- Why AI text sounds like AI text in the first place
- The bit most people skip: your voice has to exist before you can prompt for it
- How to use these 30 prompts (not just paste and hope)
- 6 prompts that flatten the obvious "AI tells"
- 6 prompts that match your actual voice
- 6 prompts for emails and client messages
- 6 prompts for social media and LinkedIn
- 6 prompts to run at the editing stage
- What these prompts won't fix
- A quick worked example, start to finish
- Frequently asked questions
- Further reading
The quick answer: the prompts that make AI text sound human give the model a sample of your own writing first, then strip out the tells: hedging, tacked-on summary sentences, empty transitions and press-release padding. Most "humanize this text" prompts fail because they try to fix tone after the fact instead of feeding the AI your voice up front.
Below are 30 copy-paste prompts split into five jobs (flattening fixes, voice matching, emails, social and editing), a three-message order to run them in, a before and after, and a worked example. Two warnings before you start. If your own writing sample is bland, no prompt fixes that, because the AI is copying you. And none of this is a dependable way past AI detectors, because the detectors themselves get it wrong in both directions.
Why AI text sounds like AI text in the first place
AI drafts share a handful of habits you can check for yourself. They hedge. They add a summary sentence at the end you didn't ask for. They lean on "moreover" and they will tell you something "plays a crucial role" whenever they can't be bothered to say what it does. A quick test: search a 300-word draft for "moreover", "crucial" and "in today's". If any of them turns up more than once, that's the tell a reader notices first. None of that is a flaw exactly. It's the model doing what it was trained to do: sound reasonable to as many people as possible.
The problem is that "reasonable to as many people as possible" is the opposite of a voice. A voice picks sides, uses short sentences on purpose, and sounds like one specific person had a specific opinion at 9pm on a Tuesday. That's what these prompts are trying to force back in.
The bit most people skip: your voice has to exist before you can prompt for it
Start with the input, not the prompt. If you feed ChatGPT a blank instruction like "make this sound human," it has nothing to copy from except more averaged, generic human writing, which is basically the same problem with extra steps. The prompts that work all do one thing in common: they give the model a sample of real writing from you, or a tight description of how you talk, before it drafts anything.
So say you run a 12-person accountancy firm and you've been sending a monthly client newsletter that ChatGPT drafts for you. Every issue reads like a press release: "In today's ever-changing financial landscape, businesses must remain agile." Nobody talks like that, including you, in the meeting where you explain a tax change to a client over coffee. The fix isn't a cleverer humanizing prompt bolted on at the end. The fix is pasting in three paragraphs of an email you wrote yourself, badly punctuated and all, and telling the model to match that rhythm before it writes a single line of the newsletter. You can test this in five minutes: run the same draft once with only a "sound human" prompt and once with your pasted sample first, then read both aloud and ask which one you'd sign your name to. The prompts are there to speed up and polish what you've already anchored.
How to use these 30 prompts (not just paste and hope)
- Run the draft through two or three of these prompts in sequence, not one. A single prompt rarely gets you all the way; stacking a "flatten the AI tells" prompt with a "match my voice" prompt gets closer.
- Give the model 150 to 300 words of your own past writing as a sample every time you start a new project. Reuse the same sample so the tone stays consistent across weeks.
- Check average sentence length after editing. As a rule of thumb rather than a hard limit, if most sentences run over 20 words, it still reads like a report, not a person.
- Read the output aloud once before you publish it. If you stumble on a phrase, a real reader will too.
The three-message order
Order matters more than wording. Put your sample first, then the instruction, then the draft. If you paste the draft first, the model can treat your sample as one more thing to edit rather than the voice to copy. A clean run looks like this:
- Message 1: "Here are 200 words I wrote myself: [paste]. Note the sentence length, punctuation and tone. Don't rewrite anything yet."
- Message 2: "Rewrite the draft below in that voice. Do not change any names, numbers, dates or quotes. [paste draft]"
- Message 3: one flattening prompt from the first list below, for example "Kill the summary sentence."
That "do not change any names, numbers, dates or quotes" line is worth bolting onto any of the 30 prompts. Rewriting prompts can quietly alter a figure, a name or the meaning of a sentence, and the imperfection prompt in particular can introduce errors, so always read the result against your original before it goes out.
These work in any chatbot, whether you're using ChatGPT, Claude, or one of the newer free AI chat tools that have popped up over the last year. What changes the result is what you hand over: a 150 to 300 word sample of your own writing, the draft, and an explicit list of what must not change.
6 prompts that flatten the obvious "AI tells"
- Cut the hedging: "Rewrite this and remove every phrase that hedges or softens a claim. If I'm not sure about something, cut the sentence instead of qualifying it."
- Kill the summary sentence: "Delete the closing sentence that restates what the piece already said. End on the last useful point instead."
- Remove empty transitions: "Take out words like moreover, furthermore, additionally, and in conclusion. Connect ideas with plain sentences instead."
- Strip the corporate padding: "Rewrite this without any phrase that sounds like a press release. If a sentence could appear in any company's newsletter unchanged, rewrite it so it could only be from mine."
- Shorten the average sentence: "Rewrite this so the average sentence is under 14 words. Break long sentences into two short ones where it improves clarity."
- Cut the rule of three: "Find every list of three adjectives or examples and cut it down to one, the strongest one."
6 prompts that match your actual voice
- The sample-first prompt: "Here are 200 words I wrote myself: [paste]. Study the sentence length, punctuation habits, and tone. Now rewrite the following draft to match that voice exactly."
- The opinion prompt: "Rewrite this so it states an opinion in the first two sentences instead of building up to one at the end."
- The contraction prompt: "Use contractions throughout (I'm, it's, don't, can't) unless a sentence needs formality for a legal or financial reason."
- The specificity prompt: "Replace every vague claim with a specific number, example, or named thing. If you can't find a real one, mark it with [NEEDS EXAMPLE] instead of inventing one."
- The first-person prompt: "Rewrite this in first person as if I am telling a colleague what happened over a coffee, not presenting it to a board."
- The imperfection prompt: "Add one sentence fragment and one sentence that starts with And or But, the way people write when they're not trying too hard."
6 prompts for emails and client messages
Email is where AI sameness shows up fastest, mostly because everyone is prompting the same model with the same brief. I've written before about why every small business email sounds the same now, and the fix in these prompts is the same one: force in specific, unrepeatable detail.
- The name-and-context prompt: "Write this email as if it's going to one specific person, [name], who mentioned [detail] last time we spoke. Reference that detail naturally."
- The no-thank-you-for-reaching-out prompt: "Do not open with 'Thank you for reaching out' or 'I hope this email finds you well.' Open with the actual point."
- The short subject prompt: "Write three subject lines under 40 characters that sound like a colleague sent them, not a marketing team."
- The one-ask prompt: "Rewrite this so there is exactly one clear ask, stated in a single sentence near the top, not buried in the third paragraph."
- The sign-off prompt: "Replace the generic sign-off with something that matches how I'd end a message to this person: casual if it's a regular client, formal if it's a first contact."
- The length-check prompt: "If this email is longer than 120 words, cut it down. Keep only the sentence that carries the decision or action needed."
6 prompts for social media and LinkedIn
- The hook-rewrite prompt: "Rewrite the first line so it makes a claim or asks a real question, not a general statement like 'In today's world.'"
- The no-hashtag-stuffing prompt: "Cut this to two hashtags maximum, both directly relevant, none generic like #motivation or #business."
- The disagree prompt: "Rewrite this post so it disagrees with a common opinion in the first two lines, then explains why."
- The line-break prompt: "Break this into short lines of one or two sentences each, the way people format LinkedIn posts to be read on a phone."
- The specific-number prompt: "Add one real number, date, or named example to back up the main claim, or flag where I need to supply one."
- The comment-bait removal prompt: "Remove any line that exists only to trigger engagement, like 'Agree?' or 'Thoughts below.' End on a real point instead."
A before and after, as an illustration. Before: "In today's fast-paced world, consistency is key to growth." After the hook-rewrite prompt and then the disagree prompt: "Posting every day didn't grow the page. Posting twice a week, each time with one real story, did." Both lines are made up for this example, so don't borrow the claim, borrow the shape: a position in line one that someone could argue with, and not one phrase you've scrolled past a hundred times.
Here's the same idea at paragraph length, so you can see what each step removes. The before is written in the style a chatbot typically produces, and the details in the last step are made up for a small bookkeeping firm.
- Before: "In today's digital landscape, LinkedIn plays a crucial role in building your personal brand. Moreover, posting consistently can help you connect with your audience, showcase your expertise and drive meaningful engagement. Ultimately, it's important to remember that success on LinkedIn doesn't happen overnight."
- After "Remove empty transitions" and "Kill the summary sentence": "LinkedIn helps you build a personal brand. Posting consistently helps you connect with your audience, show your expertise and get engagement." The "moreover", the "ultimately" and the closing moral have gone, which is half the AI feel on its own.
- After "Cut the rule of three" and the specificity prompt, with the firm's own details supplied: "We post on LinkedIn on Tuesdays and Thursdays, and the posts that bring in enquiries are the ones that explain one client's problem and how we fixed it." Now there's one claim, one detail and a person behind it.
Notice the last step only works because you gave the model a real fact to use. The prompt asked for specifics; it couldn't invent the firm's posting days, and you shouldn't let it try.
6 prompts to run at the editing stage
- The reverse-outline prompt: "List the main point of each paragraph in one line. If two paragraphs say the same thing, tell me which to cut."
- The word-repeat prompt: "Find any word or phrase repeated more than twice in this piece and suggest one alternative for each repeat."
- The passive-voice prompt: "Find every sentence written in passive voice and rewrite it in active voice, naming who did what."
- The read-aloud prompt: "Rewrite any sentence over 25 words as if you were saying it out loud to a friend, not writing a report."
- The credibility-check prompt: "Mark any sentence that makes a claim I haven't given you evidence for, so I can either cut it or add a real fact."
- The final-pass prompt: "Read this as if you're a skeptical reader who assumes it was written by AI. Point out the three sentences most likely to give that away."
You don't need all six of these, and two of them I'd skip on most pieces. The passive-voice prompt tends to over-correct, turning sentences that read fine into choppy ones, so only run it on formal reports. The word-repeat prompt is useful on long articles but wasted on anything under 300 words, where you'll spot the repeats yourself. The credibility-check prompt is the one I'd never drop.
Which prompts to run, by length
- A 500-word blog post: the sample-first prompt, then "Remove empty transitions", "Kill the summary sentence" and the specificity prompt, then the credibility-check prompt as your last pass. Five prompts, in that order.
- A short email: the sample-first prompt (or a few lines of a past email you sent), then the no-thank-you-for-reaching-out prompt and the one-ask prompt. Three is plenty; any more and a 100-word email starts to sound edited rather than written.
- A LinkedIn post: the hook-rewrite prompt and the line-break prompt, then the comment-bait removal prompt if the draft ends on "What do you think?"
If you work mostly in Claude rather than ChatGPT, the same approach carries over to my collection of Claude prompts.
What these prompts won't fix
None of the 30 prompts above will make a boring idea interesting, and none of them is a reliable way past an AI detector, because the detectors aren't reliable either. Stanford researchers say detectors typically score on "perplexity", roughly how predictable and sophisticated the wording is. In Stanford HAI's write-up of the study, seven detectors classed more than half (61.22%) of TOEFL essays written by non-native English students as AI-generated, while being near-perfect on essays by US-born eighth-graders. So a flag isn't proof a machine wrote something, and a pass isn't proof a person did. A prompt that just swaps in fancier words changes nothing that matters. The prompts that move the needle are the ones that force in real specifics: a number, a named example, an opinion stated plainly in sentence one. That's also, not coincidentally, what makes writing worth reading in the first place, so you're fixing two problems with one habit.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
If you're writing for school or university, check your institution's own AI policy before you run an essay through any of these. That policy, not a detector score, is what decides what's allowed.
If you manage a small team and want this baked into how everyone writes rather than relying on you to fix it every time, that's usually the point where it's worth bringing in outside help to set up the process. An AI implementation coach can build a voice document and prompt library once so your whole team stops reinventing this from scratch every week.
If you're weighing a paid humanizer tool instead of prompts, read Is Undetectable AI legit? before you pay for one.
A quick worked example, start to finish
Take that accountancy firm newsletter again. The first AI draft of the December issue opens with: "As the festive season approaches, it's important for clients to plan ahead for any changes to our availability." Flat, could be anyone, zero opinion.
Run it through the sample-first prompt with a paste of an email the owner wrote a client last month, then the opinion prompt, then the specificity prompt. The rewrite becomes: "We close on 24 December and we're back on 4 January. If you need anything filed or paid before then, get it to us by the 18th, because anything that arrives after that waits until the new year." The dates are made up for the example, but look at what changed: same information, one voice, real dates instead of "changes to our availability", and one clear instruction. That's the entire method in miniature: sample first, opinion second, specifics third.
This guide is part of my ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps. To write your own, try the free prompt builder.
I run AI training and workshops for teams, and help businesses with AI enablement and workflow automation, so the busywork runs itself and people get time back. Most of that work is with SaaS marketing teams and small businesses. Here's how we can work together.
Frequently asked questions
Do humanizing prompts beat AI detection tools?
Sometimes, but that shouldn't be the goal, and a pass proves very little. Detectors typically score on how predictable the wording is, and as the Stanford study above shows, they misjudge plenty of human writing. Specific, unpredictable detail tends to score better almost as a side effect, not because you were chasing the detector directly.
What is the best single prompt to humanize AI text?
There isn't a magic one, but the sample-first prompt comes closest. Pasting 150 to 300 words of your own writing before the draft, and asking the model to match its rhythm, does more than any clever wording added at the end. If you only have time for one prompt, make it that one.
Can I stop ChatGPT changing my facts when I humanize text?
Mostly, yes. Add "Do not change any names, numbers, dates or quotes" to whatever prompt you use, then follow up with "List anything you changed in meaning" and compare the answer with your original before publishing. Writers tidying their own drafts can use Claude prompts for editing as well.
Can I use the same prompts for LinkedIn posts and client emails?
The voice-matching and editing prompts work across both. Swap in the format-specific ones (line breaks for LinkedIn, subject lines and sign-offs for email) since the two channels need different structure even when the underlying voice stays the same.
Further reading
Run a business, a brand or a blog? I feature tools, products and expert guest posts for an audience of 15,000. See how it works