Humanize

When to use this skill

Activate when the user asks to humanize text, remove AI tells, make writing sound less robotic, or edit prose that feels generated. Accept text pasted inline, from a file, or attached.

What the research says

AI text detectors work by finding predictable statistical signatures. Turnitin's own product officer describes the difference as "the more idiosyncratic, unpredictable nature of human writing and the very predictable statistical signatures of AI generated text." DetectGPT found that AI text sits in "negative curvature regions" of probability — meaning it follows the model's most likely next-token paths. When Weber-Wulff's team tested 14 commercial detectors, they found the tools hunt repetition and statistical regularity. Paraphrasing breaks them because it introduces the irregularity human writing naturally has.

The goal of this skill is not to "beat detection." It is to write prose that lacks the mechanical regularity of generated text — the same regularity that makes prose feel lifeless to human readers.

Core Rules

1. Break probability patterns. AI text follows the model's highest-likelihood paths. Human writing surprises. Include an unexpected detail, a minor contradiction, or a personal aside that a model trained on averages would not generate. If every sentence feels like the "most likely" continuation of the last, rewrite one to be the "second most likely."

2. Name the actor. Every sentence needs a human subject doing something. No passive voice. No inanimate objects performing human verbs ("the data suggests," "the model reveals," "the culture shifts"). Name who decided, who built, who observed.

3. Be specific, not declarative. "The implications are significant" → name the specific implication. "This is a complex problem" → describe the actual conflict. AI text defaults to vague summaries because it lacks a point of view. Give it one.