---
name: humanizer-lite
description: |
  Remove signs of AI-generated writing from text. Use when editing or reviewing
  text to make it sound more natural and human-written. Based on Humanizer skill.
license: MIT
metadata:
  version: "3.0.0"
---

# Humanizer Lite

Rewrite text to remove AI tells. Three rules govern everything below:

1. **Keep the information, not the shape.** Every claim survives, but depth need not be uniform. Compress dull stretches, dwell where a human would, merge or split paragraphs freely.
2. **Never invent facts.** No name, number, date, quote, or citation that isn't in the source or from the user. When a vague sentence can't be made specific honestly, cut it or write the plain version. A fabrication is a defect even when it reads better. (Fiction excepted.)
3. **Match the voice.** If the user supplies a sample of their own writing, read it first and copy its habits: sentence length, vocabulary, paragraph openings, punctuation, transitions. Don't upgrade casual words or regularize deliberate quirks. A sample outranks every style rule here, including the em dash ban.

Return the rewrite. No draft, no audit bullets, no summary unless asked.

## Voice

Sterile is as obvious as slop. In blog posts, essays, opinion, and personal writing, let the writer have opinions, uncertainty, mixed feelings, humor, asides, and uneven rhythm. In encyclopedic, technical, legal, or reference text, plain and neutral *is* the human voice; don't inject first person or attitude there.

## Patterns

**Puffery.** Inflated significance, notability, and ad copy. *stands/serves as, testament, pivotal, key role, underscores its importance, reflects broader, marking a shift, evolving landscape, indelible mark, boasts a, vibrant, rich (figurative), nestled, in the heart of, renowned, breathtaking, stunning, must-visit, commitment to.* Also the generic upbeat send-off ("exciting times lie ahead," "a step in the right direction"). Cut it and end on the last concrete fact. And notability padding: long lists of outlets, follower counts. Keep one citation with real context; drop the rest.

> Nestled within the breathtaking region of Gonder, Alamata Raya Kobo stands as a vibrant town with a rich cultural heritage.
> → Alamata Raya Kobo is a town in the Gonder region of Ethiopia.

**Fake depth.** Participle tails that add nothing (*highlighting, underscoring, reflecting, symbolizing, ensuring, fostering, showcasing*), aphorism formulas (*X is the language of Y, X becomes a trap, the currency of, the architecture of*), and authority tropes that promise a deeper truth before restating an ordinary point (*the real question is, at its core, what really matters, fundamentally, the deeper issue*). Replace each with the concrete claim it gestures at.

**AI vocabulary.** *actually, additionally, align with, crucial, delve, emphasize, enduring, enhance, foster, garner, highlight (v.), interplay, intricate, key (adj.), landscape (abstract), pivotal, showcase, tapestry, testament, underscore (v.), valuable, vibrant.* They cluster; one alone means little.

**Copula avoidance and hidden actors.** *serves as / stands as / represents / boasts / features* where *is* or *has* would do. Same for subjectless fragments and passives that bury who acts: "No configuration file needed" → "You don't need a configuration file."

**Rhetorical formulas.** Negative parallelism (*not only... but*, *it's not just X, it's Y*), forced groups of three, false ranges (*from X to Y* where X and Y aren't on one scale), and runs of short declaratives manufacturing drama. One clipped sentence for emphasis is fine; four in a row is engineered. Also tailing negations tacked on as fragments:

> The options come from the selected item, no guessing.
> → The options come from the selected item without forcing the user to guess.

**Synonym cycling.** *The protagonist... the main character... the central figure... the hero.* Repetition-penalty artifact. Use one noun.

**Vague attribution and gap-filling.** *Experts argue, observers have cited, industry reports, several sources.* Name the source or cut the claim; never invent one. Related: when a model can't find information it writes *about* not finding it, then invents filler: *while specific details are limited, maintains a low profile, keeps personal details private, likely grew up, it is believed that.* Also stale disclaimers (*as of my last training update*).

> Information about her early life is not publicly available, suggesting she maintains a low profile. She likely grew up in a middle-class household.
> → Her early life is not documented in the available sources. (Or cut the section.)

**Filler and hedging.** *In order to* → *to*. *Due to the fact that* → *because*. *At this point in time* → *now*. *Has the ability to* → *can*. *It is important to note that.* Stacked qualifiers: *could potentially possibly be argued that it might.*

**Uniform hyphenation.** *data-driven, high-quality, cross-functional, real-time, long-term, end-to-end.* AI hyphenates these everywhere. Humans hyphenate attributively and usually drop it in predicate position: keep "a high-quality report," write "the report is high quality."

**Formatting tells.** Cut em and en dashes entirely, replacing them with a period, comma, colon, parentheses, or a restructure, and scan the final text for `—`, `–`, and ` -- ` before returning it. Also: mechanical boldface, bulleted lists where each item is a bold header plus a colon (usually better as prose), Title Case In Headings, emoji decoration, curly quotes.

**Chat artifacts.** *I hope this helps, Certainly!, You're absolutely right, let me know, would you like me to.* Sycophantic openers (*Great question!*). Signposting that announces instead of doing (*let's dive in, here's what you need to know, let's break this down*). Fake-candid hooks used as theatrical pauses (*Honestly? Look, Here's the thing*). A person being honest just says the thing.

**Structural tells.** Formulaic "Challenges and Future Prospects" sections. A heading followed by a one-line paragraph restating the heading. Documentation written as a diff ("this replaces the previous approach of...") when it should describe the thing as it is, unless the document is genuinely version-scoped.

## What not to flag

Clean human writing hits these too. Not tells on their own: polished grammar; mixed casual and formal register; dry prose without specific tells; formal vocabulary generally; salutations and sign-offs; a lone *however* or *moreover*; curly quotes (every editor auto-curls); a single em dash; one short emphatic sentence; *honestly* or *look* used mid-sentence; unsourced claims; clean formatting. Never rewrite a watched phrase inside a quotation, title, proper name, or an example discussing the phrase.

Look for **clusters**. One em dash is nothing; em dashes plus rule-of-three plus *vibrant tapestry* plus a "Conclusion" section is a confession.

## Signs of a human, leave these alone

Specific, hard-to-fabricate detail (a real address, a weird quote). Mixed feelings and unresolved tension. Dated slang or in-jokes tied to a year and a subculture. Real variety in sentence length. Genuine asides and self-corrections. Anything written before November 2022.