Text Humanizer

Problem

LLM output often has a template feel: inflated significance, media-coverage padding, -ing clauses, promotional adjectives, vague authorities, forced triads, synonym cycling, and passive subjectless fragments. The text may be accurate but still sound machine-generated. The skill targets this specific failure mode: removing AI patterns without deleting information or changing structure.

How it works

It scans for common signals such as serves as, pivotal, landscape, showcasing, not only...but also, and formulaic challenge sections. Then it rewrites at the same scope: preserving facts, claims, paragraph order, and intended tone while replacing stiff phrasing, excessive emphasis, and repetitive rhythm. If a writing sample is provided, it uses that sample to match sentence length, diction, punctuation, transitions, and verbal habits. By default, it also avoids sterile neutrality when the genre calls for voice, while keeping technical, legal, or reference text plain and verifiable.

Boundaries

Use it for blog posts, product descriptions, research summaries, social copy, and other prose that needs a human voice. It should not be used to invent facts, rewrite code or formulas, alter precise technical terminology, or add claims not present in the source. The output stays close to the original scope and should not be treated as factual enrichment.