The email you send angry costs you more than time. Ten seconds of AI reading your draft is cheaper than the apology you will owe tomorrow.
| Check time | About 15 seconds per message |
|---|---|
| Tone accuracy gap | Senders expected 78% accuracy conveying tone by email, recipients hit about 56% |
| Best for | High-stakes messages: bosses, clients, direct reports, anything written while annoyed |
| Not for | Legal matters, formal complaints, resignations – use human judgment instead |
1. Write one standing prompt
Open your notes app and write this once: ‘Read this message the way a tired, busy person would read it. Tell me if it sounds harsher than I mean it to. Tell me if there is a request buried in here that could get missed. Tell me if anything reads as blame, even if I do not mean it that way. Then fix only those parts. Leave everything else exactly as I wrote it.’ Save it somewhere you open ten times a day, a pinned note or a text file.
2. Paste it before anything with stakes
Before you send an email to your boss, feedback to someone who reports to you, or a reply typed while annoyed, copy your draft and the standing prompt into any AI chat. This takes fifteen seconds and costs nothing. Do it especially the moment you notice you are typing fast.
3. Read the flags, not a rewrite
The AI should point at specific lines, not hand you a whole new message. That is the point, your voice stays yours. If it flags a line as sounding like blame, look at that one line and decide if you agree. You decide, the AI is just a second pair of eyes that is not emotionally involved right now.
4. Make it a rule for the moments that matter
Pick one trigger and commit to it: any message you write while annoyed, tired, or rushed goes through this check first, no exceptions. A rule you follow every time beats a habit you remember sometimes. Put a sticky note on your screen for the first week if that is what it takes.
5. Calibrate it on a message you already regret
Find an old email or text you wish you had worded differently and run it through the prompt now. See what it flags and whether you agree. This shows you how the check thinks, so you trust it the first time it matters for real, not the tenth.
A worked example
A manager is about to reply in a team Slack channel after a deadline slipped for the second time.
Before
This is the second time this has happened. I don’t understand how the deadline was missed again. We need to fix this immediately, this is not okay.
After
Flag 1: ‘second time this has happened’ names the team publicly, this could feel like a callout in a group chat. Flag 2: ‘I don’t understand how’ reads as an accusation, not a question. Flag 3: ‘this is not okay’ has no next step, the team will not know what to do differently. Suggested lines: ‘This deadline slipped again, and I want to understand what got in the way so it does not repeat. Can we grab 15 minutes today to look at it and agree on a buffer for next time.’
Three ways to bend it
For managers giving feedback
Change the question to: does this sound like coaching or does it sound like blame? Use it before any written feedback to a direct report, especially the kind you are tempted to send instead of say out loud.
For job seekers replying to recruiters
After a rejection or a low offer, paste your reply before sending and ask: does this sound bitter or does it sound professional? A calm reply keeps the door open for the next role at the same company.
For customer facing roles
Before replying to an unhappy client, ask the AI to flag anything that sounds defensive rather than helpful. Customer replies live longer than you think, they get forwarded and screenshotted.
When it goes wrong
The AI keeps saying the message looks fine, even when I know it is harsh.
Tell it to assume the worst case reading. Add one line to the prompt: ‘Assume the reader is already annoyed with me, what in here would make that worse.’ A skeptical instruction gets a skeptical answer.
It rewrites my whole message and it does not sound like me anymore.
Remind it in the prompt to fix only the flagged parts and leave the rest untouched, word for word. If it still rewrites everything, ask it to show you just the flags first, then decide line by line yourself.
I don’t have time to do this for every message.
You don’t need to. Save it for messages with real stakes, a boss, a client, a direct report, anything typed while annoyed. Routine chats with a teammate you trust don’t need it.
When not to use this
This is not for every message, and it is not a substitute for a real conversation. If something is genuinely serious, a conflict with legal weight, a formal complaint, a resignation, get advice from a person you trust, not just a second read from AI. It also will not fix a relationship that needs an actual conversation, it only stops you from making one bad moment worse in writing.
What you can now do
You get to send the message that says what you mean, without the line that turns a work problem into a personal one.
Part of Making Life Easier With AI, a weekly walk-through of one small thing AI can take off your plate.
Take this further
Act as a blunt hiring manager who has read ten thousand resumes, not a career coach. I will paste my full resume and the job description I want. Rewrite the whole resume for that role, section by section, in this order: summary, experience, skills, education. Rules: every experience line leads with impact, not duty. Use bracketed placeholders like [8 percent] for any number I did not give you, and list at the end every placeholder I need to replace with a real figure. Keep it to one page of text. Plain formatting only, no tables or columns, so screening software can parse it. After the rewrite, tell me the three weakest claims that need evidence before I send this anywhere. My resume: [paste resume]. The role: [paste job description].
Takeaways
- Write your standing tone-check prompt once and pin it somewhere you open multiple times a day, so it takes seconds to use.
- Run any message typed while annoyed, tired, or rushed through the check before sending, especially to a boss, client, or direct report.
- Test the prompt on an old email or text you regret first, so you know how it flags things before you trust it for something real.
- If it keeps calling harsh messages fine, add a line telling it to assume the reader is already annoyed and flag what would make that worse.
- For anything with legal weight, a formal complaint, or a resignation, talk to a person instead of relying on an AI read.
Questions people actually ask
How do I get AI to check the tone of my email before I send it?
Write a standing prompt once and save it somewhere you open often, like a pinned note. Before sending anything with stakes, paste your draft plus that prompt into any AI chat. Ask it to flag lines that sound harsher than intended, missed requests, or blame, then fix only those parts and leave the rest as is.
What should I put in the prompt so AI doesn’t rewrite my whole message?
Tell it explicitly to fix only the parts it flags and leave everything else exactly as written, word for word. If it still rewrites the whole thing, ask it to list the flagged lines first without touching anything, then decide line by line yourself whether to change them. That keeps your voice intact.
Why does AI keep saying my message sounds fine when I know it’s harsh?
The AI may be reading your draft too generously. Add a line telling it to assume the worst case reading, for example, assume the reader is already annoyed with you and ask what in the message would make that worse. A more skeptical instruction produces a more skeptical, useful answer.
Is it actually true that people misjudge tone in emails?
Yes. Research by Justin Kruger and Nicholas Epley found senders expected to convey their intended tone correctly about 78 percent of the time, but recipients only understood it correctly around 56 percent of the time, barely better than guessing. The gap comes from egocentrism, since writers hear their own intended tone in their head.
When should I not rely on an AI tone check?
Skip it for routine chats with people you trust, and don’t lean on it for anything with real weight, like a formal complaint, a legal matter, or a resignation. Those need a real conversation or advice from a person, not a second read from AI. It also can’t repair a relationship that needs an honest talk.
Sources
- Egocentrism Over E-Mail: Can We Communicate as Well as We Think?NYU Stern School of Business
- E-mails and egosAmerican Psychological Association
- Writing Tone Detector and Tone SuggestionsGrammarly
- Prompting best practicesAnthropic
What happens next
Tone-checking prompts like this will likely get built directly into email and chat clients, similar to how Grammarly already flags tone in Gmail and Slack, rather than staying a separate AI chat step. Watch for AI assistants that read a whole thread’s history before flagging a reply, since context across a conversation matters more than a single message in isolation. The open question is whether people keep the habit of checking messages typed while emotional, since even well-known research on email misunderstanding hasn’t made most people slow down before hitting send.
