so.
how to tell someone their work is ai slop without making it weird.
a coworker, someone on your team, or your boss sent you a thing nobody read. you'd like it fixed. you'd also like to keep having lunch with them.
ask for what's missing, not how it was made.
§1 · the bind
the person who sent it saved an hour. you're about to spend it.
when we read through what people say about ai slop at work, most of it came from the receiving end: the recap that got a decision wrong, the deck with a number nobody can find, the email that's four paragraphs around one question. the sender moved on. the reader is left holding it.
you fix it yourself, quietly, every time. it becomes your job. nobody gave you the job.
you're right, probably. it still lands as an accusation, and now the conversation is about you.
a normal, specific work request for the thing that's missing. it fixes the doc and it doesn't need anyone to confess.
§2 · the method
three moves
they work on slop. they also work on regular, human-made unclear work, which is the point. you never have to guess which one you're holding.
-
point at one thing.
"this feels off" gives them nothing to fix. one sentence, one number, one claim does.
instead ofthis whole doc feels kind of generic?
trythe second paragraph says we chose bluepeak. i think we chose northwind.
-
ask, don't accuse.
a question they can answer beats a verdict they have to defend. "where's that from?" is a normal thing to ask about any number, from anyone.
instead ofdid you just paste this from chatgpt?
trywhere's the $50,000 from? i couldn't find it in the budget sheet.
-
make the fix small.
ask for the one thing that turns it from a draft into something you can act on: the decision up top, the source of a number, or what they need from you and by when.
instead ofcan you redo this?
trycan you put the decision and the deadline in the first two lines?
§3 · the tool
pick who sent it and what's wrong. copy the message.
fill in the [brackets], read it once (you know why), and send. it all happens on this page. nothing you click goes anywhere.
your message, to a coworker
notice none of these say "ai." that's on purpose.
§4 · upward
when it's your boss
the hardest version, because "please fix this" doesn't flow uphill. so don't ask them to fix it. offer to help it land.
- make it about the decision. "which of these is the call you want me to act on?" helps them. it doesn't grade them.
- offer to do the check. "want me to pull the current number before this goes out?" turns a correction into a favor.
- always in private. a quiet message fixes the doc. a reply-all fixes nothing and costs you something.
- let them send the fix. you found it. they get to be the one who corrected it. that trade is worth it.
§5 · please don't
four things that backfire
- ask "did you use ai for this?" they can say no, and then you're stuck. it's also the wrong question. the problem is that nobody read it.
- run it through an ai detector. they guess. a wrong guess about someone's work is very hard to take back.1
- reply with your own ai's summary of it. a summary of a summary is a rumor with formatting. now there are two documents nobody read.
- fix it silently, forever. it's kind once. after that it's a second job, and the sender never finds out there was a problem.
§6 · the low-effort option
or just send them a link
some people take a page better than a person. these were written to be forwarded. neither one says "you."
send a summary, not a grenade.
five rules for meeting summaries, plus a free check they can paste into copilot, chatgpt or claude. it marks anything it can't find in the transcript.
for the meeting recap person → /slop/ai slop bingo.
paste an email or a recap and the card lights up the classic tells. it's a mirror, not a polygraph. good for the self-aware.
for everyone else →§7 · faq
questions people ask right before they don't say anything
is it rude to point this out?
it can be, if you lead with the verdict. asking for the decision, a source, or what they need from you is just work. people ask each other those things all day.
what if they wrote it themselves?
then nothing is lost. none of the scripts mention ai, so they're fair to send to anyone, about anything. unclear work was around long before the robots.
what if it keeps happening?
then it's a team habit, not one doc. suggest a shared rule for everything that goes out: decision in the first two lines, every number traceable to where it came from. a rule for everyone is easier to say than a note to one person.
what if it's me?
welcome. it's most of us, some of the time. read it once before you send it, and check the numbers against where they came from. the meeting summary check does the tedious part for recaps.
did an ai write these scripts?
an ai drafted them. a human read every one, out loud, picturing their own boss. several did not survive.
sharpwork makes ai toolkits for people in legal, project management, pr, accounting and sales. you bring your own ai. we bring the prompts and the checks, so what leaves your desk is work you can stand behind.