AI to write messages in English

Grais reads the conversation open in your desktop browser — WhatsApp Web, Telegram Web, Instagram, LinkedIn, or Tinder — and writes the reply there. It reads and writes in more than thirty languages, and the useful part is that it takes its cue from the thread: how formal the other person is being, how long their messages are, whether they use contractions, whether anyone is using full stops.

That is a different job from correction. A correction tool takes what you already wrote and makes it grammatical, which leaves the register wherever your instincts put it. If your instinct in a second language is to be more formal than the situation — and that is the common failure, not grammar — a clean sentence that is two notches too polite is still the wrong message. Grais drafts against the conversation rather than against a style guide, so the target is the thread, not correctness in the abstract.

• It reads the incoming message too. Tone and intent are summarised with the lines they came from, which is the part that is genuinely harder in a language you learned later.
• It matches how you write. Drafts follow your own past messages to that contact, so the voice stays yours across languages rather than becoming a textbook voice.
• Review is the default. Every contact starts at the review tier, so you read the draft in full unless you later give that known contact more autonomy.

People searching that phrase are usually not asking for spell-check on a phone. They are asking for something that stops their messages reading wrong — too stiff, too abrupt, too eager — in a language they did not grow up writing casually. Correction tools are good at grammar and spelling and are worth using for that. They are working on the sentence you produced, which means they cannot tell you the sentence was the wrong temperature for this thread.

Register in messaging is unwritten and moves fast. A colleague who wrote "sounds good!!" is not asking for "Thank you for your message. I would be happy to proceed." Both are correct English. Only one belongs in that thread, and knowing which is a thing you absorb from years of casual exposure rather than from study.

Because Grais drafts from the conversation, the register question is answered by the evidence in front of it. It also holds register across languages: switching from your first language to English in the middle of a thread does not have to mean becoming a more formal person for the rest of it.

Writing is only half of it. Understatement, hedging, and politeness formulas are where a second language stays hard longest, because the literal meaning and the intended meaning drift apart and nothing on the surface tells you by how much. "Interesting, let me think about it" is not a request for time in every workplace, and native speakers rarely notice they are encoding anything at all.

Grais summarises what an incoming message appears to be asking for from the visible context. Treat that reading as something to check against the thread rather than something you have to accept. When you disagree with it, that is useful information: you can identify the ambiguity and rewrite the draft to address it directly instead of guessing.