An SEO engine that writes itself

DeMejico is a Spanish and Mexican furniture manufacturer in Valencia, CA, run by one owner doing all the marketing himself. His blog sat dormant for years while search demand for his exact products went unanswered. I designed and built a system that pulls live GA4 and Search Console data, selects the highest-opportunity keyword, researches competitors, writes in his brand voice, and publishes to WordPress four times a week on a staggered schedule.

The user is a business owner with zero spare hours. Every meaningful decision in this project is a UX decision about a person delegating to software.

The owner talks to the system through Discord: a quick-questions channel routed to a fast, cheap model, a content channel on a mid-tier model, and a strategy channel on the strongest one. Model-per-channel routing is a design pattern. It maps cost and capability to the weight of the question, invisibly, so the owner just picks the room that matches his intent.

On one run the system found a keyword adjacent to an existing content cluster, wrote the post, and flagged the overlap for the owner to consolidate or keep, rather than deciding silently. Designing what an agent decides alone versus what it raises is the same hierarchy-of-decisions work as designing a checkout, with the same failure mode when you get it wrong.

The keyword universe is self-replenishing: as posts rank, Search Console surfaces new opportunities, so the system feeds itself instead of burning through a fixed list and going thin.