Coworker Uses AI for Everything? How to Handle AI Slop

What is AI slop at work?

The researchers who named it call it workslop: AI-generated content that looks like good work but doesn't have the substance to move the task forward. A 12-slide deck with no decision in it. A project update that summarizes the project instead of updating it. A reply to your question that answers a different, easier question.

The BetterUp and Stanford survey, written up in Harvard Business Review in September 2025, put some numbers on it:

• People estimated that 15.4% of the work they receive is workslop.
• Each incident took about two hours to deal with (1 hour 51 minutes in BetterUp's own write-up of the survey, 1 hour 56 minutes in HBR's).
• The researchers valued that lost time at $186 per employee per month, or more than $9 million a year for a company of 10,000 people.
• Most of it moved sideways: 40% came from peers, about 18–19% went up to managers (the figure differs slightly between the two write-ups), and 16% came down from leaders.

A newer survey from Founder Reports (April 2026, 2,078 US workers) found that 45% of workers had fixed or redone a coworker's work because it relied too heavily on AI. At companies that require AI use, that jumped to 73%. Mandates, it turns out, produce a lot of output.

The cost goes beyond time. About half of the BetterUp respondents said they saw the sender as less capable and less reliable afterward, 42% saw them as less trustworthy, and nearly one in three said they'd rather not work with that person again. Your coworker probably thinks they're saving time. Their reputation is paying for it.

How can you tell a coworker's work is AI slop?

Everyone has a list of tells by now. The em dashes (a real loss for those of us who loved them before the robots did). Three-item lists, every time. Headings on a two-paragraph email. Words nobody says out loud, like 'delve,' 'tapestry' and 'robust.' Things that happen 'quietly.' Wikipedia's volunteer editors keep a long, very good guide called Signs of AI writing if you want the full catalog.

Style tells are weak evidence, though. Plenty of people write in bullets, and some of us were using em dashes in 2009. The better test is what the work does:

• It's long, but you can't find the answer, the decision or the number you needed.
• It's confident about facts that turn out to be wrong, or cites a source that doesn't say that (or doesn't exist).
• It's generic. Swap your company's name for a competitor's and nothing breaks.
• It ignores context the person definitely had: last week's meeting, the client's actual request, the budget.
• When you ask a follow-up question, they need a few minutes to "check."

Should you say something?

Not always. Before you do, ask yourself three questions.

• Does it actually affect you? If a coworker's AI-written status update is bland but accurate, that's a style choice, and it's between them and their manager. If you're spending your afternoon rebuilding their analysis, it's your business.
• Is it wrong, or just obviously AI? Being annoyed that something sounds like ChatGPT isn't a work problem. Made-up numbers, missed requirements and work you have to redo are.
• What does your company want? Many employers now encourage AI use, and some measure it. Criticizing the tool can make you look like the problem. Criticizing the output never does.

If the answers are yes, wrong, and 'they'd care about the quality,' say something. Fixing it silently feels kinder, but it teaches your coworker that the slop works, and it moves their job onto your plate permanently.

How do you talk to a coworker who uses AI for everything?

Keep it private, keep it specific, and keep AI out of the first sentence. 'Did you just ChatGPT this?' puts them on the defensive, and now you're arguing about honesty instead of the report. Talk about the work in front of you.

Start with a question that makes them engage with their own output. 'Can you walk me through how you got to this number?' works better than any accusation; either they can explain it (great, maybe it's fine) or they realize mid-sentence that they can't.

Then name the specific problem and the effect it had on you: 'The vendor list includes two companies we stopped using last year, so I had to rebuild it before the client call.' Specific, factual, no adjectives.

Finally, agree on what you need next time. Maybe it's a one-line summary at the top. Maybe it's 'check the numbers against the dashboard before you send it.' Make the ask small and concrete enough that they can't misunderstand it.

Some lines that work, sorted by what's going on:

For more lines in the same spirit, we have a whole list of polite ways to tell someone they're stupid at work. Every one passes the HR test.

If the conversation goes nowhere and the slop keeps coming, go to your manager. Don't frame it as 'they use AI.' Frame it as workload and risk: 'I've spent about six hours this month redoing deliverables from this project. Here are three examples. I'd like to figure out a better handoff.' Bring the examples; managers act on evidence, not vibes. Our office politics survival guide covers how to raise it without looking like you're throwing someone under the bus.

What if your boss uses ChatGPT for everything?

Then you're getting slop from above, which is harder, because you can't really send your manager feedback on their prose. (In the BetterUp survey, 16% of workslop came down from leaders, so you're not alone.)

Focus on clarity, not authorship. When a boss sends you a 600-word AI-flavored brief, you don't need to know who wrote it; you need to know what they want. Ask for it directly:

• "Before I start, can you tell me the top priority here in one line?"
• "This lists eight goals. If I can only hit two by Friday, which two?"
• "I want to make sure I get this right. What does 'done' look like for you?"

If your boss uses AI to write your performance review or make decisions about you, it's fair to ask how the conclusion was reached and to respond with your own record of results. Keep notes. For the bigger picture on managing up, see how to deal with a difficult boss.

What should a manager do when an employee relies too much on AI?

Treat it like any other quality problem, because that's what it is. If an employee's output looks polished but falls apart under questions, the conversation is the same one you'd have about any missed standard: what was expected, what arrived instead, what it cost the team, and what's getting in the way.

A few things help more than banning anything:

1. Set clear expectations for AI use. Which tasks are fine to draft with AI, which need a human check, and which shouldn't go near it (anything confidential, legal or about a specific person). BetterUp's first recommendation was exactly this: clear guidelines.
2. Make people own their output. "AI wrote it" is never the excuse. If your name is on it, you checked it.
3. Ask them to explain their work. A two-minute "talk me through this" in a one-on-one shows you very quickly whether they understand what they sent.
4. Look at your incentives. If you reward volume and speed, you'll get more slop. If you praise someone for a short update with one clear decision in it, you'll get more of those.

If nothing changes after a clear conversation, handle it like any other performance issue. Our guide on how to deal with an underperforming employee has the scripts, and HR Approved Ways to Lead Without Losing Your Mind covers the rest of the job nobody trained you for.

The tools in the next section work for teams too. A shared word list and a shared self-check prompt do more than a memo telling people to 'use AI responsibly.'

How do you make sure you're not the one sending slop?

Uncomfortable fact: in the BetterUp survey, 53% of people admitted some of their own work might be workslop. Odds are decent that someone has rolled their eyes at something you sent.

Most slop comes from a thin prompt, not a lazy person. Ask a chatbot to 'write a project update' and nothing else, and it has to guess, so it writes the average project update on the internet. Five fixes, roughly in the order you'd set them up. Copy whatever is useful.

1. Cut the words that give AI away

Chatbots lean on a small set of words far more than people do. One or two in a document mean nothing. Five in a paragraph and readers stop trusting the rest. Paste this list into your prompts as a ban list, and search your drafts for it before you send.

Phrases and habits to cut too:

• "I hope this email finds you well." Start with the point.
• "It's important to note that…" Just say the thing.
• "In today's fast-paced world…" Delete the sentence.
• "It's not just X, it's Y." Say what it is.
• "Great question!" Answer the question.
• "Let me know if you have any other questions!" Your coworker knows they can ask.
• Em dashes in every paragraph, and lists that always come in threes.
• Headings and bullets on anything shorter than a page.

2. Give the AI your voice before you give it the task

A chatbot with no instructions writes like a press release. Write a short voice brief once and paste it in at the start, or save it where your tool keeps standing instructions (custom instructions in ChatGPT, a project or style in Claude, a Gem in Gemini). Adjust this one until it sounds like you:

My writing voice. Follow this for everything you write for me.

- I write for busy coworkers who read on their phones. Put the main point or the ask in the first two sentences.
- Short sentences, plain words. Write the way I'd say it out loud to a coworker.
- Friendly, direct, a little dry. Never gushing, never salesy.
- Use real numbers, names and dates from what I give you. If something is missing, write [NEED: what's missing] instead of making it up.
- No headings or bullets in anything under 150 words. No greeting beyond "Hi [name]," and no sign-off beyond "Thanks."
- No em dashes. Don't use any word or phrase from this list: [paste the list above].
- Shorter is better. If it can be one paragraph, make it one paragraph.
- Here are two things I wrote that sound like me: [paste two real emails or messages you're happy with].

Those two samples at the end matter more than the rest of it put together. A model copies examples far better than it follows adjectives like 'friendly.'

3. If the output is generic, give it more context and data

Generic output is a sign the AI didn't have enough to work with. It can't know about last week's meeting, the client's budget or the thing your director hates unless you tell it. Compare these two prompts for the same Friday update:

Write a project update for the website redesign.

Write a Friday update on the website redesign for my director, Priya, who reads these on her phone. One short paragraph.

- What happened: homepage and pricing page went live Tuesday. The checkout page slipped a week because our payment vendor changed their API.
- Numbers: homepage bounce rate is 41%, down from 52%. We've spent $38k of the $50k budget.
- What I need: a yes or no by Tuesday on paying the vendor $2,500 to rush the fix.
- Leave out: design history, team names, anything from last week's update.

The second prompt takes two minutes longer to write and saves the twenty you'd spend rewriting. A good prompt covers who's reading, what actually happened, the real numbers, the decision or action you need, how long it should be, and what to leave out.

If you're not sure what's missing, let the AI tell you. Add this line to the end of any prompt: 'Before you write anything, ask me up to five questions you'd need answered to make this specific to our team.' Answer them, then let it write.

One rule from earlier still applies: keep confidential data, client details and anything about a specific person out of public AI tools unless your company has approved one for that.

4. Make the AI check its own work before you do

Chatbots are surprisingly good at catching their own habits when you ask them to look. Paste this after your request, or add it to your saved instructions:

Before you give me the final version, check your draft and fix it:

- Is the main point or the ask in the first two sentences? If not, move it there.
- Did every number, name, date and quote come from what I gave you? Mark anything you added, guessed or couldn't confirm with [CHECK].
- Did you use any word from my ban list, an em dash, or an offer to help further? Take them out.
- Delete every sentence that would still be true for a different company.
- Cut the length by a fifth without losing a fact.
- Then show me the final version, followed by a two-line list of what you changed and what's marked [CHECK].

It has limits. A model checking itself will catch the style problems and flag its own guesses, but it can't spot a wrong fact it believes is right. The [CHECK] tags tell you where to look first; the last read is still yours. Before you hit send, ask yourself four things. Can you explain every claim in it? Did you verify the numbers and names? Is the main point in the first two lines? Would it be shorter if you'd written it yourself? If the answer to that last one is yes, cut it down.

5. Know what good work looks like

Workslop examples are everywhere. Examples of the opposite are harder to find, so here are three, each next to the slop version of the same thing. All six could come from the same chatbot. The good ones got better input.

Put any of the good versions next to its slop twin and the differences are plain: the answer comes first, the numbers are real, there's one clear ask, and it would make no sense if you swapped in another company's name. That last test is the fastest one you have, for your own work and for whatever lands in your inbox.

AI is very good at drafts. Your coworkers are counting on you for the part after the draft. That part is also the job security, by the way; we went through the research in will AI take my job.

Questions people actually ask

Start here

1. Next time it happens: pick one line from the table above and use it on the actual document, not the person.
2. If it keeps happening: write down the hours you spend fixing it. Numbers make the manager conversation easy.
3. If you use AI yourself: save the voice brief and the self-check in your chatbot's instructions today. It takes ten minutes, once.
4. For the lines you're too polite to say: HR Approved Ways to Tell Co-Workers They're Stupid translates every one of them into something you can put in writing. No chatbot required.

Sources

• BetterUp, What is AI workslop? Research on costs and solutions (BetterUp Labs and Stanford Social Media Lab survey, September 2025)
• Harvard Business Review, AI-Generated "Workslop" Is Destroying Productivity (Niederhoffer, Kellerman, Lee, Liebscher, Rapuano and Hancock, September 22, 2025)
• CNN, AI isn't taking over your job, but "workslop" is (September 26, 2025)
• Founder Reports, AI in the Workplace Statistics for 2026 (survey of 2,078 US workers, April 2026)
• Wikipedia, Signs of AI writing (WikiProject AI Cleanup advice page)