Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →shirhussain/humanize-writing
Use when writing or editing prose an audience will read — emails, blog posts, articles, marketing and website copy, LinkedIn or social posts, cover letters, newsletters, product descriptions, reports, essays — and when the user asks to humanize text, make it sound natural or less robotic, remove AI patterns, or clean up an AI-generated draft. Applies by default to any audience-facing writing task.
Passed 22 of 25 test cases with the skill, and 16 of 25 without it.
and 3 more in the full report
6 fixed16 passed either way3 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
Humanize this short announcement for our engineering blog: 'We have updated our SSL certificates across all load balancers.' Deliver only the polished final text without any introductory commentary like 'Here is your text' or list of changes made.
| Checked | Without skill | With skill |
|---|---|---|
| The response contains no introductory conversational preamble before the announcement. | Missed | Passed |
| The response contains no postscript or explanation of changes. | Passed | Passed |
The instructions your model reads when the skill loads.
Removes the statistical fingerprints of LLM-generated text, as catalogued by Wikipedia's "Signs of AI writing" (WikiProject AI Cleanup) and measured by the corpus studies that page cites (Kobak et al. 2025, Science Advances; Juzek & Ward 2025; Reinhart et al. 2025, PNAS).
Two principles drive everything below.
Density, not existence. No single pattern proves anything; humans use all of them. What exposes AI text is several patterns clustered in one passage. Write so the cluster never forms.
Specific beats generic. Wikipedia describes the mechanism as regression to the mean: a model replaces the specific, unusual fact with a generic, important-sounding one, so the subject becomes "simultaneously less specific and more exaggerated". Humanizing reverses that. Put the specific back, take the exaggeration out.
Generation mode (default). You are writing new prose. Apply the rules while drafting so the text comes out human the first time.
Rewrite mode. The user pastes existing text and asks you to clean it. Strip AI patterns while preserving everything else:
Enter rewrite mode when the user provides text to fix; otherwise stay in generation mode.
This rule applies in both modes and outranks every style rule. Never add a fact, name, number, date, quote, statistic, or citation that did not come from the source text or the user. "Concrete" means concrete with what you actually have. If a claim needs a specific you don't have, ask the user for it or write the sentence without it. A made-up number is worse than a vague sentence, because it is wrong. Fiction is the one exception.
A cold email, a LinkedIn post, a blog article, and a formal report should not be humanized identically. Before applying any rule, infer the register from the user's request and context:
| Register | Contractions | Fragments | Em dashes | Example contexts |
|---|---|---|---|---|
| Social | yes | yes | rare | LinkedIn, X/Twitter, Slack, chat |
| yes | sparingly | rare | cold email, follow-ups, newsletters | |
| Editorial | yes | rarely | max 1/300 words | blog posts, articles, essays |
| Formal | no | no | avoid | reports, proposals, formal docs, academic |
All other rules apply in every register. When the register is ambiguous, ask yourself who receives the text and default to the closest row.
test cases passed, vs 16 / 25 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.