AI-drafted, human-refined: our editorial process
A look inside how we combine machine speed with human judgement.
We are often asked how a boutique studio publishes as much considered content as it does for its clients. The honest answer is that we lean on AI heavily — but not for the parts of the work most people assume. This is the process, in full.
Step one: the voice audit
Before we write a single post, we spend a week collecting a client's raw voice: interview transcripts, voice notes, old emails, the way they explain their work in a discovery call. We are not looking for polish. We are looking for cadence, vocabulary, and the moments where their real point of view surfaces.
Step two: the idea library
Every client has a private document of one-line ideas — provocations, frameworks, client stories worth telling. We add to it every week. AI never populates this document. It is the source material for every post, and it is where authority actually lives.
Step three: structured drafting
This is where the machine helps. We feed one idea from the library, plus voice reference material, into a private model prompt tuned to the client's cadence. The output is a structural draft — never final copy. Its job is to save us the blank page.
The blank page is where good writers give up. AI's most honest use is to remove it.
Step four: human editing
A human editor rewrites the opening, sharpens the middle, and always rewrites the closing. We cut anything that sounds like a model. We add the specific details only the client would know. Roughly 60% of the finished post is human-written, and 100% is human-approved.
Step five: client approval and scheduling
Nothing publishes without the client's signature. Ever. That is the entire quality control system.
It is not a magic process. It is a disciplined one. And it is the reason clients tell us the posts sound more like them than the ones they wrote themselves.