Sound like you, every single time

A 60-second setup

A short onboarding turns your topics, audience, goal, and the way you write into a voice profile. It takes about a minute, and it is what lets every other AI feature draft in a way that sounds like you rather than like everyone else's AI output.

Learns from your posts, keeps none of them

To sharpen the profile, the AI reads a sample of your own posts and distills how you write into structured style notes. Those notes are all we keep. The raw posts are never saved, and your timeline is never turned into an archive on our servers.

A profile you review during onboarding

The onboarding flow shows the profile before it becomes the basis for generation. This keeps the setup explicit and avoids silently scraping your timeline in the background.

It copies how you SOUND, never how you post

This is the deliberate split, and it is unusual. The profile records your sound: deadpan one-liners, lowercase openers, terse clipped sentences, no emojis, dry sarcasm. It specifically does NOT copy the way you build a post, your hooks, formats, structures or engagement tactics. The reasoning is blunt: most people arrive here because their account is not growing, so mirroring their structural habits would faithfully reproduce the thing holding them back. It takes the voice and supplies the playbook.

What the profile actually stores

A handful of tight fields, each one injected into every generation. VOICE FACTS, the specific textures of how you sound. A MODE, either amplify (your best posts already show a winning pattern, so lean into it) or establish (too few posts, or engagement too scattered to learn from, so it builds from your stated voice and niche instead). A CONFIDENCE rating, high only when there is a real body of posts to read. PRIORITY FORMATS. A one-sentence DIAGNOSIS of your account's biggest bottleneck. And NEVER RULES, the genres and moves to keep out of your posts.

Your best posts count for more

The sample it reads is ranked by engagement, and higher-performing posts are weighted more heavily, because those are the evidence of what actually lands for you. Where your stated voice and your real posts disagree, the posts win. It also strips em dashes and spaced-dash asides from everything it writes, since those are the clearest tell that a human did not write a post.

The same profile drives your replies

Replies are not a separate voice. The generator draws five replies at a time from your profile, varying them across five distinct moves so you get real options rather than one guess: a counterpoint that disagrees with a sharper angle, an addition from your own first-hand experience, a deadpan joke that fits the thread, a build on their point, and a fair contrarian take that invites discussion. Each reply also matches the language and register of the post you are replying to, and when a post is just an image it reads the text inside the image to get the language right.

Do you store my tweets to learn my voice?

No. The AI reads a sample of your posts to distill structured style notes about how you write, and then the raw posts are never saved. No archive of your timeline is stored on our servers.

Can it suggest post ideas and write replies in my usual voice?

Both, from one profile. A 60-second onboarding reads a sample of your own posts, weights the higher-performing ones, and distills how you sound into structured style notes. Those notes are injected into every generation after that: the write prompts sharpen your own thirty-second answer into three posts in your voice, and the reply generator drafts five replies at a time in the same voice, each taking a different angle and matching the language of the post you are replying to.

How is this different from an AI trained on my tweets?

It deliberately learns only how you sound, not how you construct a post. Hooks, formats and structure come from our own craft instead of being copied back to you, because most accounts that arrive here are not growing, and faithfully mirroring their posting habits would reproduce exactly what is holding them back. You keep your voice; the structure is the part that gets upgraded.

How long does the voice setup take?

About 60 seconds. It captures your topics, audience, goal, and writing style, which is enough for every AI feature to draft in your voice.

Can I continually re-train the voice from Settings?

No. In v5 the voice is distilled during onboarding and there is no ongoing re-tuning editor. This page describes the current product rather than the older Settings workflow.