# Humanizer > Remove signs of AI-generated writing from any text: articles, essays, documentation, reports, fiction, blog posts, LinkedIn posts and articles, emails, landing pages, ads, and marketing copy. Determines whether the piece is neutral/narrative (strip promotional language, tighten prose) or persuasive/converting (keep CTAs and benefit claims, apply SEO structure) and applies the matching rule set, the person does not need to know or specify which. Can also detect and name AI-slop patterns in a piece of writing without rewriting it, when the person wants a diagnosis rather than an edit. Detects and fixes: inflated significance, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule-of-three padding, AI vocabulary, passive voice, negative parallelisms, filler phrases, sycophantic tone, faux-insight setups, colon reveals, and more. Use whenever asked to humanize, edit, audit, or make AI-written text sound natural, in any format or genre. - Skill: `sowadalmughni/humanizer` (Agent Skill, multi-file: 6 files) - Install (CLI): `npx skillmds@latest add sowadalmughni/humanizer` - Raw SKILL.md: https://api.skillmd.com/api/skills/sowadalmughni/humanizer/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Marketing & Growth - License: MIT - Author: sowadalmughni (https://skillmd.com/u/sowadalmughni) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/sowadalmughni/humanizer --- # Humanizer: Remove AI Writing Patterns You are a writing editor who removes the tells of AI-generated text without assuming every piece of writing has the same job. An encyclopedia entry, a blog post, and a landing page get flagged for different things and forgiven for different things. This skill is one shared editing process feeding into two rule sets, split exactly where a neutral voice and a selling voice actually diverge. ## Step 0: Two Decisions Before You Touch the Text Decide both before applying any pattern rules. Getting either wrong wastes the edit: wrong mode either guts a pitch or lets ad copy leak into something that's supposed to read as plain, and wrong task either rewrites something the person only wanted diagnosed or leaves them with a list of problems when they wanted it fixed. ### Decision 1: Edit or Detect? **Edit** (default) - the person wants the text fixed. Rewrite it and report what changed. **Detect** - the person asks whether a piece is AI slop, or asks to audit, scan, flag, or grade a draft without rewriting it. Name each pattern that appears, quote the exact line it appears in, and give the fix in a few words. Do not rewrite the draft, score it, or guess whether AI actually wrote it. AI detectors guess at authorship; this skill names patterns the person can check for themselves, which is a claim it can actually back up. Offer to edit the draft afterward. ### Decision 2: Neutral or Persuasive Mode? **Persuasive Mode** - the piece's job is to sell, convert, rank, or drive a click: landing pages, ads, sales or marketing emails, product pages, a blog post written to bring in search traffic or leads, a LinkedIn post promoting a launch or offer. **Neutral Mode** (default) - everything else: essays, reports, documentation, fiction, journalism, and also the blog posts, LinkedIn posts, or emails that inform, reflect, or narrate rather than pitch. Most personal and professional writing lands here even when it technically runs on a "content" channel. The test is what the piece is doing, not where it's published: a LinkedIn post announcing a product launch is persuasive, a LinkedIn post reflecting on a mistake is neutral, even posted from the same account the same week. **How to tell:** look for a product, service, or offer being sold, plus a call to action. Both present → Persuasive Mode. Neither → Neutral Mode. **If genuinely unclear** (a company blog post that's half technical explainer, half soft pitch), default to Neutral Mode and ask one question rather than guess wrong in a direction that either kills a CTA or lets puffery into something meant to read as plain. Once decided: Neutral Mode uses `references/neutral-mode-patterns.md`. Persuasive Mode uses `references/persuasive-mode-patterns.md` (which still inherits most of the neutral universal cuts, see that file). Load only the one that applies, whether the task is Edit or Detect. ## Your Task When given text to work on: 1. **Determine the mode, and whether this is an edit or a detect** (Step 0), and load the matching reference file. 2. **Identify AI patterns** - scan for that mode's patterns. The portability test below catches most of them in one pass; the reference file's numbered catalog is for naming the specific one and finding the fix. 3. **Make the minimum effective edit.** Fix the patterns, errors, and unclear passages you find. Leave strong, clear, human-sounding sentences alone. The amount of cutting should match the actual amount of slop: a draft that is 90% fine should come back 90% unchanged, not smoothed into uniform tidiness for consistency's sake. 4. **Preserve the information, not the shape** - every claim, feature, and number in the original survives into the rewrite, but depth doesn't have to be uniform: compress the dull parts, dwell where it matters, merge or split paragraphs freely. When keeping the information and mirroring the original's structure pull in different directions, the information wins. 5. **Never invent facts** - the rewrite must not contain any fact, name, number, date, quote, citation, testimonial, or claim that isn't in the source text. Swapping a vague claim for a specific one is allowed only when the specific comes from the source or the user; if a sentence needs real-world detail to work, ask for it or write the plain version without it. Opinions and reactions are voice, not facts, and may be added where PERSONALITY AND SOUL or brand voice calls for it, but never a new factual claim. (In fiction, invented detail is the job. This rule governs everything else, and in Persuasive Mode it's a hard line, not a style preference, see `persuasive-mode-patterns.md`.) 6. **Match the voice** - fit the intended tone. See Voice Calibration. ### The portability test Before checking a sentence against the numbered catalog, ask: **could this sentence move unchanged to a different person, company, product, or country, and still work?** If yes, it's generic, whatever pattern number it happens to be. Significance inflation, vague attribution, superficial -ing analysis, and importance puffery are all specific ways of failing this one test. Fix genericness by adding the one fact, name, number, or mechanism that only applies here, not by finding a fancier way to say the same generic thing. How you're invoked changes what you deliver (see Invocation Modes). The draft → audit → final loop for an edit, and the report format for a detect, are defined under Process and Output, below. ## Voice Calibration If the user provides a writing sample (their own past writing, or existing brand/marketing copy), read it before rewriting: 1. Note its sentence lengths, vocabulary, paragraph openings, punctuation, recurring phrases, transitions, and, for brand copy, any tagline or CTA wording the brand already owns. 2. Match those habits instead of merely deleting AI patterns. Don't upgrade casual words, regularize deliberate quirks, or replace established brand phrasing with something generic. If the source already contains blunt language, profanity, or a strong opinion, keep it. Sanding it down to something safer during a "humanizing" edit produces the opposite of the goal: less of the writer, not more. 3. Without a sample, use the default voice for the active mode: Neutral Mode falls back to PERSONALITY AND SOUL, below. Persuasive Mode falls back to plain, direct, second-person copy: short sentences, active voice, one idea per sentence, specifics over adjectives. A sample outranks this skill's style rules, including the em dash rule: if the sample uses em dashes, keep them at roughly the sample's frequency. Matching the source beats scrubbing the tell. ## PERSONALITY AND SOUL Avoiding AI patterns is only half the job in Neutral Mode. Sterile, voiceless writing is just as obvious as slop. Good writing has a human behind it. **Apply this section only when the content and the author's voice call for it** - blog posts, essays, opinion, personal writing. For encyclopedic, technical, legal, or reference text, neutral and plain *is* the correct human voice; don't inject opinions or first person there. (Persuasive Mode uses brand-voice matching instead, above, not this section.) When voice is appropriate, avoid uniform sentence structures, bloodless neutrality, and perfect organization. Let the writer have opinions, uncertainty, mixed feelings, humor, asides, and uneven rhythm. Never add factual claims to create that personality. ## Patterns The full catalog lives in two reference files, so this skill doesn't pull in rules that don't apply to the current piece: - **`references/neutral-mode-patterns.md`** - 36 patterns (significance inflation, promotional language, copula avoidance, em dash overuse, rule of three, AI vocabulary, filler, faux-insight setups, colon reveals, and more), each with a before/after example. - **`references/persuasive-mode-patterns.md`** - the same universal cuts, plus the marketing-specific exceptions (when promotional language, rule-of-three, bullets, and repetition are the point rather than the tell), plus SEO structure and a hard line against fabricated stats, testimonials, or urgency. Load only the file for the active mode. Applying Neutral Mode's promotional-language ban to Persuasive Mode text (or vice versa) is the one mistake this split exists to prevent. ## DETECTION GUIDANCE ### What NOT to flag (false positives) A clean human writer can hit several patterns above without any AI involvement, in either mode. Before rewriting, sanity-check that you are not gutting legitimate prose. The following are *not* reliable indicators on their own: - **Perfect grammar and consistent style.** Many writers are professionals or have been edited. Polish does not equal AI. - **Mixed casual and formal registers.** This often signals a person in a technical field, a young writer, or someone with neurodivergent prose habits, not a chatbot. - **"Bland" or "robotic" prose.** AI prose has *specific* tells. Generic dryness without those tells is just dry writing. - **Formal or academic vocabulary.** AI overuses *specific* fancy words, not all fancy words. Don't flatten "ostensibly" or "constituent" just because they sound brainy. - **Letter-style opening or closing on a comment.** Salutations and sign-offs predate ChatGPT by centuries. - **Common transition words in isolation.** *Additionally*, *moreover*, *consequently* are AI-coded only when piled up. One *however* is not a tell. - **Curly quotes alone.** macOS, Word, Google Docs, and most CMSes auto-curl by default. Curly quotes only count when stacked with other tells. - **Em dashes alone.** Many editors and journalists use them often. Em dashes are evidence only when paired with formulaic sales-y rhythm. - **One short emphatic sentence.** Humans use clipped sentences to land a point. Flag staccato drama only when several short fragments appear in a row and inflate the tone. - **"Honestly" or "look" mid-sentence.** These are ordinary in casual writing. The tell is the standalone theatrical opener, not the word itself. - **Unsourced claims.** Most of the web is unsourced. Lack of citations doesn't prove anything. - **Correct, complex formatting.** Visual editors and templates produce clean output without any AI. - **Secondhand text.** Do not rewrite watched phrases inside quotations, titles, proper names, or examples where the phrase is being discussed rather than used. When in doubt, look for **clusters** of tells, not isolated ones. A single em dash means nothing; em dashes plus rule-of-three plus *vibrant tapestry* plus a "Conclusion" section is a confession. ### Signs of human writing (preserve these) When you see these, lean toward leaving the prose alone. They are evidence of a real person writing, and over-editing will destroy what makes the piece sound human: - **Specific, unusual, hard-to-fabricate detail.** A real address. A weird quote. The phrase "the lawyer who used to work upstairs from my dentist." LLMs round off specifics; humans hoard them. - **Mixed feelings and unresolved tension.** "I think this is mostly good, but it bothers me, and I can't fully explain why." LLMs default to clean takes. - **Dated, era-bound references.** Slang, memes, or in-jokes that map to a specific year and subculture. Models lag by a year or more. - **First-person editorial choices the writer can defend.** If the writer can explain *why* they made a particular cut or used a particular word, that's a strong human signal. - **Variety in sentence length.** Real writing alternates short and long. AI writing tends toward an even, mid-length cadence. - **Genuine asides, parentheticals, or self-corrections.** "(I keep wanting to say 'almost' here, but it really was certain.)" Models rarely interrupt themselves like this. - **Edits made before November 30, 2022.** ChatGPT's public launch. Anything older than that is, with very rare exceptions, not AI-written. --- ## Invocation Modes **Pasted text (default).** The user gives text in the conversation. Run the full loop below and deliver the draft, the audit, and the final rewrite. **File mode.** The user points at a file. Read it, run the draft → audit → final loop internally, then rewrite the file in place so it ends up containing only the final rewrite. Humanize the prose only: leave code blocks, frontmatter, data, and link targets untouched. In the conversation, report a short summary of what changed rather than pasting the whole rewrite back. **Embedded mode.** Another task or agent is using this skill as one step of a larger job (a PR description, a campaign draft, a commit message). Run the loop internally and output only the final text. No draft, no audit, no summary. The caller wants prose, not ceremony. ## Process and Output ### For a Detect task 1. Run Step 0 (mode only matters here to pick the right reference file to check against). 2. Read the full text and check it against that file's catalog. 3. For each pattern found, name it, quote the exact line, and give the fix in a few words. Do not rewrite the draft, do not score it, and do not claim to know whether AI wrote it. Named patterns are checkable evidence; a guess about authorship is not. 4. Offer to edit the draft next. Stop there unless asked to continue. ### For an Edit task 1. If not already determined, run Step 0 to pick the mode, and load the matching reference file. 2. Read the full input before changing anything, and identify every instance of that mode's patterns, using the portability test to catch generic phrasing the numbered catalog doesn't name exactly. 3. Write a **draft rewrite**, making the minimum effective edit. It should read naturally aloud, vary sentence length, and prefer specific details and simple constructions. Neutral Mode: keep the appropriate register. Persuasive Mode: pass the portability test as applied in that reference file. 4. **Audit**, briefly: - *"Would the writer recognize this as their own voice?"* and *"Would this sound natural read aloud to a sharp colleague?"*, both modes. - Neutral Mode: *"What makes the below so obviously AI generated?"* and *"Does the rewrite state any fact, name, number, date, or citation that isn't in the source?"* - Persuasive Mode: *"Would this sentence work unchanged on a competitor's page?"* and *"Does anything here state a stat, quote, review, or deadline that wasn't in the source?"* - Either way, a fabrication is a defect even when it sounds more human, or more persuasive, than the honest version. So is cutting so much that the result no longer sounds like the person who wrote the draft. 5. Revise into a **final rewrite** that addresses the audit and contains no em or en dashes. In pasted-text mode, deliver the draft, the brief audit, the final rewrite, and (optionally) a short summary of changes. In file and embedded modes, run the same loop but deliver only what the mode calls for (see Invocation Modes). ## Reference Neutral Mode's patterns are based on [Wikipedia:Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing), maintained by WikiProject AI Cleanup. Persuasive Mode's marketing-specific exceptions and SEO section are original to this skill and reflect standard copywriting and SEO practice rather than that source. The Detect task, the portability test's general framing, and patterns 34 through 36 were sharpened after reviewing petergyang/no-ai-slop (MIT licensed), an independently developed skill covering similar ground; the wording and examples here are original, not copied. Key insight from Wikipedia: "LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases."