You wrote that reply angry. Don’t send it yet. A badly timed blunt message can cost trust that takes months to rebuild, and a ten minute setup catches it before you hit send.
| Setup time | About 10 minutes |
|---|---|
| Tools used | ChatGPT or Claude |
| Output format | Risk score plus line edits |
| Trust in leadership | Drops for 40% of workers after poor communication |
1. Set up a standing reviewer
Open a new project or custom instructions in whatever AI chat tool you use, ChatGPT, Claude, it does not matter. Paste this once: You are my communication risk checker. When I paste a message, rate how it might land, low, medium or high risk, and suggest fixes to specific lines, not a full rewrite. Save it so it is always there when you need it.
2. Add two lines of context
Before you paste your draft, tell it who is reading and what you want to happen. Example: this goes to my VP in Mumbai, we work together but he outranks me, I want the budget cut reversed without burning the relationship. The advice changes completely once AI knows the room you are in.
3. Ask for a score, not a rewrite
Ask for a one line risk read first: low, medium, high, and why. Only then ask for fixes to the risky sentences. This keeps your voice intact instead of getting back a generic corporate reply.
4. Compare three versions before sending
Ask for the message as is, a softened version, and a direct but respectful version. Pick based on who is reading, not on which one sounds best to you. A message that works for a peer can land badly with someone senior.
5. Save your own red flag list
Every time AI flags a phrase, note it somewhere simple, a notes app is enough. After a few weeks you will start catching your own risky lines before you even open the AI tool.
A worked example
A finance manager in Dubai wants to push back on a VP in Mumbai who cut his team’s budget by 30 percent without warning, over Slack.
Before
This is not okay. You cut our budget by 30% without even telling me first. I need this fixed today or I’m escalating to the CFO.
After
Risk read: high, this reads as a threat and skips the hierarchy, likely to make him defensive before you even talk. Rewrite: I noticed our team’s budget was reduced by 30% and wanted to flag it before it affects this quarter’s deliverables. Could we find 15 minutes today to understand the reasoning? Happy to loop in the CFO together if we can’t resolve it between us.
Three ways to bend it
For salary negotiation
Job seekers can run a counteroffer email through the same check before sending it to a recruiter, catching lines that sound demanding instead of confident.
For manager feedback
Before sending a written warning or hard feedback to a direct report, check it for firmness without cruelty, especially useful when English is a second language for either of you.
For customer complaints
Customer facing teams can run a reply to an angry client through the same check to catch anything that might escalate a bad situation instead of calming it.
When it goes wrong
The rewrite sounds fake and corporate
Tell it explicitly to keep your own words and only fix the risky lines, not rewrite the whole message.
I don’t have time to check every message
Only use this for messages that could affect a relationship, a decision, or your reputation. Skip it for routine chat.
It softens everything into mush
Ask for firm but respectful, not polite. Tell it directness is fine, blame language is not.
When not to use this
Skip this for quick internal chat, routine updates, or anything where speed matters more than polish. Checking every message with AI will slow you down and make you sound stiff over time. Save it for the messages that could actually cost you something if they land wrong: pushback to someone senior, conflict, negotiation, or anything you would regret sending in anger.
What you can now do
You can send the hard message today instead of sitting on it for two days, and send it in a way that gets you what you want instead of a fight.
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
- Paste a standing communication risk checker instruction into ChatGPT custom instructions or a Claude Project so it is there every time, not retyped per message.
- Before pasting a draft, add two lines naming who is reading it and what outcome you want, since the advice changes based on hierarchy and relationship.
- Ask for a one line risk read (low, medium, high) before asking for any fixes, then request edits only to the flagged lines to keep your own voice.
- Request three versions, as is, softened, and direct but respectful, and choose based on the reader’s seniority rather than personal preference.
- Keep a running note of phrases the AI flags as risky so you start catching your own blunt lines before opening the tool.
Questions people actually ask
How do I get ChatGPT or Claude to check my message before I send it?
Set up a standing instruction once, in ChatGPT’s custom instructions or a Claude Project, telling the AI to act as a communication risk checker. Tell it to rate incoming messages as low, medium or high risk and suggest fixes to specific lines rather than a full rewrite. Save it so it applies automatically every time you paste a draft, without retyping the setup.
Why does the AI rewrite make my message sound fake or corporate?
This happens when the AI defaults to a full rewrite instead of targeted edits. Tell it explicitly to keep your own wording and only change the lines it flagged as risky. Asking for a one line risk score first, then fixes only to specific sentences, keeps more of your original voice than asking for a complete redo.
What information should I give the AI before pasting my draft message?
Name the reader and the outcome you want, for example the person’s role, whether they outrank you, and what result you are after. This context changes the advice significantly, since a message to a senior VP needs different handling than one to a peer. Skipping this step is the most common reason the suggested fixes miss the mark.
Is this tone check useful for every message I send at work?
No. It is meant for messages that could affect a relationship, a decision, or your reputation, such as pushback to someone senior, conflict, or negotiation. Running it on routine updates or quick internal chat will slow you down and make your everyday writing sound stiff over time, so save it for higher stakes situations.
Can this same method help with salary negotiation or performance feedback emails?
Yes. The article describes running a counteroffer email through the same risk check before sending it to a recruiter, to catch phrasing that reads as demanding rather than confident. The same setup also works for a manager drafting hard feedback or a written warning, especially when checking for firmness without cruelty.
Sources
- Collaborate with Claude on ProjectsAnthropic
- 5 consequences of ineffective workplace communicationAxios HQ
- A Simple Way to Be More Assertive (Without Being Pushy)Harvard Business Review
- How to Reduce Miscommunication in Global TeamsTalaera
What happens next
Expect more of this kind of workflow to move from manual copy paste into built in features, as both OpenAI and Anthropic keep expanding persistent, project level instructions that stay active across an entire chat history. Watch whether tone or risk checking becomes a native email and chat client feature rather than something people set up themselves. The main open question is whether cross cultural and cross seniority nuance gets handled well enough that a canned instruction set does not need constant manual adjustment per reader.
