The before-you-hit-send tone check (Claude project)

What you're building

A saved Claude project with one standing instruction, so every message you drop in gets rewritten to keep your exact point but land as clear and kind, not cutting. A project in Claude is a reusable workspace: you set its instructions once, and every chat you open inside it follows them automatically, so you never re-paste "make this kinder" again. Projects are available on every Claude plan, including free (free accounts can create up to five), per Anthropic's help center.

Set it up once (about 3 minutes)

1. 🗂️ Open projects. Go to claude.ai/projects (or hover the left side of Claude and click Projects), then click + New Project in the top right.
2. ✏️ Name it "before I hit send" and give it a short description like "tone check for messages before I send them." Claude does not read the name or description, they are just for you.
3. 📌 Add the standing instruction. Click Set project instructions, paste the prompt in the box below, and click Save instructions. Claude now applies this to every chat inside the project (source).
4. 📥 Use it. Any time a message makes your jaw tense, open the project, paste the harsh draft, send. You get a softer version plus a note on what to cut.

The standing instruction to paste

You are my before-I-hit-send tone check. When I paste a message, do all of this:
1. Rewrite it so it keeps my exact point and every real ask, but lands as clear and kind, never cutting, sarcastic, or passive-aggressive.
2. Name what in my draft is genuinely helpful to say, so I keep it.
3. Cut what does not need to be said at all, and tell me what you cut and why.
4. Flag any line that reads as belittling, blaming, or condescending, and show the fix.
5. Keep my natural voice. Do not make it corporate, gushy, or over-apologetic. Match the formality of the original.
6. If the kindest and most effective move is to send nothing, or to say this in person instead of in writing, tell me that plainly.
Give me the rewrite first, then a short bullet list of what you changed.

Why this works (the actual mechanic)

The reason harsh feedback backfires is not politeness, it is defensiveness: sharp wording makes the other person protect themselves instead of hearing the point, so the message you cared about gets lost. Clarity lands better than criticism because it keeps the other person's attention on the fix, not on defending themselves. The prompt above forces Claude to separate two things people usually blur together: the point (which you keep) and the tone (which you soften). That is why it does not just "make it nicer," it protects the actual message.

3 bonus prompts to run next

• 📨 The email version: "Here is an email I am about to send while annoyed. Rewrite it so it is warm and direct, gets my ask across in the first two lines, and does not sound cold or passive-aggressive. Then tell me if any line could be read the wrong way."
• 🪞 The read-it-back check: "Read this message as if you were the person receiving it on a bad day. Tell me honestly how it would land, what they would feel, and the one line most likely to start a fight. Then give me a version that avoids that."
• 🤝 The hard-conversation opener: "I need to give someone difficult feedback. Here is the situation. Draft an opening that is honest and specific, leads with what I actually want to happen, and does not bury the point in filler. Keep it short."

How to get the most out of it

• 📎 For beginners: you do not need a project to start. Open any Claude chat, paste the standing instruction, then paste your message underneath. The project just saves you from re-pasting the instruction every time.
• 🎨 For power users: turn the same rule into a Claude custom style so it follows you into any chat, not just this project. Click the + next to the message box, hover Use style, then Create & edit styles, and build one from a writing sample or a description. A style is lighter than a project when you just want the tone applied everywhere.

An honest heads-up