Getty Images; Tyler Le/BI
Office workers everywhere are awash in “workslop.” This is the term researchers are using to call AI-generated content that might look like it completes a task at work: pretty on paper, well-organized and neatly formatted, but lacks substance upon a closer read — and it often becomes a headache for the person receiving the work.
The term was coined in research published this month from the Stanford Social Media Lab and BetterUp, a professional training and coaching company. Surveying 1,150 desk workers in the US, the researchers found that 40% of respondents said they believed they had received this sort of AI-made sloppy work from their colleagues, which can take on many deceiving forms, like visually pleasing slide shows, long reports, or code that doesn’t actually translate to a thoughtful product. On average, employees surveyed said they spent nearly two hours sorting through or cleaning up each instance of workslop.
In short: Some meetings might be better off as emails, and some AI-generated emails might be better off left in the drafts.
“We see that people are just sort of proliferating these documents, offloading the work onto another human, and then having this unintended consequence — there’s a whole lack of trust as a result,” says Kate Niederhoffer, head of research at BetterUp Labs. It’s not that AI use at work can’t be effective, she says, but “using it without high enough agency” leads to issues. That’s a problem not just for the bottom line, but for relationships between employees. If your coworker foists lengthy, useless docs generated by AI onto your desk, it can feel like they’re not pulling their weight or not capable of doing the work themselves. “It’s the human dynamic that suffers because of our usage of this tool,” says Niederhoffer.
The likes of Zoom CEO Eric Yuan, Bill Gates, and JPMorgan Chase CEO Jamie Dimon have estimated that with advancements in generative AI, white collar workers may soon only clock in three or four days a week. In Gen AI’s current state, few of us are working less, and for some, it’s creating more work or shifting how they spend their days rather than making them more productive. A recent MIT Media Lab report found that 95% of AI pilot programs have not seen measurable savings or boosts in profit.
Instead of freedom, some desk workers find themselves slogging through with low-quality work; experienced software engineers are now debugging code, graphic designers are making generative-AI images look like something humans actually want to see, and writers are editing the words that large language models spit out for accuracy, and then rewriting it to cover up ChatGPT’s telltale signs. When people think they’ve received messages generated by AI, they don’t like them. A 2023 study from Cornell University found that Google-written smart replies could make conversations more efficient and led to more positive language, but those who received the messages held more negative opinions of the senders of algorithmic responses.
No doubt there have always been workers who squeaked by while another clocked late hours and cleaned up messes, but gen AI has streamlined the process of phoning it in.
Some meetings might be better off as emails, and some AI-generated emails might be better off left in the drafts.
On Upwork, a job platform for freelancers, graphic design, English, and fact-checking were among the top 10 most in-demand AI skills in August. Demand for freelancers with content writing or language tutoring and interpretation were up 15% and 162% respectively, according to Upwork’s Research Institute. These were among skills most at risk of automation thanks to large language models, but the high-demand shows how companies need expert freelancers in this skillsets, and aren’t leaning on AI to replace them, according to Upwork’s analysis. “Humans must remain in the loop from the start to ensure that AI work products are delivering the levels of quality, accuracy and usability that we expect and need them to,” Teng Liu, an economist for Upwork’s Research Institute, tells me in a statement. “Otherwise, the need for ‘AI clean-up’ jobs does frequently occur.”
Big Tech companies have turned to generative AI for coding, and job listings for junior software developers have fallen. AI has killed the coding test and given rise to vibe coding. A deeper look shows that AI might not be making developers more efficient yet, and is more often shuffling around how they spend their time, moving the bulk of their focus from writing code to reviewing it. In 2024, Uplevel, a software development company, studied 800 software developers and compared the productivity of those with GitHub’s Copilot access to those without. They found that the developers who used Copilot did not become more efficient or less burnt out, but did have bugs in code 41% more frequently than those without access to the tool (GitHub published its own smaller study a month later, showing that developers who used Copilot wrote about 18 lines of code without error, compared to 16 lines for those without). The effects of AI on productivity for engineers has been uneven, says Amy Carrillo Cotten, Uplevel’s director of customer transformation. “For a lot of engineers, the only thing that looks different is where they spend their time, not exactly how much time it took.”
Still, the ease and availability of AI coding tools have led customers to think engineering and design work should be faster and cheaper, says Minh Phan, founder of Studio Init, a software development company. Phan says there’s still complex work to be done by human developers and designers to get to an end product, and he has to convince some customers that AI for coding isn’t the lightning bolt they think it is. “What they want is a Mercedes, but the prompt is giving a bicycle,” he says.
More people using tools like ChatGPT also means more sloppy content being sent in cold outreach. Caitlin Ner, director of operations at PsyMed Ventures, a VC firm that focuses on mental and brain health startups, says her firm has received more messages from founders and pitches that seem AI-generated; either they’re spammy and don’t fit in with the firm’s niche, or they’re so like other pitches that it’s hard to decipher how well the founder can communicate if the product stands out. As a small firm, it’s a waste of time to read such generic pitches, and a bigger waste of time to start conversations only to find the founder isn’t a good fit on a call, Ner tells me. “When you’re an early stage investor, you’re essentially really taking a bet on the founder,” she says. “It’s very easy to tell if you’re pulling something just templated, versus you have a technology that actually stands out and you’re portraying that with your own words.”
A marketing worker named Shawna, for example, recently got a slew of drafted emails from a coworker at 4 p.m. meant to be sent out the next day. It’s a standard part of the gig, but these emails needed a lot of extra love: They lacked her company’s formatting, style, and tone, she tells me. That’s because Shawna’s colleague didn’t actually write the drafts, but prompted a chatbot and then copied and pasted its output without much thought.
“I had to drop everything and scramble at the end of the day to fix this series of emails,” says Shawna, who asked I not use her full name because she’s not authorized by her company to speak about this. Shawna herself uses ChatGPT to get her work done faster, and has spent time learning the best ways to prompt it, pulling together the best outputs and throwing out the worst. But she says her coworkers don’t always put as much editing into the output. While the person who sent the sloppy emails could blissfully check the item off their to-do list, Shawna had to punch up the writing and fact-check. “It made their life easier, then it made my life harder. I had to do the critical thinking on their behalf.”
There are already too many notifications, systems, processes, and standing meetings that eat up worker’s time. AI “tools are being used because people feel like there’s just too much,” says Jeff Hancock, director of the Stanford Social Media Lab. Some AI tools have helped people organize and filter their notifications, cutting down on the unnecessary pings they receive each day. But when used without thought, AI can ironically add to the flood of too much information. In all the hype around the AI boom, many are failing to use the tools with the most important question in mind, and that’s: “What are people’s needs?” Hancock says. “I think that is so often ignored when talking about AI.”
There’s a lot that all of us have to read each day — texts, emails, Slack messages, recipes, that book that’s been sitting on your nightstand neglected. Using AI in responses might seem like a fix to take something off your plate, but an overreliance on it can derail conversation. Writing, coding, or designing is about communicating ideas to others. If that’s outsourced to gen AI with little to no human oversight, there may be little value for the human on the other end who has to read it.
Amanda Hoover is a senior correspondent at Business Insider covering the tech industry. She writes about the biggest tech companies and trends.
