A free guide from the Finally archive
How I designed AI content to keep sounding like me in week six — not like everyone else on the internet.
The what & the why. No prompts required to understand it.
Who's teaching this
A licensed clinician with 27 years in practice. I built an AI content engine for my own brand: one idea became a week of posts in my voice, with review before publication. This is an archive of that build; the scheduled agents are now off.
The problem nobody warns you about
Week one, it sounds like you. By week six it sounds like every other account in your niche — same hooks, same rhythm, same beige advice. You can feel it go generic, but you can't see why.
Drift isn't bad luck. It's four specific failures — and every one is fixable.
The four silent drift failures
You prompt "in my voice" — but your voice was never actually written down. So the model guesses, and guesses average.
The same prompt writes, designs, and schedules. A jack of all trades that masters none of them.
Nothing scores the post before it ships. The off-brand ones slip straight through to your feed.
You start from what's trending, not from what you know. So you sound like the niche, not like the expert in it.
The fix
Here's the shape of the system I built. The exact templates, prompts, and rubric were part of that historical build — this is the map.
Fix 01 · the foundation
Pull your best-ever posts, emails, and talks — the ones that sounded unmistakably like you — and turn the patterns in them into a reusable spec: how you open, what you refuse to say, your rhythm, your metaphors, your point of view. The model stops guessing and starts matching.
Your voice comes from your best writing — not from a personality slider.
Fix 02 · the structure
Not one do-everything agent. A handful of workers, each with a single job:
Turns one idea into a post — held to your voice spec.
Scores it against your standard before anyone sees it.
One idea → a week of formats across channels.
Was designed to queue approved pieces for publication.
Fix 03 · the guardrail
Nothing ships on vibes. Each piece gets scored against a fixed standard, and the score decides what happens next:
The gate is what stops week-six drift before it happens.
Fix 04 · the source
Seed the whole engine with the questions your patients and clients actually ask — the things you explain over and over. Now the content is yours at the root: your field, your answers, your take. Trends decorate it; they don't drive it.
This is the difference between sounding informed and sounding like the expert.
Put together
Voice spec + narrow workers + a quality gate + my own expertise as the source — that was the container. The approval gate kept publication tied to a human yes.
Where this goes next
This is the architecture I used for my own content system. The former Finally service offers are closed. My current work lives at Local Current.