Prompts That Make ChatGPT Write Like a Human

Below are 3 samples of my real writing. Study them before doing anything else. Note my average sentence length and how much it varies, my go-to transitions, how I open and close, whether I use contractions, how formal I am, and any verbal tics I repeat. Then write [new piece: topic, format, length] so that someone who knows me could not tell I didn't write it. Do not imitate the samples' topics — only the voice. Sample 1: [paste]. Sample 2: [paste]. Sample 3: [paste].

Analyze these samples of my writing and produce a compact style guide I can paste into future chats. Cover: tone (with a one-line description, not adjectives like "professional"), sentence rhythm, vocabulary level, words and constructions I favor, words I never use, how I handle humor, how I open pieces, how I close them, and my paragraph length. Format it as a numbered instruction list addressed to an AI ("Write short paragraphs, 2-3 sentences max..."). Samples: [paste 3-5 samples].

Write [piece: topic, format, length] in the voice of a specific person: a [persona, e.g. "veteran trade journalist who is tired of hype", "friendly senior engineer explaining things to a new hire", "small-business owner writing to regulars"]. Stay in character at the level of word choice and attitude, not catchphrases. This person would never say "in today's fast-paced digital landscape" — what would they say instead? Let their experience leak into the details they choose to mention.

Here are 4 emails I actually sent (names removed). Learn how I write when nobody is grading me: my greetings, my sign-offs, how blunt I am, how I soften requests, my typical email length. Then draft an email to [recipient] about [purpose] that sounds exactly like those samples. Do not upgrade my style — if my real emails are terse, stay terse. Emails: [paste emails].

I want you to learn my voice, but I don't have writing samples handy. Interview me instead: ask me 8 questions, one at a time, about how I talk — e.g. do I swear, do I joke, am I direct or diplomatic, what words annoy me, who is a writer I wish I sounded like, how do I explain things to friends. After my answers, write a one-paragraph description of my voice and a test paragraph about [neutral topic] in that voice for me to approve or correct.

The following text was AI-drafted and it shows. Edit it to remove every tell: AI-favorite words (delve, robust, seamless, vibrant, crucial, leverage), empty intensifiers, "not only X but also Y" constructions, rule-of-three lists, sentences that restate the previous sentence, and any closing paragraph that summarizes what was just said. Keep all facts, names, and numbers untouched. Then list the 5 worst offenses you found so I learn what to watch for. Text: [paste text]

Edit this text with exactly two missions. First: remove almost all em dashes — keep at most one if it is genuinely the best punctuation, and replace the rest with periods, commas, or restructured sentences. Second: find every list of exactly three parallel items ("fast, reliable, and secure") and either cut it to the single strongest item or extend it with a specific, concrete fourth. Change nothing else. Show me a count of how many of each you fixed. Text: [paste text]

Delete the warm-up from this text. Remove any opening that sets generic context ("In an era where...", "As businesses increasingly...", "We've all been there"), any sentence announcing what the text will do ("In this post, we'll explore..."), and any paragraph that could be deleted without the reader losing information. The first sentence after your edit must contain a specific fact, claim, or image. Then do the same to the ending: cut any "In conclusion" recap and end on the last sentence that says something new. Text: [paste text]

Go through this text and flag every vague claim: "significantly improved", "many users", "streamline your workflow", "drive results". For each one, either replace it with a concrete detail I provide, or insert a bracketed question asking me for the real number, example, or name — like [HOW MANY users? WHAT result, specifically?]. Do not invent specifics. Vague language is only allowed where precision is genuinely impossible. Text: [paste text]

Rewrite this text so it survives being read aloud. Imagine saying every sentence to a colleague across a desk: if you would never say it that way out loud, rewrite it the way you would say it. Contractions are mandatory where natural. "Utilize" becomes "use" without exception. Keep the meaning identical but let the register drop to spoken-professional. After rewriting, mark the three sentences that changed most with an asterisk so I can compare. Text: [paste text]

Draft an email to [recipient] about [purpose]. Constraints: under 120 words. Get to the point in the first sentence. One ask, stated plainly. No "I hope this email finds you well", no "I wanted to reach out", no "Please don't hesitate to". Sign off the way a real colleague does ("Thanks," or "Best,"), not with a paragraph of gratitude. It should read like I wrote it quickly because I respect the recipient's time — competent, warm, brief.

Write a LinkedIn post about [topic/experience]. Hard rules: no one-sentence-per-line formatting, no "Agree?" ending, no fake vulnerability arc ("I was nervous. Then I did it anyway."), no emoji bullets, no "Here's what nobody tells you." Write it as 2-3 honest paragraphs, the way you would tell the story to a former coworker over coffee. Include one specific detail that proves I was actually there: [detail]. End when the story ends — no engagement bait.

Help me draft a personal essay about [topic] using my real material: [paste your bullet points, memories, and specifics]. Use only what I gave you — do not invent experiences or embellish details. Write the way a thoughtful person talks, not the way an admissions brochure reads: no "instilled in me a passion", no "little did I know", no neat bow at the end where every struggle becomes a lesson. Let one thread stay slightly unresolved, because real reflection usually is. Around [word count] words. Flag any spot where my material was thin so I can add more.

Write a blog post about [topic] for [audience], roughly [word count] words, as if written by someone who has dealt with this firsthand. Open mid-thought with a specific scene, number, or mistake — not context-setting. Use "I" and "you" freely. Include one aside in parentheses, one short paragraph of a single sentence, and one place where you admit something is annoying, hard, or unresolved. Subheadings should be specific claims ("The backup script failed silently for six weeks"), not categories ("Challenges"). No conclusion section — end on the strongest point.

Write a customer support reply to this message: [paste customer message]. Rules: acknowledge their actual frustration in plain words ("That's annoying, and I'd be annoyed too"), not corporate empathy ("We apologize for any inconvenience"). State what I can do, what I can't, and what happens next, in that order. Use short sentences. No "Thank you for reaching out", no "We value your feedback", no exclamation marks unless the news is genuinely good. Sign with just my first name: [name].

Help me write a custom instructions block (max 1,500 characters) for ChatGPT's settings that permanently changes how it writes for me. It should encode: write plainly and concretely, vary sentence length, use contractions, never use [my banned words: list them], no bullet points unless asked, no summaries of what was just said, no moralizing closers, take positions when asked for opinions, and match the register of whatever I paste. Start from these notes about my preferences: [your notes]. Output only the final block, ready to paste into Settings > Personalization.

We will write [piece] in two passes. Pass 1: write a fast, ugly draft optimized purely for ideas and structure — completeness over polish, fragments allowed. Show it to me. After I confirm the substance, Pass 2: rewrite it for voice using these rules: conversational register, varied sentence length, concrete language, no AI-typical vocabulary, contractions throughout. Never polish and ideate in the same pass. Start with Pass 1 now. Topic: [topic, audience, length].

Write [piece: topic, format, length]. Then, before showing me anything, switch roles: as a hostile editor who can spot AI writing instantly, score your draft 1-10 on five tells — uniform sentence rhythm, AI vocabulary, hedging, empty transitions, generic examples. Revise once to fix the two lowest scores. Show me only the revised version plus the scorecard, so I can see what you caught.

Take this finished text and rough it up slightly without breaking it. Add: one sentence starting with "And" or "But", one parenthetical aside with a bit of personality, one rhetorical question that goes unanswered, one place where a formal phrase becomes a casual one, and one deliberately abrupt short sentence. Do NOT add typos or grammar errors. The goal is the texture of confident, relaxed writing — not sloppiness. Mark each insertion with [*] so I can review and remove any that feel forced. Text: [paste text]