Old humanizer prompts went after the em dash. The newest Claude and ChatGPT models barely use em dashes anymore, and the tells that give AI writing away now are different words and phrases. I rebuilt my prompt from my own writing rules. You paste it in once and it works on every chat from then on.
In the video I said to comment UNBOT. This is that guide. The tells I see now are stock phrases like "Here's the thing," "Let that sink in," "And that matters" and "This changes everything," the "it's not this, it's that" line, and words like delve, seamless, game-changer and quietly. I hate quietly. If your saved prompt doesn't name them, it can't catch them.
The top half of the prompt tells AI what a real person sounds like. The bottom half bans the words and phrases AI leans on right now.
Write like a real person talking to one person across a table.
- Short, plain sentences. Vary the length. One idea per sentence.
- Start with the point. No warm-up.
- Say the thing directly. Never name the wrong thing to correct it: no "it's not X, it's Y," "this isn't X, this is Y," "less X, more Y."
- Use real numbers, names and details. Give one number, never a range.
- No em dashes. Use commas, colons or full stops.
- Don't list things in threes by habit. Give the real number.
- No one-word fragments for drama ("Fast. Simple.").
- Never end a paragraph by restating the point.
- Use "is," "has," "does," never "serves as," "stands as," "boasts."
Never use: delve, harness, unlock, leverage, game-changer, seamless, robust, empower, streamline, elevate, transformative, landscape, navigate, foster, utilize, crucial, comprehensive, nuanced, compelling, resonate, journey, ecosystem, unpack, quietly, truly, actually, genuinely, essentially.
Never write: "Here's the thing," "Let's dive in," "Picture this," "In today's world," "That said," "Ultimately," "At the end of the day," "And that matters," "This changes everything," "Let that sink in," "Read that again," "Think of it as."The prompt is 1,191 characters. ChatGPT Free and Go accounts can save 1,500 characters of custom instructions, and Plus and up can save 5,000. If you're on Free or Go and already saved my FACTCHECK or CODES prompt, the two together go over the limit, so keep the one you use most. To use it in every Claude chat, click your initials in the bottom left, choose Settings and paste it into Instructions for Claude.
It's a set of standing instructions that makes AI writing sound less like AI. Older ones mostly banned the em dash. Mine describes how a real person writes, with short sentences and real numbers, then bans the words and phrases AI leans on now, like "delve," "seamless" and "Let that sink in." Save it once in ChatGPT or a Claude project.
In my writing, the newest models barely use them anymore. OpenAI CEO Sam Altman posted in November 2025 that ChatGPT now follows a custom instruction to skip em dashes, and OpenAI said it still uses them by default if you don't ask. The prompt on this page includes a no-em-dash rule, so you're covered either way.
On chatgpt.com, open Settings, then Personalization, turn on Enable customization and paste into the Custom instructions field. Free and Go accounts can save 1,500 characters and Plus and up can save 5,000. This prompt is 1,191 characters, so it fits on every plan.
Put them in a project. Open the project, click Set project instructions, paste and click Save instructions. Claude follows them in every chat in that project. For every chat on your account, click your initials, choose Settings and paste into Instructions for Claude. Free accounts can make up to five projects.
The ones I ban are words like delve, seamless, game-changer, leverage, robust and quietly, and phrases like "Here's the thing," "Let that sink in" and "This changes everything." I also cut the "it's not X, it's Y" line and one-word fragments written for drama. The full list is in the prompt above.
A prompt that scans a draft for every tell on my list and rewrites only the lines that have them. Plus how to save it as a Claude skill or a ChatGPT project so it's one paste away.
Check my draft below for signs that AI wrote it. Look for: - Em dashes. - Stock phrases like "Here's the thing," "Let's dive in," "Picture this," "That said," "Ultimately," "At the end of the day," "And that matters," "This changes everything," "Let that sink in," "Read that again," "Think of it as." - Lines that name the wrong thing to correct it: "it's not X, it's Y," "this isn't X, this is Y," "less X, more Y." - These words: delve, harness, unlock, leverage, game-changer, seamless, robust, empower, streamline, elevate, transformative, landscape, navigate, foster, utilize, crucial, comprehensive, nuanced, compelling, resonate, journey, ecosystem, unpack, quietly, truly, actually, genuinely, essentially. - "Serves as," "stands as" or "boasts" where "is" or "has" would do. - Lists of three that are there for rhythm, one-word fragments for drama, and paragraphs that end by restating the point. - Ranges where one number would do. Then: 1. Give me a table with every line that has a problem: the line, what the problem is, and your rewrite. 2. Rewrite only those lines. Keep every other sentence exactly as I wrote it, and keep all my facts and numbers. 3. Don't make up a number. If a range needs to become one number, ask me which. 4. Give me the full draft with your changes in place. If you find nothing, say so and change nothing. My draft: [paste your draft]
Read the table before you take the rewrite. A flagged line can be your own voice, and you get the final say.
Prompts, step-by-step guides and checklists you can put to work today.