25 pages · updated 2026-09-20 · no signup to read any of them
Each page below takes one specific thing that makes text read as machine-written — an em dash pile-up, a cover letter that opens with "I am writing to express my sincere interest", a product description that lists features instead of scenes — and shows the rewrite. They are short on purpose: read one, fix your draft, close the tab.
What is here
25 guides, one problem each, with before/after examples pulled from real drafts.
Every page says plainly what it will not do — no promise of passing an AI detector, no rewrite of your facts or numbers.
Written to be read on a phone during a commute, not to keep you scrolling.
All 25 guides
How to make ChatGPT text sound human Six concrete edits that make ChatGPT output read like a person wrote it — sentence rhythm, unhedged claims, specific detail, and the words to cut.
Why your cold email reads as AI-written The delivery tells that make a cold email read as generated, plus a 90-word structure that survives a real reader's first two seconds.
Why em dashes give away AI writing Em dashes are a weak but real AI signal. Here is how readers notice them, why the cluster matters more than the mark, and what to do instead.
Rewriting a blog intro that reads as AI Why the first two sentences get skipped, four openings that work, and a method for cutting the warm-up paragraph entirely.
Making AI text sound like you wrote it How to extract your own writing patterns and apply them to generated drafts — the four things worth copying from your own past writing.
Résumé bullets that sound AI-written Six patterns that make AI-written résumé bullets obvious to a recruiter, and the rewrite that turns each one into a line you can defend in an interview.
YouTube descriptions that sound AI-written What makes a generated YouTube description obvious in the first two lines, and how to write one that survives past the preview cut.
Support replies that sound like a bot Generated support replies fail in a predictable way: they restate the question before answering it. Here is the ordering that makes a reply read as help.
Etsy 13-tag filler — type one product phrase, get thirteen tags to edit, each capped at the 20 characters Etsy enforces. Runs in your browser, nothing uploaded.
AI-tell checker — paste a draft and see the phrases and rhythms that make it read as generated.
If you would rather not do it by hand
ShipCopy rewrites a draft you already have — in one of three modes: make it read human, turn a product description into launch copy, or turn a plain-language request into a reusable prompt. Six output formats, 13 languages. 10 generations free, no card required.
If you want this done to your own draft: ShipCopy takes a piece of text you already wrote and rewrites it in one of three modes — rewrite it to read human, turn a product description into launch copy, or turn a plain-language request into a reusable prompt. Six output formats, 13 languages. 10 generations free, no card.
Get the checklist as a file you can actually use. One email with the full list, plus the
three tells that get caught most often. No drip, no "just checking in" follow-up.
The three that get caught most often
1. Every paragraph the same length. Real writing is lumpy.
2. A summary sentence at the end of each section that repeats the heading.
3. “Not just X, but Y” used more than once in a document.
The full 12-point list: the checklist.
30 tells with before/after examples: signals.
What this page will not do. It will not tell you that any rewrite is detectable-proof, and it will not rewrite facts: names, figures and dates stay exactly as you wrote them. Detection tools disagree with each other, and we do not claim to beat them.