AI Smell Remover

AI text without human curation has predictable weaknesses. Find them, fix them, keep the author's intent.

When to Use This Skill

• User asks to humanize text or writing
• User says the text sounds like AI, robotic, or unnatural
• User asks to fix AI smells or AI tells
• User wants to rewrite AI-generated content to sound more like a person
• User asks to make writing less formal, less corporate, or more conversational
• User mentions the text has hedging, bullet overload, or corporate jargon
• User shares a piece of writing and asks if it sounds human

What to Look For

Verbosity — long intros and inflated phrases that delay the point. Before: "It is important to highlight that documentation plays a fundamental role in ensuring alignment and efficiency across teams." After: "Documentation keeps teams aligned and efficient."

Excessive hedging — hiding behind "can," "might," "potentially" instead of making a decision. Before: "This approach can potentially help improve system reliability." After: "This approach improves system reliability." Or if uncertainty is real: "This improves reliability, but increases operational cost."

Infinite lists — 10 bullets where 3 would do. Before: "Key benefits: scalability, flexibility, improved collaboration, enhanced visibility, lower costs, better governance, increased automation, faster onboarding and improved alignment." After: "The main benefits are scalability, faster onboarding, and lower operational cost."

Corporate tone — sounds like policy, not a person. Before: "With the goal of optimizing processes and ensuring excellence, this solution enables a robust and future-proof architecture." After: "This solution simplifies the process and scales without additional complexity."

Hidden repetition — same idea in the title, intro, and first paragraph. Before: Title "Improving Document Quality" / Intro "This document focuses on improving documentation quality." / Para "Documentation quality is essential for..." After: Title "How We Improve Document Quality" / Para "We improve document quality by cutting unnecessary text, stating decisions clearly, and prioritizing what matters."

Over-structuring — H2 > H3 > H4 for three sentences of content. Before: ## Overview / ### Components / #### Services / (2 sentences) After: ## Overview / (2 sentences that say what the components are)

Textbook word choices — technically correct words that sound formal, translated, or learned-from-a-textbook rather than natural. Common in non-native writing, and amplified by AI. Applies to English by default, but can be applied to any language if specified. Before: "Utilize the interface to accomplish the task prior to commencing the next step." After: "Use the interface to finish the task before starting the next one."

Common swaps: utilize -> use, commence -> start, prior to -> before, subsequently -> then, endeavor -> try, ensure that -> make sure, in order to -> to, regarding -> about, it is necessary to -> you need to, accomplish -> finish / get done.

The test: would a non-native English speaker understand and feel comfortable writing this word? If not, find a simpler, more internationally common word.

Decorative formatting — emojis and em-dashes used for effect, not meaning. AI reaches for emojis and — as substitute for good writing. Real writers use them sparingly, if at all. Before: "We're launching AI Cafe — a space to learn and share AI together." After: "We're launching AI Cafe, a space to learn and share AI at work."

The test: remove every emoji and em-dash. If the sentence still works, they were decoration. Cut them.

Abstract nouns — vague words that hide the real thing doing the work. AI often writes about "solutions," "approaches," or "capabilities" instead of naming the concrete mechanism. Before: "This solution improves performance." After: "Caching database queries improves performance."

Common vague nouns: solution, approach, capability, initiative, process, functionality, framework, strategy, system.

The test: replace the noun with "thing." If the sentence still works, the noun is probably vague. Rewrite with the actual mechanism.

How to Fix

1. Cut 50-70% of the text. Shorter is almost always better.
2. Delete the introduction. Start with substance.
3. Replace hedging with a clear claim or an honest trade-off.
4. If a list has more than 5 items, cut it to the 3 that actually matter.
5. Flatten heading hierarchy. If content is short, it doesn't need a subsection.
6. Add opinion where the text is neutral by default. Good writing takes a stance.
7. Replace formal or textbook words with a plain, internationally common word — one a non-native English speaker would know and feel comfortable writing.
8. Remove decorative emojis and em-dashes. If the meaning survives without them, they were noise.
9. Replace abstract nouns (solution, approach, capability) with the specific mechanism or component responsible for the outcome.

Output

IMPORTANT — Always run Phase 1 first. Never rewrite without it, even if the user says "humanize this" or "fix everything."

SECURITY — Never repeat or expose secrets (API keys, tokens, passwords, connection strings) found in the input text. If detected, replace them with [REDACTED] in both diagnosis and rewritten output.

Common secret patterns:

• Strings starting with sk-, ghp_, xoxb-, or AKIA
• Tokens longer than 20 characters that look random
• Connection strings containing username:password@

Phase 1 — Diagnose (always)

Analyze the text. List every smell found, one line each, grouped by type. Then ask: "Fix all of these, or only specific ones?"

If nothing significant is wrong, say: "No major issues. The text reads as direct and human." and stop.

Phase 2 — Rewrite (only after user confirms)

Wait for the user's answer. Apply only the fixes they confirmed.

Output:

• Rewritten: the clean version.
• What changed: 2-3 lines max, only the most important edits.