The number worth knowing

New 2026 research on "workslop" — AI-generated content that looks finished but is generic, wrong, or missing what the project actually needed — puts real figures on something most desk workers have already felt. Around 40% of workers say they've received it in the past month. Fixing each instance takes close to two hours on average. More than half of the people who received it said it left them annoyed; over a fifth said it left them feeling offended. The uncomfortable twist: nearly one in five people who use AI at work admit they've sent exactly that kind of output themselves.

Why it's not the same as a typo

Workslop doesn't look sloppy. It's properly formatted, confidently written, and complete-looking — which is exactly why it slips past a quick skim and lands as extra work on whoever has to actually use it. They read it, realise it's missing the one number they needed or doesn't reflect what was actually decided, and now have to come back and ask you for the real version. The time you saved writing it gets spent twice over, just by someone else.

Where it usually comes from

It's rarely laziness. It's a quick, under-specified prompt — "write a summary of this," "draft an update for the team" — with none of the detail that makes a document actually usable: the real figures, the decision that was made, the name of the client it's about. Read on its own, the output sounds fine. It just wasn't written for this project, and that shows the moment someone who knows the details tries to act on it.

The three-question check before you hit send

Before sending anything AI drafted for you, run it past three questions. Does it contain a specific fact, name or number from this actual project, or could it apply to almost anyone's situation? Would the person reading it need to come back and ask you a follow-up question just to use it? And, honestly, did you read the whole thing properly, or only the first two lines?

Read this draft before I send it to [recipient]. Flag anything that: - Could apply to any project, not just this one - States a fact, figure or claim I haven't actually verified - A reader would need to come back and ask me to clarify Then tell me plainly: does this read like it was written specifically for this situation, or could it have been written for anyone?

If you manage people, not just yourself

If you're reviewing work from a team that uses AI, the same three questions work as a quick screen before anything goes external. It's worth saying out loud once, rather than assuming everyone already knows: a fast AI draft is a starting point for editing, not a finished deliverable, and the person who sends it is still responsible for what's in it.

💡 This week's action

Before your next AI-drafted email or report goes out, run it through the three-question check above. It costs you two minutes. The research says skipping it costs whoever receives it closer to two hours.

The actual fix

None of this means using AI less. It means putting the real, specific details into the prompt in the first place — the client's actual name, the real figure, the decision that was actually made — so the first draft doesn't need the two-hour rescue later. A slightly longer prompt now is cheaper than a confused colleague later.