How to Edit an AI Email Draft So It Sounds Like You
The short answer
Edit an AI email draft in three passes: first check facts and commitments, then cut the opener and move the ask to the first line, then swap in your contractions, break the symmetry, and restore your sign-off. These eight targeted changes make most drafts sound like you in ninety seconds without rewriting the whole thing.
Edit an AI email draft in three passes: fix facts, fix structure, then fix voice. Eight targeted edits do most of the work in under 90 seconds.
On this page
The AI draft is sitting there. It is structurally correct, it covers the points, and it reads like it was written by a pleasant, overly cautious person who is not you. The opener is warm in a way you never are. The three bullets are too symmetrical. The ask is buried in paragraph two. The sign-off says 'Best regards' and you always just type 'Thanks.'
How to edit an AI email draft so it sounds like you is not the same problem as how to rewrite it. Rewriting takes longer than writing from scratch and defeats the purpose. What you actually need is a short, ordered set of targeted edits that change the draft from a competent generic reply to something a reader who knows you would not flag.
This guide gives you that: a three-pass method, eight specific edits, a table of how the same changes land across ChatGPT, Claude, Copilot, and Gmail's Gemini, and a section on what to do when a draft is so far off that targeted edits are not enough. The whole thing is designed for ninety seconds of editing, not ninety minutes.
Before You Start: Confirm the Facts#
The first pass is not about voice at all. It is about facts. Read the draft once for anything the AI might have invented: a date you did not provide, a figure you did not give it, a commitment phrased as 'as we discussed' that you did not discuss. AI models fill factual gaps with plausible-sounding values, and a wrong date or made-up price goes to a real person with real consequences.
This pass takes fifteen seconds if you wrote a tight prompt. It takes longer if you let the model draft loosely. Flag any invented detail with a placeholder, fix it from memory or check your notes, and only then move on to structure and voice. A draft that sounds perfectly like you but promises a price you never agreed to is a liability, not an asset.
Once the facts are clean, you are ready to edit for structure and voice.
Facts first, always
The Three-Pass Editing Method#
Work in three passes, in order. Mixing them up means you fix the voice on a paragraph you are about to delete, or delete a paragraph you needed for structure. The passes are: facts (done above), then structure, then voice.
Structure edits are fast and objective: move the ask, cut the opener, break the symmetry. Voice edits are the personal layer: swap in your contractions, restore your sign-off, change one word in each sentence from the model's vocabulary to yours. Running structure before voice means you are not polishing prose you are about to reorganize.
- 1
Pass 1 — Fix the facts
Read for invented dates, figures, names, and commitments. Replace any unverified claim with the confirmed value or a placeholder. This pass is done when no sentence could embarrass you if the detail turned out to be wrong.
- 2
Pass 2 — Fix the structure
Move the ask to the first line if it is buried. Cut the entire opening sentence if it is a throat-clearer ('I hope this finds you well,' 'Thank you for reaching out'). Break any symmetrical three-bullet list where only two bullets are real points. Delete any sentence that only restates the one before it.
- 3
Pass 3 — Fix the voice
Swap the model's contractions for yours (or add them where you use them). Change one or two words per paragraph from the model's defaults to words you actually say. Restore your real sign-off. Read the whole thing once at speed — if a phrase snags, replace it.
The Eight Edits That Do Most of the Work#
Within the three passes, eight specific edits account for the majority of what makes a draft sound generic. Not every draft needs all eight. Most need at least four. Run through the list on each draft until the pattern becomes automatic.
The edits are ordered roughly by how much they change the reader's impression. The opener is the first thing the recipient reads; getting it wrong sets the wrong tone for everything that follows. The ask placement determines whether the email works. The sign-off is the last impression. Voice edits in the middle matter less than those three structural ones, but they are what makes the difference between a draft that no longer reads as AI and one that reads as you specifically.
| Edit | What to remove | What to replace with |
|---|---|---|
| 1. Cut the opener | 'I hope this email finds you well' or any throat-clearing first sentence | Nothing — start on the first real point, or a one-word acknowledgment ('Got it.' / 'Thanks for this.') |
| 2. Move the ask to line 1 | The ask buried in paragraph 2 or 3 | The ask in the first sentence or the first line after a single-sentence context |
| 3. Restore your contractions | 'I would,' 'I am,' 'it is,' 'we will' — formal expansions you would not use | 'I'd,' 'I'm,' 'it's,' 'we'll' — or vice versa if you write formal |
| 4. Break the symmetry | Three bullets of exactly equal length covering three equal-weight points | Two bullets where two points matter, prose for the third, or one sentence if there was only one real point |
| 5. Swap one word per paragraph | The model's default vocabulary ('appreciate,' 'leverage,' 'endeavour,' 'please do not hesitate') | The word you would actually type — usually shorter and more direct |
| 6. Restore your sign-off | 'Best regards,' 'Kind regards,' 'Warm regards' — whatever the model defaulted to | Your actual closer: 'Thanks,' 'Cheers,' 'Best,' or nothing — whatever you always use |
| 7. Fix the register | Formality that does not match the relationship (formal to a daily teammate, casual to a new client) | The register that matches how you actually write to this specific person |
| 8. Remove the hedge | 'I just wanted to,' 'I was wondering if perhaps,' 'feel free to,' 'please do not hesitate to' | The direct version: state the ask plainly — 'Can you?' / 'Please send' / 'Let me know by Friday' |
How This Looks Across Platforms#
The edits above work on any AI draft. The workflow for applying them differs by platform. Some tools let you send a repair instruction to the draft; others require you to edit the text directly. Knowing which approach to use saves time.
| Platform | How to apply the edits | Best for |
|---|---|---|
| ChatGPT / Claude | Paste the draft with a targeted repair instruction: 'Remove the opening sentence and move the ask to line 1' | Structural fixes that require reordering — the model handles the rewrite |
| Gmail (Gemini) | Use the 'Refine' button with a short instruction, or edit the text inline | Quick voice fixes — Gemini's refinement is good for tone changes within Gmail |
| Microsoft Copilot (Outlook) | Use the coaching panel for tone feedback, then edit inline | Register checks — Copilot flags formality mismatches well |
| Any AI email client | Edit the draft directly in the compose window before approving | Fast passes 2 and 3 — the draft is already in your inbox, no context-switching |
What To Do When the Draft Is Too Far Off#
Some drafts are not worth editing. If you have made three targeted edits and the draft still sounds nothing like you, the problem is usually the prompt, not the output. The model had too little context — no example of your voice, no clear ask, no relationship context — and it wrote to the statistical average of all email. Editing the output of a bad prompt produces a mediocre result; fixing the prompt produces a good draft.
The fix is not to rewrite the email yourself. It is to run a better prompt: add one or two sentences from a previous real email you sent to this person or this type of person, specify your usual opener, name your sign-off. That single addition moves the draft from generic to close enough to edit in two passes.
If you are doing this repeatedly — spending more than ninety seconds editing most drafts — it is a signal that the way you are prompting the AI does not include enough of your voice. The solution there is a personal style guide: a short block of instructions you include in every prompt, covering your opener defaults, your contractions, your sign-off, and one or two vocabulary rules. You write it once and paste it every time.
When to stop editing and fix the prompt instead
A Faster Way: Voice Configuration Before the Draft#
Everything above is the manual craft layer — it works on any AI draft from any tool, and it gets fast with practice. But if you are running the same eight edits on every send, you are fixing a gap the tool should not leave in the first place.
We build AI Emaily. It is an AI-native email client that drafts in your voice from a user-set Context Brain and client profiles you control — the Personal Context you configure, not a guess from your sent mail. You set what the agent knows about your tone, your defaults, your sign-off, and the relationship with each client. The draft arrives already using those settings. You still review and approve before anything sends — Copilot mode keeps you as the final check — but the review is a ten-second confirmation rather than a ninety-second edit pass. A 7-day free trial is available at app.aiemaily.com/signup; pricing is at aiemaily.com/pricing. If you find yourself running this guide on every draft, that is the problem AI Emaily is built to solve.
Frequently asked
See it in AI Emaily
Keep reading
Sources
Written by
Nafiul HasanNafiul Hasan is an entrepreneur and AI automation system builder with 10+ years of experience turning messy, manual workflows into reliable automated systems. He designs and ships AI enterprise solutions end-to-end — the agent logic, the data plumbing, and the product people actually use — and founded AI Emaily to give busy professionals their attention back. He writes here from the builder's seat: what works, what breaks, and how to put AI to work without giving up control.