Humanizer

Overview

Humanizer is an agent skill from blader that solves a problem every team using AI writing tools has encountered: the output sounds like AI wrote it. Not because it is wrong, but because language models default to the most statistically likely phrasing — the choice that fits the widest audience and subject. Humanizer identifies those defaults and rewrites them into prose that reads like a specific person wrote it for a specific audience.

Key Capabilities

25 ranked detection patterns: Humanizer catalogs the specific tells of AI-generated writing, ordered by strength and frequency. The first five patterns — hedge phrases, gratuitous signposting, symmetrical structures, inflated framing, and leftover conversational scaffolding — justify an edit on a single sighting. Weaker patterns are flagged only when multiple tells cluster in the same passage, since a skilled human writer might use any one of them intentionally.

Voice matching: When given a writing sample (2-3 paragraphs), Humanizer matches the author’s specific style — rhythm, word choice, punctuation habits, and deliberate quirks like dash usage. The rewrite sounds like the person, not like a generic “more casual” setting.

Fact preservation: Humanizer never invents facts. Names, numbers, dates, quotes, citations, and other factual details are preserved verbatim from the source. If a sentence needs a detail that is missing, Humanizer asks rather than fabricating one.

File-aware editing: When pointed at a file, Humanizer rewrites only the prose — code blocks, data tables, frontmatter, and link targets are left untouched. This makes it safe to run against markdown documents, blog posts, and documentation files.

Show-your-work mode: On pasted text, Humanizer shows its process: the initial rewrite, a self-critique identifying anything that still sounds artificial, and the final version. This transparency lets writers learn the patterns and improve their own AI-assisted writing process.

Why It Matters

AI-generated text is increasingly identifiable — not by formal detection tools, but by readers who have developed pattern recognition for the style. This recognition erodes trust and credibility, especially in professional contexts. Humanizer treats AI writing tells as an engineering problem with a concrete solution: identify the specific patterns, rank them by severity, and rewrite systematically. The 25-pattern framework makes the problem legible and the fixes predictable.

Use Cases

Content teams use Humanizer to post-process AI-drafted blog posts, marketing copy, and documentation before publication. Technical writers run it on generated documentation to ensure it matches existing voice and style. Professionals who use Claude for email drafting pass output through Humanizer to maintain their personal writing voice. Teams publishing under individual bylines use voice matching to ensure consistency across AI-assisted and manually-written pieces.

Considerations

Humanizer operates on completed text — it is a post-processing step, not a generation-time constraint. The skill works best with prose; it does not attempt to humanize code, structured data, or highly formulaic content like changelogs. Voice matching requires a representative writing sample; a single paragraph may not capture enough stylistic signal for accurate matching. The 25-pattern framework is opinionated — some patterns it flags (like certain transition phrases) may be intentional choices in specific writing contexts.

Who It’s For

Writers, content teams, and professionals who use AI to draft text but need the output to sound like them, not like a language model. Technical writers maintaining voice consistency across large documentation sets. Anyone who has had a reader say “this sounds like ChatGPT wrote it” and wants a systematic fix rather than manual editing.