Hi Reader,

My client Donna and I have worked together for almost two years. She's a private chef to celebrities and conceptual artist behind Recipes4Survival — a larger-than-life, unmistakably New York personality, quick, funny, honest to a fault, with a rich history, real skills, and an art background that shows up in everything she touches. She stands for sustainability and mindfulness, and in nearly two decades of building her name, nothing about her has ever been generic.

I convinced her that building an online presence takes consistent publishing. But consistent publishing, on top of a full-time career as a celebrity chef, is a full-time job of its own. So we turned to AI to help her keep up. I already knew going in that AI left without real guardrails defaults to a flat, generic tone — and very quickly, in our first pass, that played out exactly as expected. Handing AI a topic and a format was never going to produce anything that actually sounded like her.

So we did something simple instead — full intake, years of her actual writing gathered and used to build a real voice profile, not a style guide, not three adjectives, an actual profile the AI could draw from every time instead of guessing.

This isn't just a hunch. OpenAI's own May 2026 release notes say their model update specifically improves how it uses your past chats, files, and context — and shows you what it used, so you can see it working. Anthropic's research backs it from the usage side too: people who've stuck with one AI tool for six months or more see 10% higher success in their conversations, not because they got smarter, but because the tool finally had something real to work with. (Full citation on the blog.)

Here's where most people go sideways, including us before we fixed it: they try to teach five different tools their voice at once. Same work, five times, done badly each time, instead of once, done well. Depth before breadth isn't a nice-to-have — it's the actual mechanism.

The exact method we used — what actually goes into a real voice profile, and how to build one from material you already have — is what's waiting on the other side of this link.

​Read the full post →​

P.S. — the full method, plus more of her story, is on the blog.