30 ChatGPT Prompts for Human Responses (Copy & Paste)

AI writes quickly. Sounding like a person is a separate problem, and if you have ever read a reply that was technically correct and completely flat, you already know the difference. A good ChatGPT prompt for human responses works by directing rhythm, register and reasoning — not by adding the word "natural" and hoping.

There are 30 below, covering everyday tone, empathetic replies, professional writing, stripping out the tells, and teaching a model your own voice. Paste them into ChatGPT, or run them in Chat Smith, a free AI chatbot that keeps several models in one place — useful here, since models differ sharply in how naturally they write.

• Give it a role. “A thoughtful friend giving honest advice” changes rhythm and word choice more than any tone adjective.
• Describe the feeling, not the tone. “Calm and quietly reassuring” works; “professional” doesn’t.
• Name the reader. Knowing who is listening sets the register automatically.
• Constrain the shape. Varied sentence length, no bullet points, at least one very short sentence — structure matters more than vocabulary.
• Ask for reasoning, not just conclusions. Showing the thinking adds texture, and a long-context model like GPT-5.6 Sol keeps a voice consistent across a long thread.
• Stay honest about authorship. Sounding natural is not the same as passing AI writing off as your own — where disclosure is expected, it still applies.

Five prompts for everyday replies. Notice how many forbid structure — no headings, no bullets, no summary opening. Most of what makes text feel machine-written is formatting rather than vocabulary, and some models need less of this correction than others: Claude Sonnet 5 tends to vary its rhythm well out of the box.

1. The thinking-out-loud reply

Answer this as a person thinking out loud rather than delivering a conclusion: [paste your question]. Show the consideration — start somewhere, notice a complication, arrive somewhere. Vary your sentence lengths. It is fine to end less certain than you began. Do not open with a summary of what you are about to say.

2. The coffee conversation

Explain [topic] the way you would to a friend across a table who asked out of genuine curiosity. Assume they are intelligent but know nothing about this. Use one concrete example from ordinary life. No headings, no bullet points, no numbered structure — just talk. Under 200 words.

3. The register dial

Here is something I wrote: [paste it]. Rewrite it three times at different levels of formality: how I would say it to a close colleague, to a client I have met twice, and to a stranger in writing. Then tell me which register the original was actually in, whether or not I intended it.

4. The rhythm pass

Here is a passage: [paste it]. It reads stiffly. Fix the rhythm without changing the meaning: use contractions where a person would, break up any sentence carrying more than two clauses, and let at least one sentence be very short. Show me the before and after of the three sentences you changed most.

5. The specificity injector

Here is my draft: [paste it]. It is vague in a way that makes it sound generated. Find every general statement and tell me what specific detail, number, name or example would replace it. Do not invent the details — ask me for the ones you need. Human writing is specific, and that is most of the difference.

Five prompts for messages that matter to the person receiving them. The common failure here is rushing to resolve, which is what makes a reply feel automated even when the words are kind. For sensitive replies, a careful model such as Claude Sonnet 4.6 is good at holding back instead of jumping straight to advice.

6. The acknowledge-first reply

Someone told me this: [paste what they said]. Write a reply that acknowledges what they are feeling before offering anything else, and that does not rush to fix it. If advice belongs at all, it goes last and briefly. Do not use the phrase "I understand how you feel".

7. The difficult news message

I need to tell someone [describe the news and my relationship to them]. Write it so the news is in the first two sentences, the reasoning is brief and honest, and the close gives them something concrete. No corporate euphemism, no excessive apology, and nothing that sounds like it was written by a department rather than a person.

8. The apology that lands

I need to apologise for [describe what happened honestly, including my part in it]. Write an apology that names what I did rather than how they felt, does not use the word "if" anywhere, offers what I will do differently, and asks for nothing in return. Under 120 words.

9. The disagreement without the edge

I need to disagree with [describe the person and situation]. My position: [state it]. Write a reply that takes their view seriously enough to state it back accurately, then explains where I differ and why. Warm but not soft — I am not conceding the point. Do not open with "I hear you".

10. The check-in message

I want to check in with someone going through [describe the situation]. We are [describe the relationship and how close you are]. Write a short message that makes it easy for them to reply with anything, including nothing at all. Avoid platitudes, do not ask them to explain their situation, and do not offer help I have not actually thought through.

Five prompts for work writing. The goal is to sound like a competent person rather than a department — the same standard worth applying to anything an AI email writer produces for you.

11. The email that sounds like a person

Write an email to [describe the recipient and my relationship to them] about [topic and what I need]. Sound like a competent person rather than a department: contractions where natural, one clear ask, no "I hope this email finds you well", no "please do not hesitate to reach out". Under 120 words.

12. The de-corporatiser

Here is something written in corporate language: [paste it]. Translate it into what a person would actually say. Keep every piece of substance, remove every phrase that exists only to sound professional, and tell me if stripping the jargon reveals that a sentence was not saying anything at all.

13. The plain-language explainer

Explain [technical or complex topic] to [describe the audience and what they already know]. Use their existing knowledge as the bridge rather than simplifying the substance. Define any term you keep. Then tell me which part of the explanation is most likely to be the one they still do not follow.

14. The hedge remover

Here is my draft: [paste it]. I am hedging — "I think maybe", "it might be worth", "just wanted to check". Remove the hedges without making me sound arrogant, and flag anywhere the hedge was actually appropriate because the uncertainty is real. There is a difference between hedging and honesty.

15. The meeting follow-up

Here are my notes from a conversation: [paste them]. Write the follow-up message: what we agreed, what is outstanding, who does what by when. Keep it in the voice of someone who was actually in the room — reference something specific that was said. Nothing that reads like minutes generated by software.

Six prompts for text that already reads as generated. Run 16 before 17 — diagnosing the specific tells produces a far better rewrite than asking for a natural version, which is also why an AI sentence rewriter or AI rewording tool alone rarely fixes the problem.

16. The tell detector

Here is a passage that reads as machine-generated: [paste it]. Identify the specific tells: hedged claims, three-item lists, "it is not just X, it is Y" constructions, empty transitions like moreover and in conclusion, uniform sentence length, and paragraphs that all run the same size. Quote each instance. Do not rewrite anything yet.

17. The tell remover

Using the tells you just identified, rewrite the passage to remove them while keeping the argument identical. Do not replace one formula with another — if you find yourself reaching for a different stock structure, break the pattern instead. Vary paragraph length deliberately rather than evenly.

18. The over-explanation cut

Here is my draft: [paste it]. Find every place where I explain something I have already shown, restate a point in different words, or add a sentence summarising the paragraph that just ended. List them, then show me the version with them cut. This habit is one of the strongest signals of generated text.

19. The transition audit

Here is a draft: [paste it]. Look only at how sentences and paragraphs connect. Flag every transitional phrase that adds nothing — moreover, furthermore, that being said, it is worth noting that. For each, tell me whether the connection should be implicit instead, or whether the two ideas were never actually connected in the first place.

20. The enthusiasm calibrator

Here is a draft: [paste it]. It is enthusiastic in a way that reads as artificial — everything is exciting, powerful or game-changing. Strip the adjectives doing emotional work that the facts should be doing, then tell me which claims collapse once the enthusiasm is removed. Those either need evidence or need deleting.

21. The paragraph shape breaker

Here is my draft: [paste it]. Every paragraph runs roughly the same length and follows the same internal shape. Restructure so the rhythm varies: some long, at least one a single sentence, and the ordering driven by the argument rather than by symmetry. Do not change the content itself.

Five prompts that go further than any tone instruction: extract your actual voice from things you have already written, then hold the model to it. Run 22 once and reuse the output forever, alongside whatever AI writing tools you use day to day — and if you work in Claude as well, our Claude prompts for writing pair well with the style brief from prompt 22.

22. The voice extractor

Here are three things I have written: [paste them]. Analyse my voice specifically: sentence length pattern, vocabulary level, how I handle humour, punctuation habits, structures I favour, and what I never do. Write it as a style brief under 200 words that I can paste into future prompts.

23. The voice matcher

Here is my style brief: [paste the output of the previous prompt]. Write [the deliverable] in that voice. Then tell me which choices you made specifically to match it, and which parts of my voice you could not reproduce from the brief — those are the gaps I need to describe better next time.

24. The did-it-work check

Here is something I wrote: [paste it]. Here is something you wrote in my voice: [paste it]. Compare them honestly. Where does yours diverge from mine, and is that because my brief was incomplete or because you defaulted to a generic register? Be specific about which sentences.

25. The banned list builder

Read these samples of my writing: [paste them]. Build a banned list for future prompts: words and constructions I never use, plus the ones I overuse and should be warned about. Include the phrases that would immediately signal to someone who knows my writing that I did not write it.

26. The sound-like-me test

Here is a draft written for me: [paste it]. I need to know whether a colleague who reads my writing regularly would notice. Read it as that person. Which sentences would make them pause? What would they say felt off? Do not be reassuring about it.

Four prompts for contexts where sounding human is the entire job — a support reply, a story carrying a point, or the same content in three different voices. For the more playful registers, Grok 4.5 is worth comparing against your usual model.

27. The support reply

Write a customer support reply to this message: [paste the customer's message]. What actually happened and what I can offer: [describe]. Acknowledge the problem specifically rather than generically, say plainly what went wrong if we were at fault, and give the next step. No scripted phrasing, and do not thank them for their patience if they have not been patient.

28. The short personal story

Turn this point into a short story that carries it: [state the point]. The situation I would draw on: [describe it]. Keep it under 150 words, ground it in one specific scene, and let the point emerge from what happens rather than being stated at the end. Ask me for the details you need rather than inventing them.

29. The analogy finder

I need to explain [concept] to [audience]. Give me five analogies from different domains — physical, everyday, social, mechanical, natural. For each, say what it captures and exactly where it misleads. Then tell me which one this audience is most likely to already have intuitions about.

30. The register experiment

Take this content: [paste it] and rewrite it in three voices: someone explaining it to a colleague, someone writing it as a personal essay, and someone answering it as a question on a forum. Same information in all three. Then tell me which suits [my actual context], and what the other two revealed about the content itself.

The voice extractor is the prompt worth saving above all the others. Run it once on your own writing, keep the style brief and your banned list somewhere handy, and paste them into every new chat — so you are not rebuilding your own voice from scratch each session.

Running several models matters more for this task than for most. Models have distinctly different default registers — one hedges, one over-explains, one reaches for enthusiasm — and sending the same voice-matching prompt to two of them shows you which tendencies belong to the model rather than to your brief. The one that needs least correcting is the one to keep for that kind of writing.

Start with the tell detector on something you have already written with AI help and were not quite happy with. Seeing the specific patterns named is what makes them impossible to unsee afterwards. For everything beyond tone, there is a wider set of ChatGPT prompts covering writing, research and work tasks.