How to Fix the Tone of an Email Using AI

Fix Email Tone AI: Diagnose First

Here’s the one-line version: read your draft once for tone before you read it for content, then sort what you find into “too blunt” or “too buried.” Everything else in this guide follows from which bucket you land in.

Signs Your Draft Reads Too Harsh

• Short, clipped sentences with no context (“Send it today.”)
• No acknowledgment of the reader’s situation or effort
• Absolute words like “unacceptable,” “again,” or “clearly”
• It reads fine in your head and cold on someone else’s screen

Signs Your Draft Reads Too Soft

• The actual ask shows up in the second half of paragraph three
• No deadline, or one wrapped in “whenever you get a chance”
• Multiple apologies for a request that needed none
• A reader could skim it and walk away with no clear action

Most people try to run this check in their head, which is exactly why it fails — you already know what you meant to say, so you read intention instead of words on the page. A manager rereading a pushback email to a vendor almost always rates it as “fine,” because they can hear their own reasonable tone of voice while reading it. The vendor, with no access to that internal voice, just sees the blunt sentences.

Don’t fix anything yet. Read top to bottom and notice where you flinch, on either side — too sharp or too soft.

Instead of “this sounds bad,” land on one specific thing: no deadline, three apologies, an absolute word in paragraph two. That specific signal is what you hand the AI next — it’s what turns a generic rewrite into a targeted one, and it’s the difference between a fix that takes one pass and one that takes four.

If you’re starting an email from nothing rather than repairing a draft, this diagnosis step doesn’t apply the same way — writing a hard email from scratch is a related but different job.

What AI Rewrites Still Miss

AI can shift register — formal to friendly, clipped to warm — but it can’t read the actual relationship behind the email. That’s the honest limit worth knowing before you lean on any of the tools below.

The trade-off shows up most on the firmness question. Soften a message too far and a genuinely overdue invoice reads like a gentle reminder with no consequence attached; leave it too sharp and a first-time slip reads like an accusation. Neither the tool nor a generic prompt can weigh that for you — it’s a judgment call you make before you ask for the rewrite, not after.

There’s also a data question worth being deliberate about, since you’re pasting real work content into a chat window.

None of this means skip the AI pass — it means treat it as a second opinion on wording, not a replacement for judgment about what to say and how firm to be.

Fixing Tone in ChatGPT

ChatGPT doesn’t have a single “fix my tone” button. Canvas’s built-in shortcuts cover length and reading level, not tone directly, so the actual fix happens by asking for it in plain language — in a normal chat, or inside Canvas if you want the edit visibly tracked.

Give ChatGPT the specific problem you found in the diagnosis step, not just “make this better.” The more context about the relationship, the less generic the result.

Here's an email I'm about to send. Read it for tone only.

The problem: [too blunt / too buried — name the specific signal, e.g. "no deadline" or "an absolute word in paragraph two"]
My relationship with the reader: [e.g. "colleague I know well" or "client I haven't met yet"]

Rewrite it to fix that specific problem. Keep my original wording everywhere else — don't rewrite sentences that weren't the problem, and don't soften or remove the actual ask or the deadline.

Original email:
[paste draft here]

If the first rewrite overcorrects — too stiff after fixing “too blunt,” too pushy after fixing “too buried” — say so directly: “that’s too formal now, dial it back one notch.” Treating this as a two- or three-turn conversation beats trying to nail the exact phrasing in one prompt.

If the draft is long, or you want to see exactly what changed, open it in Canvas first and ask for the same fix there — the “Change reading level” shortcut is the closest built-in tool to a tone adjustment, useful when the problem is more “too stiff” than “too harsh.”

One plan note: Canvas isn’t available on ChatGPT’s pro-series reasoning models, so if you don’t see it, that’s why — check which ChatGPT plan you’re on if the shortcuts menu is missing entirely.

Fixing Tone in Gmail With Gemini

Gmail’s Help Me Write panel gives you three fixed directions, plus a box for typing anything more specific — which matters, because none of the three buttons is really “fix the harshness.”

Formalize is the one people reach for by habit, and it’s the wrong choice for a harsh draft — it tends to make blunt sound formally blunt, not warmer. Friendly is the closer match for that job. For a buried draft, Shorten is worth trying before Friendly: trimming the hedging often surfaces the ask on its own, without needing a separate tone pass at all.

Click the pencil-and-star icon in the compose window, then use Friendly for a too-harsh draft, or type a specific instruction for anything the three buttons don’t cover.

Keep the deadline and the specific ask exactly as written. Make the opening line warmer and remove the word "[the absolute word you flagged]."

Worth knowing before you rely on this daily: Gemini’s tools inside Gmail need an eligible Google Workspace or Google AI plan, or a personal account restricted to the US — check your access if the prompt bar isn’t showing up. Google’s own support documentation covers the Formalize, Friendly, and Shorten options in more detail if you want the full picture, including how the mobile version differs from desktop.

Fixing Tone in Outlook With Copilot

Outlook actually gives you two different entry points, and picking the wrong one wastes a step: Draft with Copilot sets a tone before anything is written, while Coaching by Copilot reviews a draft you already wrote. As a rule of thumb, reach for Draft with Copilot when you’re starting from nothing and already know it’s a sensitive send; reach for Coaching by Copilot when you’ve written something and just aren’t sure how it lands.

Starting from a blank email, select Draft with Copilot, describe what you need to say, then use the Adjust control to set a tone and length before Copilot generates anything.

Let [reader's name] know [the situation in one sentence]. Ask for [the specific action] by [the deadline]. Keep the tone [warm but direct / firm but respectful — pick one].

Already have a draft? Select the Copilot icon, choose Coaching by Copilot, and it reviews what you wrote for tone, clarity, and how a reader is likely to take it — then offers suggestions you can apply one at a time or all at once.

Here’s what trips people up: Coaching by Copilot needs a Microsoft 365 Copilot work licence. According to Microsoft’s own support documentation, a personal or free Outlook account won’t show the option at all, no matter how you search for it. Copilot inside Outlook covers the licensing question in more depth if you’re not sure what your organization has.

Fixing Tone in Claude

Claude doesn’t have a one-click “make this friendlier” button either — you ask directly, the same as ChatGPT, with one extra option: a saved Style that sets your register once instead of every time.

Paste the draft and tell Claude what’s wrong and who it’s for. The relationship context is what separates a rewrite that sounds like you from one that sounds like a form letter.

Here's an email I'm about to send. Read it for tone only.

The problem: [too blunt / too buried]
Who it's for: [e.g. "a colleague I know well" or "a client I haven't met yet"]

Fix that specific problem and keep everything else — including the ask and any deadline — exactly as written.

Original email:
[paste draft here]

For a longer email, ask Claude to open it in a Claude Artifact first — you’ll see the edited version alongside the original, which makes it easier to confirm the ask and deadline survived the rewrite.

If you’re doing this daily, a custom Style is worth the five minutes it takes to set up. Claude’s presets are Normal, Concise, Formal, and Explanatory, and a custom one can be built from a few emails that already sound like you, so every draft starts closer to your voice instead of a generic professional default. Feed it two or three emails you’re genuinely happy with, and it picks up your habits — shorter sentences, a particular sign-off, however you actually write.

Styles work on every Claude.ai plan, including the free one — which Claude plan you’re on only matters for things like usage limits, not whether this feature is available to you.

Match the Fix to What You Have

You don’t need all four tools open — you need to know which one you already have, and what it’s actually good at. However you fix email tone AI advice points you toward, the fifteen-second habit matters more than the specific app you run it in.

If you’re inside Gmail or Outlook already, use what’s built in — it’s one click and doesn’t break your flow. If you’re in ChatGPT or Claude for other reasons, it’s just as fast to paste the draft in rather than switching apps. The real trade-off is Outlook’s Coaching feature: it’s the most thorough review of the four, but only if your organization has the Copilot licence for it.

A freelancer on a personal Gmail account and a manager on a corporate Microsoft 365 seat are, in practice, working from two different toolkits for this exact task. The diagnosis stays identical either way — only the entry point changes, which is really the whole point of building the habit around a test rather than around one specific app.

If None of These Are Open

Any general-purpose chat AI can run this fix — the prompt from the ChatGPT section works nearly word-for-word anywhere.

Once you find a prompt phrasing that works for you, save it instead of retyping it. In ChatGPT, that’s custom instructions; in Claude, it’s a saved Style.

When I ask you to fix the tone of an email, always:
1. Ask me whether the problem is "too blunt" or "too buried" if I haven't said.
2. Keep the specific ask, deadline, and any numbers exactly as written.
3. Change only what's causing the tone problem — don't rewrite the whole email.

PromptPeakAI’s free AI Work Templates include a version of this prompt already formatted to paste in, if you’d rather not build your own.

When the AI Rewrite Still Misses

Sometimes the fix overshoots, undershoots, or just doesn’t show up. Here’s what each symptom usually means.

The Blunt-or-Buried Test isn’t a one-time trick — it’s a habit worth keeping for anything you’re about to send under your own name. Most of the fix is the fifteen seconds of diagnosis before you ever open an AI tool; the tool itself is just doing what you tell it, in whichever app happens to be open. Get the diagnosis right consistently, and which of the four tools you reach for stops mattering nearly as much as people expect it to.