What Makes AI Replies Actually Sound Like You

You’ve probably tried asking ChatGPT to write a message for you. The result is usually polite, grammatically perfect, and immediately identifiable as not written by a human. It starts with “I hope this message finds you well” or uses words like “certainly” and “furthermore” — words you’ve never typed in a WhatsApp message in your life.

This is the fundamental problem with AI-generated messages: they sound like AI, not like you. And in personal messaging, that gap isn’t just noticeable — it’s a dealbreaker. Nobody wants to receive a reply that feels like it was written by a customer service bot.

So how do you get AI to actually sound like a real person? Specifically, like you?

Why Generic AI Fails at Personal Messages

Large language models are trained on vast amounts of text, and they’ve learned to produce output that’s statistically likely — which means generic. When you ask an LLM to “reply to this message,” it defaults to a register that’s:

Overly formal — complete sentences, proper punctuation, no contractions
Universally polite — hedging phrases, qualifiers, diplomatic tone
Context-blind — the same style regardless of whether you’re talking to your boss, your best friend, or your mum
Emotionally flat — no personality, no humor, no warmth

This isn’t a bug — it’s what these models are optimized for. They’re designed to be helpful to everyone, which means they sound like no one in particular.

Personal messaging is the opposite of generic. It’s deeply contextual. The way you text your partner is different from how you text a colleague. You might use lowercase with friends and proper capitalization with your parents. You might use “haha” instead of “lol,” or no laugh indicators at all. These micro-patterns are your voice, and they’re invisible to a generic AI prompt.

How Tone Matching Works in Nudge

Nudge doesn’t use a generic “write a reply” prompt. It analyzes your recent messages in a conversation to understand how you communicate with that specific person, then instructs the AI to match your style.

Here’s what we look at:

Message Length

Some people write novels. Others reply in three words. The length of your typical messages in a conversation is one of the strongest style signals. If your last ten messages to someone averaged eight words, Nudge won’t generate a three-paragraph reply.

Punctuation Patterns

Do you use periods at the end of messages? Most people under 40 don’t — in casual WhatsApp conversations, a period can actually read as curt or passive-aggressive. Nudge picks up on whether you end messages with periods, use exclamation marks, employ ellipses, or just let sentences trail off without punctuation.

Emoji and Expression

Some people punctuate every message with emoji. Others haven’t used one since 2019. Some use “haha” generously; others are more restrained. These expression patterns are personal and consistent, and they’re immediately obvious when an AI gets them wrong.

Nudge detects your emoji frequency, the specific emoji you tend to use, and whether you use text-based expressions like “haha,” “lol,” or “lmao.” If you’re a no-emoji person, your drafts won’t suddenly sprout smiley faces.

Vocabulary and Register

The words you choose define your voice. Do you say “sounds good” or “perfect”? “Can’t make it” or “I won’t be able to attend”? “Yeah” or “Yes”? These seem like tiny differences, but they’re the difference between a message that reads as yours and one that reads as someone else’s.

Nudge identifies the vocabulary register of your recent messages — casual, semi-formal, formal — and matches it. The same person might text differently across conversations, and the AI adapts to each one.

Conversation Rhythm

Beyond individual message style, Nudge considers the rhythm of the conversation. Is this a rapid back-and-forth exchange or a slower thread where messages arrive hours apart? Are you initiating or responding? Is the other person asking questions or making statements? The draft is shaped to fit the natural flow of the specific conversation.

AI drafts that sound like you, not like a robot. Nudge matches your tone, length, and style for every conversation.

Join the waitlist

Before and After

To make this concrete, here’s what the difference looks like.

Incoming message: “Hey! Are we still on for dinner Friday? Want to try that new Thai place?”

Generic AI reply:

“Hi! Yes, I would love to join for dinner on Friday. The new Thai place sounds great. Looking forward to it!”

Nudge draft (for someone who texts casually):

“yeah definitely, thai place sounds good. what time were you thinking?”

Nudge draft (for someone who’s more expressive):

“Yes!! I’ve been wanting to try that place. Friday works, what time?”

Same incoming message, same user, different conversations — different style. The generic version is fine. The Nudge versions are yours.

What We Don’t Do

It’s worth being clear about the boundaries:

We don’t learn a permanent model of your voice. Each draft uses the recent messages in that conversation as context. There’s no persistent profile of your writing style stored anywhere.
We don’t send drafts automatically. Every draft is presented for your review. You can edit it, rewrite it, or throw it away. Nudge drafts. You decide.
We don’t read messages in the background. Draft generation only happens when you request it for a specific conversation.

The AI is a starting point, not a replacement. Most people edit their drafts before sending — shortening them, adding a detail, adjusting the tone. That’s the intended workflow. The value isn’t that the AI writes the perfect message. It’s that you don’t have to start from a blank screen. (For more on why starting is the hardest part — and the psychology behind it — see why you’re bad at replying.)

Why Starting Is the Hard Part

If you’ve ever stared at a message for ten minutes trying to figure out what to say, you already know this intuitively. The blank reply field is the enemy. Once you have something — even something imperfect — you can work with it. You can react to it, improve it, make it yours.

This is a well-documented cognitive phenomenon. In creative and communication tasks, the hardest part is almost always initiation. The first draft doesn’t need to be good. It needs to exist. Once it does, your brain switches from generation mode (hard, effortful) to editing mode (easier, more intuitive).

Nudge generates that first draft. You turn it into the real thing. The whole process takes under thirty seconds, and the result is a message that sounds like you wrote it — because, in the ways that matter, you did.

The Technical Reality

Behind the scenes, Nudge sends up to 20 recent messages from the conversation to our AI proxy, which forwards them to Claude (by Anthropic) with instructions to match your style. The draft comes back in seconds. The message context is discarded immediately — we don’t store your conversations on our servers.

For the privacy-conscious: this is the one moment where data briefly leaves your device. We’ve written a detailed breakdown of the privacy architecture if you want the full picture.

If you’re tired of AI messages that sound like they were written by a committee, join the waitlist. Nudge is currently in private beta, and we’re building it for people who want AI that adapts to them — not the other way around.