AI slop is machine-generated content that looks polished but wastes the reader’s time: writing that is fluent sentence by sentence and empty as a whole. The workplace version has its own name, workslop, coined by researchers at Stanford and BetterUp for AI work that masquerades as progress while pushing the real effort onto whoever receives it.
Slop is not the same thing as AI-assisted writing. A draft the author verified, tightened, and stands behind is just writing. Slop is what ships when nobody did that second half.
You can recognize it by feel, and you can also measure it. AI-heavy prose overuses a few hundred style words; researchers traced the most famous of them spiking across 15 million scientific abstracts right after ChatGPT launched. It keeps sentence lengths unnaturally even where people scatter short next to long. It leans on templates: the rule of three, the hollow opener, the tidy summary that repeats the headline. Each of these is countable, which is why a draft can carry a slop score from 0 to 100. Known human writing lands between 9 and 21, unedited AI drafts average 77, and 25 is the line where readers start to notice.
The costs are documented, not argued. Fixing one piece of workslop takes a receiving colleague nearly two hours, 42% of recipients trust the sender less afterward, and about 40% of the time AI saves gets spent correcting its output. LinkedIn added a button for reporting it, and over a million people have already used it.
The fix is not banning the tools. It is refusing to ship their first draft: verify the facts, read it aloud, cut what anyone could have written, and measure it before sending. That workflow exists as a free, open-source skill, and the receipts behind every number on this page are linked from our post The Slop Tax. See also the full anti-slop toolkit and the measured editor ranking.