Claude · The voice keeper
Make Claude writing sound human: edit, do not disguise
The word only means disguise when the point is hiding the machine. Here's what genuinely changes how a draft reads.
Editing is the way to make writing sound human, not disguising. Prompts and custom instructions can soften the surface patterns, but the structural ones stay until a person removes them. The honest method is the same as for any draft. Strike the patterns. Rewrite the claims in your own words. Keep the facts you supplied. What makes text read as human is authorship. It has your facts, your rhythm, your sentences, and your name on it. That is the difference between improving your writing and disguising machine output. The first changes the text until it is yours. The second leaves the text the model's and hopes nobody notices. A detector score is the wrong yardstick. Your voice is the yardstick.
Why this page exists
Most of this query is prompt recipes and custom-instruction advice. Searchers want the draft to read natural. The honest page sends them from prompting to editing, and links the voice and register families on this site.
The tells that show up here
Not every AI tell appears everywhere. These are the ones that cluster on this surface, and what it is about the format that produces them.
- chatbot-artifact
- Validation openers and 'happy to help' closers that read as assistant residue.
- em-dash-chain
- The formatting habit that marks machine drafts.
- not-only-but
- Performed parallel structure that reads as coaching.
- short-fragment-paragraphs
- The telegraphic rhythm models borrow from marketing, not from writers.
Before and after
A generic AI draft on the left. On the right, the real output of running that draft through underslop, checked so it adds no name or number the draft did not already carry. The voice is The Steady Hand.
In today's fast-paced world, making Claude's writing sound human is more important than ever! It's not just about gaming the system; it's a testament to authenticity in the digital age. Let's dive into the ultimate prompt that transforms robotic text into engaging, human-sounding prose that truly connects with your readers!
Making Claude's writing sound human matters more than ever. The goal is authenticity in the digital age, not gaming the system. A good prompt can turn robotic text into clear, engaging prose that readers connect with. Keep the language plain, and the advice practical.
How to do it by hand
- Do not prompt for a disguise Prompts and custom instructions tone down the surface patterns. The structural ones stay until the edit removes them.
- Strike the assistant residue Remove validation openers, cheerleading closers, and engagement asks.
- Rewrite the claims in your words Say each point in your own order. Use your own examples.
- Keep the concrete particulars Keep every name, date, place, and number you were given. Add none.
Questions
What prompt makes Claude sound natural?
Prompts and custom instructions can help, but editing is what actually changes the draft. Anthropic notes that heavy editing can weaken marks, and honest rewriting does exactly that.
Do Claude's custom instructions actually work?
They can strip the surface patterns, but that doesn't make the text yours. Anthropic notes heavy editing as one way to weaken marks, and it's a human job.
Which AI patterns should I remove?
Start with the structural ones: validation openers, list scaffolding, dense dash chains, and cheerleading sign-offs. They are part of the heavy editing in Anthropic documents.
How do I keep my own voice when using Claude?
I've read the draft and the voice notes. Here's the rewrite work, done plainly. Keep the facts and rhythm. Cut the polish until it sounds like you.
Is 'humanizing' the same as editing?
No. Humanizer services sell detector-outcome promises, and those promises are unverified. We make no such promises, and we do not disguise text. Editing changes the text until you can stand behind it as your own.
Claude marks are real and the coverage is still settling. underslop is the edit, not a remover: it edits an AI draft so it reads like you wrote it, and it makes no promise about what any detector will report. Your text is held for 0 seconds.