Adjectives in prompts gave me generic drafts. 50 of my own posts fixed it
When I started scheduling content for multiple platforms, I leaned on AI drafts to keep the queue full. Every draft read fine and sounded like nobody. I tried prompt tweaks for weeks: "casual but sharp", "founder tone", "no corporate speak". Each variation produced the same interchangeable voice, because thousands of other people type those same adjectives.
What worked: examples instead of instructions. I fed the model about 50 of my published posts and the drafts started using my openers, my sentence rhythm, my bluntness. Editing time per post dropped from rewrite-everything to touch-up.
Two takeaways if you are automating any part of your content:
- Describe less, show more. Your archive is the prompt.
- Optimize for consistency, not perfect posts. The queue going quiet costs you more than a mediocre Tuesday post.
I eventually built this workflow into XreplyAI (scheduling for 15 platforms, with drafts trained on your own archive): https://xreplyai.com?utm_source=indiehackers&utm_medium=social&utm_campaign=edusales-2026-07-16
The archive-as-prompt idea is much stronger than piling on tone adjectives. I'd keep 10-15 posts out of the example set and compare generated drafts against them; otherwise the model can sound familiar while mostly copying surface quirks. Did you notice whether the gains came more from topic selection, openings, or sentence rhythm?