How to use ChatGPT to reply to texts without sounding like ChatGPT

To use ChatGPT to reply to texts without sounding like ChatGPT, give it three things before you ask: three to five messages you’ve actually sent (your style sample), the last few messages of the conversation, and who the person is to you and what you want to say. Then ask for two or three short options, pick one, and rewrite it in your own words before you send. The style sample is what stops it from sounding like a polite customer-service agent.

Below: why default ChatGPT replies sound fake, the five-step method, a copy-paste prompt kit, before-and-after examples, and the privacy and dating-app caveats to know first. The same prompts work in Claude or Gemini.

Why do ChatGPT replies sound like ChatGPT?

With no examples of how you write, ChatGPT falls back on its default voice: friendly, complete and tidy. That voice has tells your friends will notice:

• Greetings and sign-offs you never use (“Hey Sam”, “Talk soon”)
• Full sentences and perfect punctuation when you normally type in lowercase
• Filler: “I hope you’re doing well”, “Thanks so much for reaching out”
• Over-enthusiasm, and long dashes where you’d use a comma or nothing
• Answering every point in order, in three sentences where you’d send one
• Emoji in places you’d never put them

None of these are wrong. They just aren’t you. The fix is to show it how you write.

The method in five steps

1. Build your style sample

Open a recent chat with someone similar (a friend for friends, a colleague for work) and copy three to five messages you sent. Pick ones that show your habits: how long they are, whether you use capitals, your usual slang and emoji, how you say yes and no.

Even better, turn them into a style card once and reuse it. Paste 15 to 20 of your messages into ChatGPT with the “describe my style” prompt below, check the result, and keep it in a note.

Example style card, made up for this guide:

Lowercase, no full stops at the end. Usually under 12 words. Says “haha” not “lol”. Uses “tbh” and “tmrw”. One emoji at most, usually none. Never says “Hey [name]”. Asks a question back when making plans.

2. Paste the thread, minus the private bits

Paste the last three to six messages so ChatGPT knows what was actually asked. Replace names with initials and remove phone numbers, addresses and anything sensitive. You’re sending someone else’s words to OpenAI.

3. Say who they are and what you want to say

“My sister, we’re close” or “a client I’ve met twice”. Then the substance: yes or no, the time that works, the thing you’re apologizing for. ChatGPT can match your style; it can’t know your plans.

4. Ask for short options

Two or three options, each under a word limit you’d really send (15 to 25 words for most texts). Short options are easier to edit and sound less written.

5. Edit, then send it yourself

Read the draft out loud. Delete any detail you didn’t give it, because AI invents plausible ones. Add the specific thing only you know. Then send it from your own phone or Mac, as you.

The prompt kit

Copy these, fill in the brackets, and paste them into ChatGPT.

The everyday reply

Here is how I text. Match this style exactly: same length, capitals, punctuation, slang and emoji habits.
[paste 3-5 of your messages, or your style card]

Conversation so far (most recent last):
[paste the last few messages, names replaced with initials]

They are: [who they are to you]
I want to say: [your answer in plain words]

Give me 3 options, each under 20 words. No greetings or sign-offs unless my samples use them.

Describe my style (do this once)

Here are 20 messages I've sent to friends. Describe my texting style in 5 short bullet points I can reuse in future prompts: typical length, capitalization, punctuation, slang, emoji, and anything I never do.
[paste messages]

Saying no kindly

[style card]
They asked: [their message]
I need to say no because [reason you're comfortable sharing]. Keep it warm, don't over-apologize, and suggest [alternative, if any]. 2 options under 20 words.

Replying late

[style card]
I didn't reply to this for [how long]: [their message]
Write a reply that owns the delay in one short phrase, then answers properly. No long apology. 2 options.

Group chat

[style card]
Group chat with [who]. Last messages: [paste]
I want to [agree / suggest a time / bow out]. One message, under 15 words, casual.

Work-ish message on WhatsApp or iMessage

[style card]
This is a colleague, not a close friend. Their message: [paste]
I want to say: [answer]. Friendly and brief, a bit more complete than my usual texts, still no formal greeting.

Fix a draft that sounds off

This sounds too formal and too long compared with my style card: [style card]
Draft: [paste]
Rewrite it shorter and closer to my style. Keep the facts, add nothing new.

Before and after

All examples are made up for this guide.

Catching up with a friend. They wrote: “hiii long time, how’ve you been? we should catch up soon”

	Reply
Default ChatGPT	“Hey, it’s so great to hear from you. I’ve been doing well, thanks for asking. Things have been busy with work, but I’d love to catch up soon. Let me know when you’re free.”
With your style card	“omg hiii. been good, work’s been a lot. yes let’s, free thurs or sun?”
After your edit	“omg hiii. good, work’s been a lot but surviving. yes pls, thurs or sun?”

Saying no to a colleague. They wrote: “any chance you can cover my shift saturday?”

	Reply
Default ChatGPT	“Hi Sam, unfortunately I won’t be able to cover your shift on Saturday as I have prior commitments. I hope you’re able to find someone else.”
With your style card	“ah sorry can’t sat, got family stuff. ask p? she mentioned wanting more hours”
After your edit	“ah sorry can’t sat, family thing. hope you find someone”

Notice what happened in the second example: the AI invented “she mentioned wanting more hours”. You never said that, so it goes. This is the most important edit you’ll make.

The privacy caveats

• You’re sharing other people’s words. Your friend didn’t agree to have their messages sent to an AI company. Paste only what’s needed, with names and identifying details removed.
• Never paste sensitive details: health, money, addresses, passwords, anything you’d regret seeing in a data leak.
• Check your ChatGPT data settings before you make this a daily habit, so you know whether your conversations are used to improve OpenAI’s models.

Our checklist on whether it’s safe to let AI read your messages compares how different tools handle this.

Using ChatGPT for dating app messages

ChatGPT can be a good wingman: ideas for an opener based on what’s in someone’s profile, a gut check on a reply you’ve written, a way out of a conversation that’s going nowhere. It shouldn’t write your messages for you, and the apps agree.

• Hinge’s terms ban content that wasn’t written by you or was automatically generated, unless Hinge authorized it, and both Tinder’s terms (effective 5 March 2026) and Hinge’s ban third-party apps or services that directly interact with the app without written consent, “including but not limited to artificial intelligence or machine learning systems” (Tinder, Hinge).
• Hinge’s own advice is to use AI for a gut check or ideas and write in your own voice, as a wingman, not a ghostwriter (Hinge).

So use ChatGPT for ideas, then write the message yourself, and never use it to misrepresent who you are. More in what Tinder, Hinge and Bumble’s terms say about AI.

When copying and pasting gets old

The method above works anywhere, but pasting the same style card twenty times a day gets tedious. Two ways to skip it:

• iMessage on a Mac: ChatGPT’s Messages plugin can read your iMessage threads itself and draft replies you approve, on any ChatGPT plan, in the app’s Codex and ChatGPT Work modes on Apple silicon (OpenAI). Setup is in how to use ChatGPT with iMessage on a Mac.
• WhatsApp, Telegram, Instagram and iMessage together, on a Mac: Tera Chat can use the ChatGPT account you already have, through OpenAI’s Codex, for free. It learns your texting style from your own messages, so there’s no style card to paste, lets you set a different tone for each tag of contacts, and drafts in the thread you’re reading. You still edit and send every message yourself. It’s Mac-only and needs Apple silicon and macOS 15 or later.

For how these tools compare, see AI that replies to texts in your style, and for more on style samples, how to make AI replies sound like you.

What to do next

Write your style card today: paste 20 of your messages into the “describe my style” prompt and save the result. Next time a text sits unanswered because you don’t know how to word it, paste the card, the thread and what you want to say, pick the option closest to you, and fix it before you hit send.