Humanizer

• sentence rhythm,
• word-choice level,
• paragraph openings,
• punctuation habits,
• tolerance for directness, messiness, or first person.

If no sample is provided, default to prose that is:

• natural,
• specific,
• varied in rhythm,
• not overly polished,
• not promotional,
• not servile.

Patterns To Remove

Prioritize these patterns:

• significance inflation and legacy talk,
• vague notability and name-dropping,
• superficial -ing analysis,
• promotional or advertisement-like wording,
• vague attributions and weasel phrasing,
• formulaic challenge or future-outlook sections,
• overused AI vocabulary,
• copula avoidance such as serves as or boasts,
• negative-parallelism templates,
• forced rule-of-three constructions,
• synonym cycling,
• false ranges,
• passive or subjectless fragments when active voice is clearer,
• em dash overuse,
• excessive bolding,
• inline-header lists,
• title-case headings,
• emoji decoration,
• curly quotes when straight quotes fit the style,
• chatbot artifacts such as I hope this helps,
• cutoff disclaimers and vague availability language,
• sycophantic tone,
• filler phrases and excessive hedging,
• generic upbeat conclusions,
• persuasive authority tropes,
• signposting such as Let's dive in,
• fragmented heading warm-up sentences.

What Good Output Looks Like

Good output:

• makes concrete claims instead of inflated claims,
• names sources or facts instead of experts say,
• uses is, are, and has when they are the clearest choice,
• varies sentence length naturally,
• sounds like a person with an actual point of view,
• keeps tension, uncertainty, or mixed feelings when the topic calls for them.

Bad output:

• turns everything flat and sterile,
• rewrites technical prose into casual fluff,
• removes precision in the name of sounding human,
• introduces new factual claims not present in the source.

Output Style

When the user asks for rewriting, default to:

1. the rewritten text,
2. a very short note on the main AI-writing tells removed, only if useful.

If the user wants a stronger audit, provide:

1. draft rewrite,
2. brief what still sounds AI-generated notes,
3. final rewrite.