I spent months drafting social posts with AI to stay consistent as a solo founder. The posts shipped on schedule and got zero traction. No replies, no inbound, nothing.
The content wasn't bad. It was generic — it sounded like every other AI post, because it came from the same default patterns: tidy three-item lists, a relentlessly even tone, the same vocabulary the model reaches for when you don't give it anything specific. On a feed, generic doesn't read as neutral. It reads as noise, and people scroll past without registering it.
The fix wasn't a cleverer prompt. It was changing the input. Instead of describing how I wanted to sound, I fed the model ~80 of my own past posts and let it work from those — my real sentence rhythm, how blunt I get, the jokes I'd actually make. The drafts came back rougher and more specific, and that's when engagement showed up.
The mental model I landed on: AI is a remix engine, not a personality. If you don't supply yours, it defaults to the average of everyone's writing — and average is invisible.
I built this into a tool (XreplyAI) — it trains on your own archive so drafts sound like you, BYOK so you pay your AI provider directly. But the lesson holds even if you wire the workflow yourself: ground the model in your real writing, not a prompt.
Top comments (0)