How it compares with other humanizer skills
The four most-used alternatives are rewrite tools that you run over a finished draft. humanizing-writing applies while Claude is writing, loads about 1,150 words per use, and says where its rules stop. Its test is also weaker than the one blader/humanizer reports. The table is from a survey in October 2026.
| Skill | When it runs | Built on | Genre and voice | Testing |
|---|---|---|---|---|
| humanizing-writing | While Claude writes, and on request over a draft | Eleven rules on claims, sentence shape, and endings. About 1,150 words loaded per use | Ten named cases, from API reference to fiction to your own voice | 18 blind pairs over two rounds, on Haiku, Sonnet, and Opus. Claude judges. Losses published |
| blader/humanizer | After the draft exists | One explanation of why model text sounds as it does, then 26 patterns ordered by strength. About 4,200 words per use | One paragraph on genre. Voice matching from a sample | Reports a blind preference test, 16 of 16 |
| conorbronsdon/avoid-ai-writing | After the draft exists. Rewrite, detect, and edit modes | A tiered word list makes up most of it | Protects code, quotes, and tables. English only by its own statement | Not covered in our survey |
| hardikpandya/stop-slop | After the draft exists | Names rhetorical moves ("cut quotables", "false agency") instead of phrases | No genre exceptions. Bans em dashes outright | A self-scored rubric, 35 of 50 to pass, with no source for the number |
| petergyang/no-ai-slop | After the draft exists | Minimum-change editing | Keeps a writer's "I think", fragments, and digressions. Caps em dashes at "1-2 max" with no source | Not covered in our survey |
When another one is the better choice
- blader/humanizer: A larger blind test, about 55,000 stars of use behind it, and install paths for more agents.
- conorbronsdon/avoid-ai-writing: It reports which checks ran, what was left alone, and why editing stopped.
- hardikpandya/stop-slop: Naming the move, not the phrase. We took the closer and the colon reveal from it.
- petergyang/no-ai-slop: The lightest touch on a person's own draft. We took "keep a real hedge" from it.
We borrowed from all four. From blader came the single explanation of why model text sounds as it does, and its five strongest patterns, none of which version 1.x covered. The closer and the colon reveal as named patterns came from stop-slop and no-ai-slop. From avoid-ai-writing came the habit of reporting the losses in a test, and not only the wins.
When this one is
Pick humanizing-writing if you’d rather the first draft came out right than run a second pass over it, if your writing crosses genres (reference docs one hour, a release post the next), or if you want Claude to draft in your own voice and leave your habits in. And if you write mostly in one genre and want the best-tested rewrite tool, use blader’s.
What we chose not to copy
Rewrite, detect, and edit modes, because a mode switch adds length that loads on every use. Numeric caps on punctuation and self-scored rubrics, because none of the four gives a source for its numbers. And “false agency” as a rule, because inanimate subjects are normal in technical writing: the function returns, the index stores.