The contrast reframe
"It's not X, it's Y." Invents a weak idea just to knock it down.
Fix: Delete the rejected half. Say the real point directly.
The Sound Human check: the prompt I run before anything leaves my laptop. It reads your draft as an editor, strips the twelve patterns that make writing sound machine-made, and changes nothing else.
Works in Claude, ChatGPT, Gemini and Copilot. Five minutes to set up as a skill.
You've also got an email on its way with links to all the free resources, so you can find them again.
Jump to the Sound Human check ↓You can't always name them, but you can smell them. Here they are, named, with what to do instead.
"It's not X, it's Y." Invents a weak idea just to knock it down.
Fix: Delete the rejected half. Say the real point directly.
Pivotal. Significant. A testament to. It declares importance instead of showing it.
Fix: State the fact. Let one specific detail show why it matters.
"...marking a significant shift." Fake analysis bolted onto the end of a sentence.
Fix: Cut it, or give the point its own sentence and a real argument.
The platform, the solution, the tool, the offering, all in one paragraph.
Fix: Pick the right word and use it again. Humans repeat words.
"Here's the kicker." "The best part?" Fanfare for a point that should stand on its own.
Fix: Delete the fanfare. Usually "But" does the whole job.
"You're not alone." "You're not imagining it." Reassurance nobody asked for.
Fix: Inform or challenge. Don't soothe.
Three adjectives, three examples, three bullets, every single time.
Fix: Use the number the point needs. Often it's one.
Additionally. Furthermore. Moreover. It's worth noting that.
Fix: Delete it. If the logic falls apart, the logic needed work.
Long dashes used as dramatic pauses, three to a paragraph.
Fix: A comma, a full stop, or a new sentence.
Every thread resolved, every section the same length, a neat summary at the end.
Fix: Be lopsided. Go deep where it matters. Cut the summary.
"Studies show." "Experts agree." No name, no year, no link.
Fix: Name the source or cut the claim.
"Look." "Let's be real." Every sentence four words long.
Fix: Vary the rhythm. Follow a long sentence with a short one.
And the words. Delve, seamless, elevate, unlock, harness, leverage, robust, tapestry. The test is simple: if you wouldn't say it out loud in a meeting, it goes.
01 is the full check. 02 is the quick one. 03 makes it a skill. 04 stops most of the tells being written in the first place.
Paste with your draft underneath. Twelve tells fixed, nothing else touched.
Read the draft below as my editor, not my writer. Check it for these twelve AI tells and fix only those: 1. "Not X, it's Y" contrasts. Delete the rejected half; say the real point directly. 2. Puffery: pivotal, significant, a testament to. State the fact instead. 3. "-ing" phrases tacked onto the end of sentences as fake analysis. Cut them. 4. Rotating synonyms for the same thing. Use one word consistently. 5. Fanfare: "here's the kicker", "the best part?". Delete it. 6. Reassurance nobody asked for: "you're not alone". Delete it. 7. Automatic groups of three. Use the number the point needs. 8. Hollow connectors: additionally, furthermore, moreover, it's worth noting. Delete them. 9. Long dashes used as pauses. Replace with a comma, a full stop or a new sentence. 10. A closing summary that repeats the piece, or suspiciously even sections. Cut the summary. 11. Unnamed sources: "studies show", "experts agree". Flag them for me to source or cut. 12. Stock casual openers and uniformly short sentences. Vary the rhythm. Also remove any word I wouldn't say out loud in a meeting (delve, seamless, elevate, unlock, harness, leverage, robust, tapestry). Keep my sentences exactly as they are unless they break a rule above. Add nothing. Don't "improve" anything else. Give me the clean version first, then a list of every change you made and which rule it broke. Draft: [paste your draft here]
For emails and short posts, when you haven't got time for the full list.
Read this back as my editor, not my writer. Strip every AI tell: dashes used as pauses, groups of three, "not X but Y" contrasts, rhetorical questions before a reveal, any sentence that summarises the paragraph above it, and any word I wouldn't say out loud. Keep my sentences exactly as they are unless they break one of those rules. Add nothing. Show me what you changed. [paste your draft here]
Run once, straight after you've used prompt 01. After this it's a forward slash.
Turn the Sound Human check we just used into a skill called sound-human. When I run it, it should take whatever draft I give it, apply the twelve rules exactly as written, keep my voice file in mind if there's one in this project, return the clean version first and the change list second, and never add new content. Save it so I can run it with /sound-human.
Paste into your global instructions (Settings, then General, then instructions for Claude) so fewer tells get written in the first place.
When you write anything for me: no long dashes; no "it's not X, it's Y" contrasts; no groups of three unless the content really has three; no hollow connectors (additionally, furthermore, moreover); no summary paragraph at the end; no words I wouldn't say out loud (delve, seamless, elevate, unlock, harness, leverage, robust). Vary sentence length. Name sources or don't make the claim.
Three ways to go further, depending on where you are.
Build your AI chief of staff in 30 days, live, with me and a small group of people running businesses. The next cohort opens when the waitlist is full. Join it and you get first access and the waitlist price.
Join the Challenge waitlist →Three live calls a month, at least four new skills a month, plugin updates. For people already using Claude daily who want to keep building. Cancel any time.
Join the membership →Four private sessions building the whole system around your actual week, or a workshop for your leadership team. Book a call and we'll work out which.
Book a call →