humanizer

intent

humanizer scans text for 24 common AI writing patterns (significance inflation, promotional language, copula avoidance, chatbot artifacts, etc.) and rewrites them into natural, voiced prose. use this when you've written something that sounds statistically generic, needs personality injected, or reads like it came from a temperature-0 LLM. the skill preserves meaning while replacing sterile constructions with specific details, varied rhythm, and actual opinions.

inputs

text to humanize (required)

• raw text you want to review and rewrite
• can be pasted directly or piped in
• works on any length (articles, emails, code comments, product copy)
• context: whether this is technical writing, marketing copy, documentation, or editorial will inform tone choices

optional context flags

• --preserve-neutral: keep editorial/news voice (tone down first-person opinions)
• --technical-first: bias toward clarity and precision over personality (useful for API docs, specs)
• --aggressive-personality: err toward strong opinions, humor, edge (useful for opinion pieces, social posts)

no external connections required

• this skill runs locally on the text you provide
• no API keys, auth, or rate limits

procedure

1. read the input text carefully and mark each instance of the 24 patterns (listed in reference section below) with line numbers or highlights. note: many pieces will have 5-15 instances; that's normal.

2. identify pattern clusters - flag which patterns co-occur (e.g., "significance inflation + promotional language" in the same paragraph often signal a single tone problem, not 10 separate issues).

3. rewrite problematic sections, replacing each pattern:

   • significance inflation ("pivotal moment", "testament") → direct statements of fact
   • copula avoidance ("serves as", "boasts") → is/are/has
   • promotional language ("vibrant", "stunning") → concrete attributes or delete
   • chatbot artifacts ("I hope this helps", "Of course!") → delete or replace with substance
   • filler phrases ("in order to", "due to the fact that") → compact equivalents
   • excessive hedging ("could potentially possibly") → single modal (may, might, likely)
   • (continue for all 24 patterns in order)

4. preserve core meaning and facts - do not lose data, dates, names, or claims during rewrites. only the phrasing changes.

5. vary sentence rhythm - after rewrite pass, audit rhythm: check that consecutive sentences don't start the same way, that some are short (3-5 words) and some longer (15-20+ words), that questions or fragments appear occasionally if natural.

6. add voice and specificity - inject opinions where appropriate, use first-person ("i noticed", "this bothers me") if the context allows, include details instead of abstractions (e.g., "the river supports three endemic fish species" instead of "plays a crucial ecological role").

7. run the ai audit - read the draft aloud or have someone else read it. ask: "does this still sound like an LLM wrote it?" list remaining tells (stiff rhythm, generic conclusions, missing opinions, etc.).

8. revise based on audit findings - make another pass to smooth rhythm, add missing specificity, or deepen voice where audit flagged flatness.

9. output the final version with a summary of changes applied.

decision points

if the text contains heavy promotional language (boasts, nestled, groundbreaking, stunning, vibrant) AND generic claims:

• rewrite to concrete facts and remove adjectives, OR
• if this is marketing copy meant to sell, dial back personality and focus on removing vagueness (replace "stunning views" with "overlooks the valley" or "250-degree vista")

if the text is heavy on hedging, disclaimers, and knowledge-cutoff language:

• assume the author is uncertain; either cite specific sources to replace vague attributions, OR acknowledge directly ("i don't know", "the docs don't say", "this isn't well documented")

if the text has strong first-person voice already (opinions, "i think", asides):

• make minimal changes; focus only on removing patterns that contradict authentic voice (remove "of course!" or "i hope this helps", keep "this worries me")

if the text is technical (API docs, code comments, specs):

• use --technical-first flag; prioritize clarity and precision over personality; remove promotional language but keep neutral tone; avoid injecting casual opinion

if the text is empty or very short (< 20 words):

• flag as insufficient for pattern analysis; ask for more context or a longer sample

if the text has no detectable AI patterns (already natural, voiced, specific):

• report that as the outcome: "this text is clean; no rewrites needed"

output contract

the skill outputs three sections:

draft rewrite

• first pass at humanized text with patterns removed
• plaintext, no markup
• length similar to input (slightly shorter is normal when filler is stripped)

ai audit

• bulleted list of remaining AI tells (if any)
• examples: "rhythm still too tidy", "conclusion is generic", "missing specificity on dates"
• if text is clean, state: "no significant AI patterns remain"

final version

• revised text after audit pass
• incorporates fixes from audit phase
• this is what the user should use

changes made

• bulleted summary of transformations applied
• e.g., "removed 12 instances of copula avoidance", "replaced 4 vague attributions with specific sources", "added 3 first-person asides", "cut 8 filler phrases"

all output is plaintext unless the user requests markdown or HTML.

outcome signal

you know the skill worked when:

• text sounds natural when read aloud - no stumbling on phrases, no awkward rhythm, no "i hope this helps" hanging at the end
• specific details replace vague claims - before: "plays a vital role in the ecosystem"; after: "supports three endemic fish species"
• opinions or uncertainty appear where they belong - "i genuinely don't know how to feel about this" instead of neutral reporting
• sentence rhythm varies - not every sentence is 15 words; some are 3-5, others are 20+
• chatbot artifacts are gone - no "of course!", "let me know", "would you like", "i hope this helps"
• promotional language is gone - no "vibrant", "stunning", "nestled", "breathtaking", "groundbreaking"
• the audit pass found < 2 remaining tells - ideally zero; 1-2 minor issues (e.g., "one metaphor still feels abstract") is acceptable for first revision
• you'd be willing to publish or send this version - it has a voice; it feels like a real person wrote it

reference: the 24 patterns

content patterns

1. significance inflation - "stands as", "testament", "pivotal moment", "evolving landscape", "marks", "reflects broader", "key turning point"
2. notability name-dropping - "cited in NYT/BBC/FT", "independent coverage", "leading expert", vague press mentions
3. superficial -ing analyses - "highlighting", "underscoring", "emphasizing", "reflecting", "symbolizing", "cultivating", "showcasing"
4. promotional language - "boasts", "vibrant", "rich" (figurative), "profound", "groundbreaking", "renowned", "breathtaking", "must-visit", "stunning", "nestled", "in the heart of"
5. vague attributions - "experts argue", "observers cite", "industry reports", "some sources", "critics argue"
6. formulaic challenges sections - "Despite challenges...", "faces several obstacles", "Challenges and Legacy", "Future Outlook"

language patterns

7. AI vocabulary - "additionally", "align with", "crucial", "delve", "emphasizing", "enduring", "enhance", "fostering", "garner", "highlight" (verb), "interplay", "intricate", "key" (adjective), "landscape" (abstract), "pivotal", "showcase", "tapestry" (abstract), "testament", "underscore" (verb), "valuable", "vibrant"
8. copula avoidance - "serves as", "stands as", "marks", "represents", "boasts", "features", "offers" (when "is" or "has" would work)
9. negative parallelisms - "not only...but also", "it's not just...it's", "not merely...but"
10. rule of three overuse - lists of three identical grammatical units where 2 or 1 would be clearer
11. synonym cycling - repeating the same idea with different words ("protagonist faces challenges, the main character must overcome obstacles")
12. false ranges - "from X to Y" where X and Y aren't meaningfully on a scale ("from the singularity to dark matter")

style patterns

13. em dash overuse - multiple dashes per paragraph; should use commas, parens, colons, or periods instead
14. boldface overuse - highlighting words that don't need emphasis
15. inline-header lists - "Performance: [description]" structure where a flowing sentence works better
16. title case headings - capitalize every word in headers (should be lowercase except proper nouns)
17. emojis in professional writing - 🚀, 💡, etc. in formal or semi-formal contexts
18. curly quotation marks - smart quotes (") instead of straight quotes (")

communication patterns

19. chatbot artifacts - "I hope this helps", "Of course!", "Certainly!", "You're absolutely right!", "Would you like...", "let me know", "here is a..."
20. knowledge-cutoff disclaimers - "as of [date]", "up to my last training", "while specific details are limited", "based on available information"
21. sycophantic tone - "Great question!", "That's an excellent point!", "You're so right"

filler and hedging

22. filler phrases - "in order to" (use "to"), "due to the fact that" (use "because"), "at this point in time" (use "now"), "it is important to note that" (delete), "has the ability to" (use "can")
23. excessive hedging - "could potentially possibly", "might arguably suggest", "seems to indicate" (use single modal: may, might, likely)
24. generic positive conclusions - "the future looks bright", "exciting times ahead", "continuing their journey toward excellence"

---

credits: original skill by zed-industries. patterns and methodology based on wikipedia:signs of ai writing, maintained by wikiproject ai cleanup.