An AI slop editor takes a machine-sounding draft and rewrites it so it reads like a person wrote it, without changing the facts. Plenty of tools claim to do this. Numbers are rarer. So we replayed the same 18 drafts through five open-source editors and scored every rewrite on the public slop score, where lower is better, anything under 25 reads clear, and known human writing sits between 9 and 21.
One honest caveat before the ranking: the benchmark is ours. The method, the drafts, and every score are public in the Zero Slop repository, so you can rerun any row and check us.
Zero Slop
mean 15.4 · 18 of 18 clear
The only editor whose rewrites landed inside the known-human range. It runs as a skill in your own agent, keeps every name, number, quote, and link, and learns your taste from the edits you keep. Free and MIT licensed.
humanizer
mean 24.3 · 13 of 18 clear
The strongest prompt-based rewriter we tested. It cuts tell words well; rhythm and structure move less.
stop-slop
mean 24.7 · 13 of 18 clear
Close behind humanizer, with the same shape: good vocabulary hygiene, less repair of cadence.
no-ai-slop
mean 28.1 · 12 of 18 clear
A solid checklist of things to avoid. As a rewriter it leaves more of the AI accent in place.
de-slop
mean 52.3 · 6 of 18 clear
Better than nothing, and well short of the pack.
The gap has a simple cause. A prompt asks one model pass to fix everything at once. A pipeline splits the job: score, interpret, rewrite, check facts, copy edit, read aloud, verify. The drafts that started at a mean of 76.3 ended at 15.4 because each stage had one job and a gate behind it.
If you write with AI anywhere, run your next draft through one of these before you hit send. The reader can tell. Start with the one-command install, or see the full anti-slop toolkit and what AI slop actually is.