed with a different tool.

Somebody still has to catch it, and reporting on the shift describes an entire generation of conscientious employees, disproportionately younger ones, quietly absorbing the role of unofficial quality control for everyone else's AI output, a job nobody assigned them and nobody is paying them for. Priya's ninety minutes did not appear on any timesheet. It came out of her actual work, the work with her name on it, because someone had to be the adult in the room and it was quietly assumed it would be her.

What does the discipline that actually stops this look like?

Not rejecting the tool. Protecting the standard.

The fix is not "use AI less." It is treating every AI draft the way a competent editor treats a first draft from a junior writer: useful raw material, not a finished product, and definitely not something that leaves the building with your name on it unverified. The specific, unglamorous protocol looks like this. Read the entire draft before sending it, not the first two lines and the formatting. Check every specific claim, quote, statistic, or client reference against its actual source before it leaves your outbox, the same ninety seconds of friction that would have caught Priya's fabricated conversation before it ever reached a colleague. Treat the draft as work to build from, never work that is done, because the moment you treat it as done, you have outsourced the one part of the job that was actually yours: the judgment.

This is Building-phase work in its most literal form, the kind nobody screenshots for social media because it produces no visible before-and-after. It is not exciting. It will not read as a productivity hack. It is the discipline equivalent of proofreading your own signature before you put it on a contract, done quietly, every single time, whether or not anyone is checking behind you.

It also happens to be exactly the differentiator the data points to. The same coverage of the AI productivity gap notes that the organizations and individuals actually capturing value from AI are the ones treating verification as a non-negotiable step in the process, not an optional one, while the 95 percent seeing no return tend to be the ones measuring success in drafts produced rather than work actually finished and trusted. Two people can use the identical tool and produce opposite outcomes, and the entire difference sits in the ten minutes one of them spent checking what the other one shipped blind.

Your name on a piece of work has always been a promise that you stand behind it. AI did not change that promise. It just made it much cheaper to break without anyone immediately noticing, and much more valuable to be the person who never does.

The debate about whether AI is good or bad for productivity will keep running exactly as long as it has run so far, which is to say indefinitely and without a winner. It is the wrong debate. The tool is neutral. The discipline around it is not.

Somewhere in every office right now, someone is reading a draft in full before sending it, checking one more citation against one more source, catching one more fabricated line before it becomes someone else's ninety minutes. Nobody is thanking them. The $9 million figure exists precisely because most people are not doing what they are doing.

Read the whole thing. Check the source. Ship what you would put your name on if someone asked you, out loud, whether you had verified it yourself.

Shine on!