How to replicate a founder's writing style

By Jack Stovell X · Instagram · LinkedIn · published 2026-09-20 · updated 2026-09-30

Every ghostwriter knows the note. "It's good. It's just not quite me." No detail, because the founder cannot say what is off; they only know it when they read it. The writer guesses, revises, and the second draft is a little more generic than the first, because generic is the safe direction.

Replicating a founder's voice is a measurement problem, and the measurement removes the guessing.

Score the drafts. Paste something the founder wrote themselves, then the ghostwritten or AI-drafted pieces. Each one is scored against the founder's own writing, with the habits that slipped named and numbered. That list is the note the founder could not give.

What the founder cannot tell you

Voice lives in habits nobody articulates: contraction rate, how many asides go in brackets, whether they open with the point or build to it, how short their shortest sentences get, whether the humour is dry or warm or absent. Ask a founder to describe their style and you get adjectives. Measure three of their emails and you get numbers that a writer, or a model, can actually match.

The three-step version

  1. Collect the founder's own writing. Emails, Slack messages, memos, anything written without help. 3,000 words is plenty; 1,000 will do to start.
  2. Measure it. A profile turns that writing into 45 attributes and a narrative. The free check on this page does a one-piece version.
  3. Score every draft before it goes to the founder. Fix the two or three features that moved. The founder's note stops being "not quite me" and becomes "yes".

If AI is doing the first draft

Then the measurement is the prompt. Give the model the founder's numbers, not "write like a direct, visionary CEO", and hold the output to them with a score. ScriptGrain's Studio and Agency plans keep one profile per client and let the whole team draft against the same measured standard, which is what "consistent voice across writers" actually means.

Most teams still treat inconsistent tone as a discipline problem, and another style document won't fix it. The real cause sits elsewhere: your writers are putting words down in different written accents, and no reminder or "quick sync" changes that. We break down why this happens and what a measured profile does about it in why team writing sounds disjointed.

Where a founder's voice actually shows up

A founder doesn't have one voice. They have four, and they overlap only a little.

The investor update is the most guarded. Numbers up front, short sentences, very little colour. The LinkedIn post is looser, with more "I" and the odd joke. The launch note sits somewhere between the two, warmer than the update but still trying to sound sure. The hiring page is the strangest of the set: it's where founders either sound like themselves or sound like a template.

So which one are you replicating? All of them, but not with the same settings.

This matters because the attributes move with the format. A founder might write at 18 words a sentence in an update and 12 in a post. Contraction rate climbs when they're talking to candidates. Their opening habit might hold steady across everything (point first, always), and that steady feature is the one to lock down before anything else.

A quick example. Update: "Revenue grew. Churn didn't. Here's why." Hiring page: "We're small, we argue a lot, and we ship on Fridays." Same person. Different length, different warmth, same habit of stating the conclusion first.

Measure each format separately if you can. The habits that survive every format are the core of the voice. The ones that shift are the dials.

Choosing the samples a founder signs off

Pick writing the founder wrote, not writing the founder approved. Those are different things.

A piece a comms team has polished will read well and measure badly. Sentences get evened out. Contractions get expanded. The dry aside gets cut because someone thought it was risky. You end up with a profile of the comms team's house style, and it will score the founder's real emails as off-voice. Fair enough, that's a strange result, but it's what you'd expect.

Avoid:

  1. Anything that went through a communications or PR review before publishing.
  2. Press releases and boilerplate, even with the founder's name on them.
  3. Posts drafted by an agency and lightly tweaked.
  4. Old pieces from a period when the founder was copying someone else's tone.

Prefer emails, Slack messages, memos and anything typed quickly without an audience in mind. Then add two or three pieces they're genuinely proud of, because pride shows up in the rhythm too.

On quantity, a profile needs at least one sample of 50+ words. Three or more pieces and about 3,000 words is the recommendation. Mixed formats beat one long document. Three emails plus a memo plus a post tells you more than 3,000 words of memo.

One check before you commit. Ask the founder whether they'd sign off on each sample as theirs, and whether they actually wrote it. You only want the yeses to both.

When the founder's writing changes

It will. Founders get more confident after a funding round, more hedged after a bad quarter, shorter after a year of writing to a board. A profile built eighteen months ago is a photograph, and the founder has kept ageing.

The signal is usually in the scores. Drafts that used to land between 0.85 and 0.95 start dropping into the 0.5 to 0.8 range, and the founder's note creeps back to "not quite me". Check the per-feature deltas before you blame the writer. If the same two features keep moving, say sentence length and hedging, the profile has gone stale rather than the drafts going wrong.

So refresh it. Add recent writing, the last few months of emails and posts, and drop the oldest samples if the gap is large. You don't need to start again. A new extraction usually takes under a minute (median 39 seconds, measured 2026-09-28).

One more thing. Change isn't always drift to fix. Sometimes the founder has deliberately moved, more direct, less formal, and the profile should follow them. Ask. If they say yes, update the target and let the score measure against where they are now.

And because the profile is read live through the MCP server and the REST API, every tool drafting against it picks up the refresh straight away. Nobody has to re-paste anything.

Questions

How much of the founder's writing do I need?

Enough to be them: 1,000 words of unassisted writing starts a profile, 3,000 makes it steady. Slack and email count; polished marketing copy someone else touched does not.

Can several writers share one founder profile?

Yes. Studio and Agency plans share profiles across a team, so every writer's draft is scored against the same measurement rather than each writer's impression of the founder.

Is this different from asking the founder to review?

It happens before the review. The founder still reads the piece; they just stop being the only instrument that can tell whether it sounds like them.

Related