AI email copy that sounds like a person wrote it

Sales and lifecycle email is where generic AI copy is most obvious and most expensive. Everyone has learned to spot it, and spotting it is enough to stop reading.

Why it matters here

Email is short, so there is nowhere for a wrong register to hide. One overwrought opener or one phrase you would never say is enough to lose the reader, and the tell is usually rhythm, not vocabulary.

How ScriptGrain does it

  1. Build a voice profile from emails and posts you have already written.
  2. ScriptGrain measures how you actually construct a sentence: contraction rate, comma density, whether you open with a hook, context or the point.
  3. Drafts hold to those measurements instead of drifting into the model's house style by the third paragraph.

Emails default to about 180 words. You can generate up to three variants from one brief and keep whichever sounds most like you.

What an email in your voice looks like

Start with rhythm. It's the first thing a reader picks up and the last thing a generic draft gets right.

Most people's emails have a shape they'd never describe out loud. One sender opens with the point and gets a second sentence to justify it. Another sets context first, then lands the ask in a short fragment at the end. A third asks a question in the opening line and answers it by line three. None of this is wrong. It's just theirs.

Then the small stuff. Do you write "I'll" or "I will"? Do you use brackets for asides, or do you split the thought into two sentences? Do you reach for semicolons, or has one never once appeared in your sent folder? Do you sign off with a line, a joke, or just your name?

Length matters too. A good email in your voice is the length you'd actually send. For most people that's shorter than a model wants to write. ScriptGrain defaults emails to about 180 words, and the format allows anything from 60 to 400, so a two-line nudge and a proper outreach note are both on the table.

So here's the test. Cover the signature. Could a colleague still tell it was you? If yes, it's on voice. If they'd shrug, it isn't.

Where AI email copy gives itself away

The tells are rarely single words. Everyone hunts for "delve" and "leverage" and misses the structure underneath.

Here's what actually gives it away:

  1. Uniform sentence length. Every sentence sits in the same comfortable band. Real writers lurch. A long sentence, then four words. Then something in between.
  2. The warm-up opener. "I hope this email finds you well" and its cousins. Nobody who writes tight emails opens that way, and readers have learned to skip it.
  3. Balanced, tidy closes. A neat summary sentence that restates the ask in fresh words. Real emails just stop.
  4. Register drift. The first paragraph sounds like you. By the third, the model's house style has crept back in. That's voice drift, and it's the usual reason a draft reads fine at the top and wrong at the bottom.

You can run a finished draft through the AI cliché checker to catch the stock phrases. Fair enough, that handles vocabulary. It won't tell you whether the rhythm is yours, and rhythm is where most emails lose the reader.

A worked example

Take an illustrative brief: a follow-up to a prospect who went quiet after a demo. Goal: get a yes or no on a next step. Tone: friendly but not pushy.

Without a profile, you'd likely get a three-paragraph note. A polite opener about hoping they're well, a recap of the demo, a soft "I'd love to hear your thoughts," and a sign-off offering to answer any questions. Every sentence about the same length. Nothing you'd say across a desk.

Now add a profile built from a sender who writes punchy, conclusion-first emails, contracts freely, and rarely uses more than one comma per sentence. The brief hasn't changed. The draft has.

The opener becomes the point: a direct question about whether the next step still makes sense. The recap shrinks to one line, because this sender assumes the reader remembers the demo. A fragment lands the ask. Contractions appear where the profile's contraction rate says they should. The warm-up disappears entirely, and so does the tidy closing summary. The email ends on a short, slightly dry line and stops.

Same content, different person. That's the whole idea. The facts of the email are identical; what changes is how the sentences are built, and the profile holds those habits steady from first line to last.

What to feed the profile

Give it emails. That sounds obvious, but people often feed a profile their best blog post or a polished case study, and email is a different register. A profile built from long, careful prose will make your emails stiffer than you are.

A profile needs at least one sample of 50+ words to start. Three or more pieces and about 3,000 words is the recommendation, and for email that means a spread: a cold outreach, a reply to a customer, an internal note, maybe a longer update. Variety shows the profile what stays constant when the situation changes, which is exactly what you want it to hold onto.

Pick things you wrote without a committee. A sent email that three people edited is a blend, not a voice. Choose the ones that genuinely sound like you on a normal day.

Extraction usually takes under a minute (the median measured on 28 September 2026 was 39 seconds), so there's little cost to trying a different set of samples and seeing how the profile shifts. If you want to see what's being measured before you commit, the free writing style analysis tool shows the countable part of it, and the full list is in the writing voice attributes reference. The profile covers 45 attributes across 8 layers, from average sentence length and its variance to opening style and pronoun mix.

Choosing between variants

You can generate up to three variants from one brief. Use them.

The variants aren't there to give you options on content. They're three honest attempts at the same voice, and they'll differ in small ways that matter. One leans a touch more formal. One opens with a question. One is shorter. You're picking the one that reads most like you, which is faster than editing a single draft toward a feeling.

Each draft is scored from 0 to 1 against your profile, the voice match score. On-voice drafts typically land between 0.85 and 0.95. Partial matches sit between 0.5 and 0.8, and a clearly different voice falls under 0.45. The method behind it is published in the voice measurement framework, so you can check the maths rather than take it on trust.

The score is useful, but the per-feature notes are where you'll actually work. Each one gives a draft value against your profile value. If the draft's average sentence length is well above yours, you know to cut. If the contraction rate is low, you know the register has gone formal. You don't have to guess why a draft feels off. Polish can revise a draft toward a target score (the default is 0.9), and the app learns from the edits you save, so the second batch starts closer than the first.

Before anything goes out, read the per-feature notes once. It takes a minute.

Drafting from the assistant you already use

You don't have to leave ChatGPT or Claude to do any of this. The profile reaches them through the MCP server and the REST API, and it's read live each time, so it's never a pasted prompt that goes stale. Both are included on every plan, Free included. More detail is on the API and MCP server page, and the setup notes for ChatGPT and Claude cover each assistant.

In practice: you connect the profile, then brief the assistant the way you normally would. It drafts with your measurements as a constraint, and the same scoring applies. If you'd rather keep prompting in the tool you know, nothing about your habits has to change.

To be fair, both assistants write a perfectly serviceable email from a plain brief. ChatGPT and Claude are strong at structure, at tightening, at catching a muddled ask. What they can't do alone is know how you construct a sentence. That's the gap the profile fills.

If you write email for clients rather than just yourself, the copywriter page covers running several profiles, and the brand voice consistency checker is there for auditing a batch.

Write the email. Check the score. Send it.

Other formats