The anxious re-read before sending a message, gone in ten seconds. Once it’s set up, every risky message gets a free second opinion before it leaves your phone.
| Setup Time | One saved instruction, created once |
|---|---|
| Check Time | About 10 seconds per message |
| Best Platforms | Claude Projects or ChatGPT Custom GPTs |
| Not Recommended For | Resignation letters or formal complaints |
1. Set up one saved instruction
Open Claude, ChatGPT, or whatever AI you already use. Create a saved prompt or a project (in ChatGPT: a Custom GPT or Project, in Claude: a Project) with one job only: check tone before you send. Paste this in as the instruction: “When I paste a message, tell me in one line how the reader will likely feel receiving it (relieved, annoyed, confused, defensive, fine). Then, only if it needs a fix, give me one rewritten version that keeps my meaning but softens or sharpens the tone. No extra commentary.” Save it. You now have a tool, not a one-time chat.
2. Use it before anything sensitive
Before you send a message to your boss, a client, HR, or anyone where the stakes feel higher than a coworker chat, paste the draft in. It takes ten seconds. Try it on something you’ve already written, like a follow-up chasing a delayed vendor, and see what comes back before you send the real one.
3. Read the flag, decide yourself
The AI is a second pair of eyes, not the final word. If it flags your message as sounding defensive and you meant it that way on purpose, ignore the suggestion and send your original. You know the relationship and the history, it doesn’t.
4. Correct it over the first week
Note any message where its read was wrong, maybe it missed that your manager likes bluntness, or that a client responds well to humor. Add one line to your saved instruction, like “my manager prefers direct messages, do not soften those.” It gets sharper every time you correct it.
5. Reuse it everywhere
The same saved prompt works for WhatsApp messages to your team, LinkedIn comments, performance review responses, even a tricky family group chat. Ten seconds before send, every time it actually matters.
A worked example
A team lead in Dubai needs to chase a vendor who has missed a deadline for the third time, without sounding like she’s escalating unnecessarily.
Before
Per my last email, this is now the third delay. Please confirm the new timeline immediately as this is affecting our launch.
After
Read: Will likely land as frustrated and confrontational, may put the vendor on the defensive and slow the reply down further. Rewrite: “Hi Ahmed, following up again on the timeline, this is our third slip so I want to flag it clearly. Can we lock a firm date today so I can plan around it? Happy to jump on a call if that’s faster.”
Three ways to bend it
For managers writing feedback
Paste a draft performance review comment before sending it to HR or the employee. Ask it to flag anything that could read as vague, unfair, or exposed if it gets read back months later in a dispute.
For job seekers writing to recruiters
Before sending a follow-up or a salary negotiation email, ask it specifically: does this sound entitled, or confident and reasonable. Recruiters read hundreds of these, tone is often the only thing separating a reply from silence.
For anyone messaging across cultures
If you work with colleagues or clients from another country, add a line to your instruction: flag anything that could read as too direct in a culture that prefers indirect language, or too vague in one that expects directness. Useful across Gulf, Indian, and Western teams inside the same company.
When it goes wrong
It keeps softening messages I actually want to keep firm.
Tell it explicitly in your saved instruction that you want directness preserved, and to only flag real rudeness, not assertiveness. AI defaults to polite unless told otherwise.
The feedback feels generic and doesn’t really help.
Give it more context in the instruction: who you usually write to, and what has gone wrong before. A specific job gets a sharper read than a vague one.
When not to use this
This will not catch everything, and it does not know office politics, personal history, or sarcasm that only makes sense between two people who know each other. Do not rely on it alone for anything genuinely high-stakes, like a resignation letter or a formal complaint, get a real person’s read too, ideally someone who knows who you’re writing to. It’s a second opinion, not a replacement for judgment.
What you can now do
You stop re-reading your own messages five times before sending, and start sending the sensitive ones with a quick, honest check instead.
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
- Set up a saved Claude Project or Custom GPT with the one line tone check instruction before your next sensitive message, not after.
- Run a message you already sent through the check retroactively to calibrate what counts as risky in your own writing.
- Log at least one wrong read from the AI this week and add a correction line to the saved instruction.
- Add a culture specific line to the instruction if you regularly message colleagues or clients across regions.
- Keep a human second reader in the loop for resignation letters, formal complaints, or anything genuinely high stakes.
Questions people actually ask
How do I get ChatGPT to check my tone before I hit send?
Create a Custom GPT or Project with one saved instruction telling it to read your draft, name the likely emotional reaction in one line, and offer a rewrite only if needed. Save it once so every future message gets the same ten second check instead of you writing a new prompt each time.
Can AI actually tell how a message will come across?
Not perfectly. Research on email specifically found people are far more confident than accurate at judging tone, so an AI second opinion catches things you are too close to notice. But it does not know your history with the reader or inside jokes, so treat its read as a flag to reconsider, not a verdict.
What is the difference between a Claude Project and just asking Claude in a normal chat?
A Project saves your instruction as a standing system prompt, so it applies automatically to everything you paste in that space, no retyping needed. A regular chat forgets your instructions once it ends, so you would have to explain the tone check rules from scratch every single time.
Is it safe to use AI to check tone on sensitive messages like a resignation letter?
Use it for everyday messages like follow ups, but not alone for anything with real consequences, since it cannot read office politics or personal history. Get a second opinion in genuinely high stakes situations, ideally from someone who knows the audience you are writing to.
Why does my AI tone checker keep softening messages I meant to be firm?
Most models default to polite phrasing unless told otherwise. Add a line to your saved instruction saying you want directness preserved and only real rudeness flagged, not assertiveness, so it stops treating a firm, confident tone as something to fix.
Sources
- E-mails and egosAmerican Psychological Association, Monitor on Psychology
- Understanding Claude’s personalization featuresAnthropic
What happens next
Expect tone checking to move from a manual saved prompt toward a built in feature, as Anthropic and OpenAI keep expanding project level and persistent memory options that could hold this kind of standing instruction automatically. Watch whether these assistants start factoring in your actual message history with a specific person, since that context is the piece a generic tone check still cannot see on its own.
