Build a Humanizer Writing Skill

This process helps you edit generic AI prose. It cannot guarantee a “0% AI” score or reliable detector bypass. AI detectors are inconsistent, and human writing can be flagged too.

What this helps you do

Turn a checklist of common AI-writing patterns into a repeatable editing instruction that preserves facts, adds specificity, and sounds closer to your real voice.

Before you start

Bring the draft, the facts it must preserve, and two or three samples of your real writing. Open Wikipedia's Signs of AI writing page and review its scope: it is an editing aid for spotting patterns, not a universal detector or proof of authorship.

ChatGPT's reusable Skills feature is limited by plan and workspace settings. OpenAI currently documents Personal Skills for eligible Business, Enterprise, Edu, and Healthcare workspaces. If you do not see Skills, save the prompt in a Project or a text file instead.

Terms to know

• Voice sample: real writing used as a reference for cadence and word choice.
• Unsupported claim: a statement not backed by a source or known fact.
• Writing pattern: a repeated habit such as canned framing, vague praise, or excessive sectioning.
• Skill: a reusable instruction package. Availability depends on the ChatGPT plan and workspace.

Step-by-step

1. Open a new chat and paste the first prompt below.
2. Attach or paste the Wikipedia checklist and your voice samples.
3. Paste the second prompt and ask for a line-level edit of one draft.
4. Check that every fact still matches the original.
5. If Skills are available, ask ChatGPT to package the tested instruction as Humanizer. Otherwise, save it in a Project or document.

Copy this

I want you to edit my writing so it sounds natural and specific to my voice. Preserve every fact. Do not invent stories, opinions, numbers, or personal details. Use the voice samples I provide as the standard for cadence and word choice.

Use the attached Wikipedia `Signs of AI writing` page as an editing checklist, not as proof that a person or AI wrote the text.

Review the draft for unsupported importance claims, canned introductions, vague attribution, generic conclusions, repetitive sentence structures, unnecessary headings, excessive lists, chatbot language, and formatting habits that do not match my voice samples.

First show the lines you would change and why. Then rewrite them in plain language. Preserve the meaning and facts. Leave specific, natural lines alone.

DRAFT:
[PASTE]

VOICE SAMPLES:
[PASTE]

If your account supports Skills:

Turn the editing process we just tested into a reusable skill named Humanizer. Include the fact-preservation rule, the checklist review, the voice-sample comparison, and a final line-by-line verification. Do not promise AI-detector bypass.

Things to know

The Wikipedia page changes over time and is written for Wikipedia editors. Not every listed pattern is bad in every format. A detector score should not be used to accuse a student, employee, or writer of misconduct. The meaningful test is whether the draft is accurate, specific, readable, and true to the named writer.

If something goes wrong

• If the rewrite erases your voice, ask for line edits only and add stronger voice samples.
• If it invents details, restore the source and require a fact-by-fact comparison.
• If the Skill option is missing, use a Project instruction or paste the prompt when needed.
• If a detector still flags the text, do not chase the score. Review the actual writing and disclosure rules for your setting.

Final check

The edit passes when every claim matches the source, no personal detail was invented, the revised lines sound natural when read aloud, and you can explain each change without referring to a detector score.