The AI-Document Slop Test
An AI-written document is fine — as long as you can stand behind every line and it took longer to make than to read.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 3
- Confidence
- 80%
A standard for using AI-generated writing in professional documents and email without producing slop. AI-generated work is welcome — often better than what most people would write by hand — but it must clear two tests: the author directed it well enough to stand behind every line, and it didn't take less time to produce than it takes the reader to read. The dividing line isn't human-vs-AI authorship; it's ownership. You must decide what it should say even if the exact sentences don't matter.
Origin
Dan Shipper's internal rule at Every for what makes AI-generated documents acceptable versus slop.
Core principles
- 01Slop is when it took the author less time to make than it takes you to read
- 02The author must stand behind every line and know what's in the document
- 03Direct the substance (what it should say) even when the exact wording doesn't matter
- 04Well-directed AI writing often beats what most people produce by hand, because most people write strategy documents badly
How to run it
- 1
Own the substance before generating
Decide what the document or email should actually say — the ideas, structure, and argument — rather than outsourcing the thinking to the model.
Pro tip For prose like email, accept that the exact sentences often don't matter much; invest your judgment in what it should say, not each word.
Watch out If you can't articulate what it should say, you're generating slop no matter how polished the output looks.
- 2
Direct the model well, then read and stand behind it
Have the model draft, then review until you can defend every line as if you wrote it. The test is whether you'd be caught out if questioned about its contents.
Pro tip Label AI-generated or AI-assisted work honestly — Shipper publishes a mix of human and AI writing and always labels it.
Watch out Sending an AI document you clearly know nothing about — being unable to discuss what's in it — is a 'big no-no' and not allowed.
- 3
Apply the time-asymmetry check
Confirm the piece didn't take you less time to produce than it takes the recipient to read; that asymmetry is the definition of slop.
Pro tip Use AI as a co-author on things where it genuinely improves the output, not as a shortcut to dump unreviewed text on someone.
In the wild
At end-of-2025 planning, everyone at Every talked to a Notion agent that asked about last year, goals, and metrics and pushed back, producing AI-generated strategy reports per team. Shipper then reviewed them to spot which were high vs low quality and which teams needed to talk.
→ He got 'incredibly good' strategy documents that were easier to process than most human-written ones, because most people write strategy documents badly.
Shipper asked Codex to send an investor email; it normally asks first but this time sent it without his review. He checked his sent folder and found it was 'exactly what I would have sent.'
→ It cleared the bar by accident — but the standard remains that he wants to be the one who decides what it should say.
Common mistakes
Sending AI work you can't stand behind
If it's clear in conversation that you have no idea what's in a document you sent, you've violated the ownership standard — the point of directing AI is to still own the substance.
Reflexively rejecting AI-generated writing
The aversion to AI writing is misplaced for internal documents and email, where well-directed AI output usually beats what most people would type by hand — refusing it forfeits real quality gains.
Is it for you?
Best for
Managers and teams setting norms for AI-assisted documents, email, and strategy work internally
Not ideal for
Contexts demanding original human authorship for its own sake, such as bylined creative writing you want to be fully human
From the transcript
“there is a difference between an AI generated document that's slop and not. And the slop one is it took them less time to make…”
“if we talk about it and it's clear you have no idea what's in it, like big no-no. Not allowed to do that.”
“the kind of strategy document that GPT-5.5 can write when it's directed well by someone on my team is way better than like them just…”
From the episode
The AI paradox: More automation, more humans, more work
Dan Shipper