You wrote a report, a long email, or a deck with AI's help. It reads fine, the deadline is close, and you're about to send it. But something feels slightly off, and you can't say what.
Here's what you get: six questions that sort good AI-assisted work from slop, a clear verdict based on how many times you answered no, and why this test beats any automated score.
Start with the basics
Slop, in the words of Nufar Gaspar, an AI consultant who wrote her own "De-Slop Playbook": work where the person who should have done the thinking skipped it, partly or entirely, and the cost of that moved to whoever receives it. A Stanford and BetterUp study found that 40% of US desk workers received this kind of work in the past month, and each incident cost the person who received it close to two hours to fix or untangle. It rarely happens out of bad intent, it happens because it's easier to send whatever came back fast.
Six questions, for anything AI helped you write
Answer honestly, without rereading what you wrote just to justify it.
- Could you defend every sentence, without rereading it to remember why it's there?If you need to check what you wrote to know if it's true, you haven't really gone through it.
- Did you read all of it, start to finish, before it goes out?Not just the first two paragraphs. A 40-page report that looks fine after page one is exactly where tiredness hands the reading over to whoever receives it.
- Is there anything in it only you, or your team, could have contributed?A point of view, a decision, real data. If anyone could sign their name to it, it won't be especially relevant or sharp.
- Is this the shortest, most effective form that still does the job?If it could be said in half the space, the rest is padding, not content.
- Does it save the reader time, or cost them time?This is the definition of slop, just turned into a question about the specific thing in front of you.
- Would you put your name on it, in capital letters, without flinching?The most intuitive question of all. Even if you can't say why not, if the answer is no, there's more work to do.
Wait, before you run this on someone else's work
This test is a mirror, not a weapon for other people's work. Questions 1 and 2, could you defend it, did you read it, you can't answer those for someone else, only they know. If you're reviewing someone else's work, lead with question 5, does it save or cost you time, and question 3, is there a point of view in it. A no on either is reason enough to send it back, without turning all six into an interrogation.
Next ten minutes: pick one thing you're about to send today, an email, a deck, a report, and run it through the six questions. Two or more noes, hold off, and decide exactly what's missing before you go back to the tool.