← What to do

How to train AI to write like you

Mike Taylor, head of evals at Every, lays out how he got AI drafts to sound like his own writing. The usable part for most people: collect 12 to 25 of your longer pieces from one format (or transcripts of your recent meetings if you don't publish), skip tweets and emails as samples, and always draft from a brief that holds your real ideas. Then save the edits you make so your instructions improve over time. His later steps use an AI 'judge' and automated optimizers, which require developer-level setup. He says his own process cut drafting from a day to about an hour; that is his result, not a guarantee.

For
Anyone who writes captions, blog posts, newsletters or client emails with AI help and wants the drafts to sound less generic. The first steps need no technical experience. The optimizer and judge steps are for people comfortable installing tools from GitHub.
Cost
The article is readable through the gift link that Ben's Bites shared; Every itself is a paid subscription. The author's writing skill is posted on GitHub. He reports that running the optimizers cost $27 to $53 per skill at API prices, compared with about $5 for a one-shot style guide, and notes he later used his Claude Code and Codex subscriptions instead.
Access
Open to anyone with the gift link. You need your own writing samples or meeting transcripts and an AI chat tool. The GitHub skill includes setup instructions and assumes comfort with developer tools.
Byte's take

Worth ten minutes if you already lean on AI for writing, because the brief-first habit and the warning against using tweets and emails as samples apply right away. Skip the optimizer section unless you're comfortable with GitHub and paying for API usage.

Your moveRead the guide through the gift link, then gather 12 to 25 of your own longer pieces in a single format and write a short brief with your actual points before your next AI draft.

Start here

More in What to do