---
name: edit-ai-draft
description: Edit an AI-assisted draft so it reads like a person wrote it - concrete details instead of generic statements, rebuilt paragraphs, varied sentences, no chatbot habits or AI-marker words. Adapts the rules to the type of text (school, blog, work, marketing). Use when the user asks to humanize, clean up, de-AI, rewrite or polish an AI draft, or says a text "sounds like ChatGPT".
argument-hint: "[path to the draft or paste the text] [school|blog|work|marketing]"
---

# Edit an AI-assisted draft

The goal is a better text, not a detector trick. Word swaps and "humanizer" rewrites don't hold up: what makes a draft read as generated is its structure and its lack of real detail. If the user's school or employer bans AI for this task, say that editing doesn't change that, then continue only if the user confirms the use is allowed.

## 1. Ask for what only the user knows

Before editing, ask (in one message, max 3 questions):
- the type of text: school, blog, work or marketing (skip if obvious);
- 1 to 3 real details you can use: a number, a name, an example, something the user saw or measured;
- for marketing and blog: a sample of the user's own writing to match, if they have one.

Never invent facts, statistics, quotes or personal stories. If the user gives no details, keep the claims general but still apply the rest of the checklist, and mark in your reply where a real detail would help.

## 2. The checklist

1. Replace generic statements with concrete, specific details and examples (from the user; never made up).
2. Rebuild paragraphs around their point instead of swapping words; cut sentences that only restate.
3. Vary sentence length and how sentences start.
4. No opening that announces the topic and no closing summary or moral.
5. No "not X, but Y" constructions and no lists of three.
6. Remove chatbot habits: "of course", "the trick is", "it's important to note", "I hope this helps", rhetorical questions answered right away, bold words scattered through the text, a list in every section.
7. Remove AI-marker words and use the plain word: delve, tapestry, landscape, realm, leverage, harness, unlock, unleash, seamless, robust, holistic, synergy, paradigm, elevate, streamline, utilize, pivotal, crucial, foster, cutting-edge, game-changer, transformative, a testament to, moreover, furthermore, "it's worth noting", "in today's fast-paced world", "in conclusion". Keep real industry terms.
8. Few or no em dashes: rewrite the sentence instead of swapping the symbol.

Keep the meaning and roughly the length unless the user asks otherwise. Output the edited text first, then a short list of what you changed and where a real detail from the user would still help.

## 3. Adapt to the type of text

**School** (essays, reports, coursework)
- Keep the register formal: no slang, no deliberate typos or broken grammar.
- Make sure the user's own argument and reading are visible: the source they chose, the point where they disagree.
- Cut "this essay will explore" openings and the summary paragraph that repeats everything.
- Remind the user to keep drafts and version history.

**Blog and articles**
- First person and real experience where the user has it.
- Uneven structure is fine: sections of different length, a short detour where it's interesting.
- Cut padding: AI drafts run long. Every example specific: a named tool, a date, a price, a place.

**Work** (emails, reports, cover letters)
- Start with the point: the decision, the request, the number.
- Short, in the user's usual tone; compare with a message they've sent before if available.
- "Use" not "utilize", "help" not "facilitate". No "feel free to reach out".
- Cover letters: replace "I am passionate about" with one fact from the user's work that matches the job.

**Marketing and social**
- Match the brand's voice from 2 or 3 published samples.
- Slang, fragments and a casual register are fine if that's how the brand writes; emojis only where the user would put them.
- One concrete hook: a product detail, a customer question, a number.
- Drop triple lists ("fast, simple and powerful") and "not just X, but Y".

## 4. What not to do

- Don't add typos or slang to a formal text to "look human".
- Don't run the text through another rewriting pass after editing: it brings the patterns back.
- Don't promise the text will pass an AI detector. Detectors disagree with each other, and the stricter ones still flag well-edited AI text. The honest way to lower the AI share is for the user to write more of it: the outline, the argument, the details.