To check the tone of a message with AI, paste your draft in with one line of context and ask a specific question: “How will this come across to my manager? Does anything sound rude or passive-aggressive?” Then ask for two or three rewrites in the tone you want. AI is good at spotting curt phrasing, missing warmth and accidental blame, and quick at offering softer versions.
On recent phones the built-in tools do quick rewrites too: Apple’s Writing Tools offer Friendly, Professional and Concise versions, and supported Android phones rewrite short messages through Gemini Nano. For messages about a coworker, a family fight or anything else you’d rather not upload, a local AI app does the same job without the text leaving your phone.
AI is a useful second reader. It can’t know your history with the person, so the final call is yours.
How to check a message’s tone with AI #
- Paste the draft. Include a line of context: who it’s to and what you want to happen.
- Ask how it reads. “How will my manager likely read this? Does anything sound defensive or passive-aggressive?”
- Ask for a rewrite. “Rewrite it to sound firm but friendly. Keep it under 80 words.”
- Ask for options, not one answer. Three versions beat accepting the first rewrite.
- Edit it back into your voice. AI rewrites often sound generic. Keep your own phrases where they work.
- Read it once as the recipient before sending.
Prompts that work #
Checking your own message #
| Situation | Prompt |
|---|---|
| Not sure how it sounds | “How would a reader likely feel after reading this? Point to the words that cause it.” |
| Too harsh | “Make this kinder without losing the point.” |
| Too soft | “Make this more direct. Remove the apologies and hedging.” |
| Passive-aggressive | “Is this passive-aggressive? Quote the phrases that make it sound that way.” |
| Too long | “Cut this to three sentences.” |
| Work email | “Make this professional and neutral for a manager I don’t know well.” |
| Saying no | “Help me decline this politely and clearly, without over-explaining.” |
| Apology | “Make this apology sincere and specific, without excuses.” |
| Angry draft | “I’m upset. Rewrite this so it says what I need calmly.” |
Rewriting it #
- “Give me three versions: one warmer, one more direct, one more professional. Keep them the same length.”
- “Make this sound less defensive without apologizing.”
- “Rate the tone from 1 (cold) to 5 (warm) and explain why.”
Reading a message you received #
- “What tone is this message? Is the sender annoyed, or am I overthinking it?”
- “What are two or three reasonable ways to read this?”
- “Suggest a calm reply that doesn’t escalate.”
Asking the AI to point at specific words is the most useful habit of all. “The phrase ‘as I already mentioned’ may read as irritated” tells you what to change and why. Asking for several interpretations of a message you received is the second: it reminds you that tone in text is ambiguous, and it stops the model sounding more certain than it should.
A before-and-after example #
Draft: “As I said last week, the report was due Friday. I need it today. Please let me know why this keeps happening.”
Prompt: “This is to a coworker who’s usually reliable. Make it firm but not accusatory.”
Rewrite: “Hi Sam, just checking on the report that was due Friday. I need it by end of day today for the client meeting. Is anything holding it up that I can help with?”
The rewrite keeps the deadline and the urgency, drops “as I said” and “why this keeps happening,” and gives the other person an easy way to respond.
Built-in tone tools on your phone #
iPhone: with Apple Intelligence on an iPhone 15 Pro or later, select your text, open Writing Tools and choose Friendly, Professional or Concise, or use Proofread. Many requests run on Apple’s on-device model; larger ones may use Apple’s Private Cloud Compute.
Android: Google’s on-device Gemini Nano model supports proofreading and rewriting short messages “in different tones or styles,” according to Google’s developer documentation, on supported phones such as the Pixel 9 and later and the Galaxy S25 and S26 series. Whether you see it depends on which apps and keyboards have added the feature.
These are fast for a quick polish. They don’t explain what sounded off, and you can’t give them your own instructions.
Where does AI misread tone? #
- Sarcasm and irony. “Great, another meeting” reads as positive to a model that misses the context.
- Your relationship. “Fine.” from your best friend and from your boss mean different things. Blunt messages between close colleagues are often fine.
- Culture and workplace norms. Directness that’s normal in one office or country reads as rude in another. Say so in the prompt.
- Emoji and punctuation. A period at the end of a short text, a thumbs-up, or no emoji at all carries meaning that models only partly pick up.
- Short messages. “Ok.” has too little text to judge.
- Your voice. Rewrites flatten personality into generic politeness, and small on-device models do this more.
- Bigger problems. If a conversation keeps going badly, a better sentence won’t fix it.
Give the model context to work with: “This is to a coworker I’ve worked with for five years; we’re usually informal.”
Why check tone privately? #
The messages people most want a second opinion on are the sensitive ones: a note to HR, a reply to an ex, feedback to a direct report, a tense family text. Pasting a message you received also shares someone else’s words, and sometimes their name, without their knowledge.
With a cloud chatbot that text is stored on the provider’s servers, where it may be reviewed or used for training depending on your settings. Our breakdown of whether humans read your AI chats covers what each service does.
A local AI app runs the model on your phone instead. Personal LLM is one option: free on iOS and Android, no account, and your chats stay on the phone. Three of its features suit this job:
- A system prompt per chat, so you can keep a dedicated “tone editor” chat that always behaves the same way.
- Presets you can switch mid-chat. Precise for the analysis, Creative when you want genuinely different rewrites.
- One-tap copy of the whole reply or just the sentence you need.
A system prompt for a tone-editor chat:
You are an editor who helps me get the tone of messages right. For each draft, first say in one or two sentences how it’s likely to come across and which words cause that. Then give one rewrite in the tone I ask for. Keep my meaning and roughly my length. Don’t add details I didn’t include.
More templates are in our system prompt examples.
How good are small offline models at tone? #
Tone and rewriting are among the tasks small models handle best. They rely on everyday language patterns, which even 3B to 4B models learn well, rather than on rare facts. A 4B model on a recent phone reliably flags curt phrasing, suggests warmer alternatives and adjusts formality. A 9B model is better at subtle cases and at explaining why something reads a certain way.
Two settings make a difference:
- Lower the temperature for analysis (around 0.3) so the reading stays consistent between runs.
- Raise it for rewrites (around 1.0 or more) so the three options are actually different.
Our guide to getting better answers from a local AI explains what each setting does.
If what you want isn’t your own message but a batch of other people’s text, such as reviews, survey answers or support emails, that’s a different job with different prompts. See AI sentiment analysis without uploading the data.
Frequently asked questions #
Can AI tell if my message sounds rude? #
It gives you a good read on how most people would take the words, and can point to the phrases causing it. It doesn’t know your relationship or history with the reader, so treat its reading as one opinion.
Is there a free AI tone checker? #
Yes. Apple Writing Tools and Gemini Nano’s rewriting are built into supported phones at no cost, and local AI apps like Personal LLM are free. Cloud chatbots also do this on their free tiers, but your message goes to their servers.
Is it safe to paste private messages into ChatGPT? #
They’re stored on OpenAI’s servers and may be used for training unless you turn that off. For messages about other people or sensitive situations, use a temporary chat or an on-device AI app that keeps the text on your phone.
How do I make an email sound more professional with AI? #
Paste it and ask: “Make this professional and neutral, keep it short, and keep all the facts.” Then check that the rewrite didn’t change any details or promise anything you didn’t mean.
Can AI detect sarcasm? #
Sometimes. Models catch obvious sarcasm, especially with context, but they often miss dry sarcasm and sarcasm between people who know each other well. If the tone depends on shared history, trust your own reading.
Can AI rewrite my message in another language with the right tone? #
Yes, most models translate and adjust tone together: “Translate this into Spanish for a formal business email.” Have a fluent speaker check anything important, since politeness conventions vary a lot between languages.