Tone checker for text messages
Subtext is an AI tone checker for text messages, emails and anything else you write to another person. Paste your draft, or the thread you are replying to, and it shows how the message could read before you send it: the exact words that could land as harsh, cold or passive-aggressive, and why. Then it gives you three versions to choose from, each flagged safe to send or worth another look. It never sends anything for you.
On iPhone, Android and in your browser. Nothing you write trains an AI model.
Can a tone checker tell you how your message will land?
Partly, and the honest version of "partly" is the reason to use one. Subtext reads the words in front of it and shows how they could come across: sharp, cold, passive, over-apologetic, sarcastic when you did not mean it. What it cannot do is know the person you are writing to or what happened between you last week. So every flag says "could read as", never "is", and every version it offers is one you can edit or ignore. The piece on what AI can and cannot tell from a message's tone goes through the evidence behind that line.
Why texts get misread in the first place
Text drops most of what you would use to read tone in person: voice, face, timing. What is left is wording, punctuation and whatever the reader brings with them. In a series of email experiments, writers expected their sarcasm, anger or sadness to be recognised far more often than it was, and knowing the reader well did not help (Egocentrism over e-mail: Can we communicate as well as we think?). Most everyday messages between people who know each other are still read correctly, so the trouble sits with a minority: the ambiguous one, the sarcastic one, the one you have rewritten four times (People correctly interpret the tone of text messages and emails). Small things carry weight there. A full stop after a one-word reply reads as less sincere, and only in texts, as the piece on the period in text messages explains.
What Subtext flags, and how it says it
Subtext marks the exact words doing the damage and names the feeling underneath them, from a vocabulary of 51 emotions and 48 kinds of trigger. Each tag explains itself: "quite direct, little softening" on a bare request with no hedge or thanks; "more than one plausible reading" on a line that could be a joke or a dig; "few warmth cues" on a reply with no acknowledgement and no sign-off; "could be read as indirectly critical" on neutral wording with an edge underneath. What you will not see is a percentage or a verdict like "hostile" stated as fact. Those are the labels human readers agree on least, so a tool that states them with confidence is bluffing. The piece Does this text sound rude? shows the same reading on real drafts.
Three versions, then your call
After the read, Subtext offers three versions, one tap apart: one close to what you wrote, one warmer, one shorter and clearer. Each carries a safe-to-send badge or a check-before-sending note. Treat that as a second reading rather than a verdict. A warning means look again at the marked words, and a green badge does not mean the other person will be fine. Tap warmer, shorter or add confidence to move any version, and it tells you how much of your wording it changed, so you can see whether you are looking at a polish or a replacement. Manual edit lets you change a word yourself before you copy it.
When it asks before it drafts
You should not have to draft some messages from one line. If you ask Subtext how to write something, give it a thin brief, or hand it a boundary, refusal or confrontation with no word on who it is for, it asks first: who the message is going to, what you want to happen, how you want to come across. Then it drafts from your answers. For everyday messages it does not interrupt; it reads how close you and the other person are from the tone of what you wrote and keeps that in the versions. The received-message page covers the other direction, working out what they meant before you reply, and the page on a tool that reads the whole message thread before suggesting a reply explains why the thread matters.
What a tone checker cannot know
Three things, and it is better you hear them from us. It cannot see your relationship. The same seven words from your manager, a new client and your best friend carry three different weights, and two of those three are invisible in the draft. It cannot read sarcasm reliably. Models that score near-perfectly on synthetic sarcasm drop to close to a coin flip on real human writing (LLMs in Sarcasm Detection? It's elementary! (Or is it?)), and people struggle with it in text too. And it cannot predict how one specific person will react. Nothing in the research we reviewed validates that for any tool, so we will not claim it. Readers also take tone differently across cultures, and one study found that AI suggestions pull writers toward one style, which the piece on whether AI makes your texts sound like a robot takes apart.
How it compares with the tone tools you already have
Checked 22 September 2026 against each vendor's own documentation. Grammarly's tone detector labels the tone of your own text and suggests rewrites; it does not read the thread you are replying to. Apple Writing Tools rewrite in three registers, Friendly, Professional and Concise, without saying what was wrong. Gemini in Gmail uses your email context and offers friendlier, more formal or shorter versions, for email. ChatGPT reads a pasted thread and will tag or rewrite if you ask it to. Subtext reads the whole exchange, including screenshots and voice notes, marks the words and explains each one, and gives three versions with a safe-to-send note. This is not a scoring table, and Grammarly is very good at what it labels. The difference is context. The longer comparisons are in the best AI tone checkers, Subtext vs Apple, Google, Grammarly and ChatGPT and Subtext vs Grammarly for personal messages and tone checking.
What happens to what you paste
You are handing over something personal, often someone else's words too. Your messages go to the AI providers that read and write them, and to our servers so you can come back to the conversation; the privacy policy names every provider. Nothing you send is used to train an AI model. Delete a conversation yourself and it is gone straight away, attachments included. Subtext is an app, not a keyboard, so it never sits between you and what you type; it only sees what you hand it. Whatever tool you use, check its training setting before you paste someone else's messages.
FAQ
Is there an app that checks the tone of a text before I send it?
Yes. Subtext reads your draft, or a screenshot of the whole thread, and marks the words that could land as harsh, cold or passive-aggressive before you send anything. It explains each flag and offers three versions, and you copy the one you choose.
What is the best app for checking the tone of a text message before you send it?
The one that shows its reasoning. A label like "aggressive" with no explanation tells you nothing you can act on. Look for a tool that marks the exact words, says why, reads the conversation you are replying to as well as your line, and lets you keep your own wording. We built Subtext to those four criteria, and no independent evaluation of any tone checker exists yet, ours included.
Is there an app that tells you if your message sounds harsh or could be misread?
Subtext does exactly that, with one limit. It can tell you a line "could read as" harsh and show you the words behind that reading. It cannot tell you how a specific person will feel, because their history with you is not in the text.
Is there a tone analyser app for iPhone that works with text messages and emails?
Subtext runs on iPhone, Android and in the browser. Paste a text, an email, a Slack message or a screenshot of the chat, and copy the version you pick back into the app you were in. It is not a keyboard, so it works with anything you can copy from.
I always come across harsher than I mean to over text. What helps?
Usually the fix is small: a bare request with no softening, a one-word reply with a full stop, a joke with no cue that it is one. Subtext points at the exact words, which is faster than rereading the whole message for the fifth time (more on that loop in how to stop overthinking every message), and its warmer version shows what one or two changes do without rewriting you.
Can it tell if my text sounds passive-aggressive?
It can flag wording that readers often take that way, and tell you which words. Whether it is passive-aggressive depends on you and the reader, so the accurate claim is "could read as", and that is the claim Subtext makes.
Does it understand sarcasm?
Sometimes. Sarcasm is a known weak spot for every model and for most humans reading text. If your message relies on the sarcasm landing, add the cue or drop the sarcasm.
Will it still sound like me?
Yes, that is what it is for. It starts from your draft instead of writing from nothing, keeps your phrasing, and tells you how much it changed. Subtext never sends anything for you. You copy the version you pick and send it yourself.
What does it cost?
Subtext gives you a few free analyses to start, then a paid subscription with weekly, monthly and yearly plans, billed through the App Store or Google Play on your phone, or through Stripe on the web. You see the price in your currency before you subscribe.
Is this therapy?
No. Subtext is a writing and communication assistant. It does not provide therapy, diagnosis or medical advice, and its suggestions are not always right, so read anything before you send it.
Subtext works in 17+ languages and answers in the one you write in. On your phone or in your browser.
Right where you need it
Subtext is built for your phone first. Your account works on the web once you have it.
Already downloaded? Your account works at web.subtext.it, synced with your phone.
Or scan to install. Click a code to enlarge it.
Works in the browser you are already using.
- Safari
- Edge
- Chrome
- Firefox
- Opera
- Brave