← Back to WorkGrowth & SEO

Self-Evolving Content Sites

Affiliate sites that run themselves. A daily AI routine reads its own traffic stats, picks one improvement, ships it, and logs it—while a browser-free pipeline auto-publishes branded Pinterest pins.

AstroCloudflarePinterest APIeBay EPNClaude Code

The Challenge

Affiliate content sites live or die on fresh, well-targeted pages and a steady stream of qualified traffic. But hand-maintaining catalogs, SEO, and social distribution is a full-time grind—and it does not scale across multiple sites at once.

The Approach

I built two Astro sites on Cloudflare Pages with first-party analytics in Workers + D1, so traffic is an asset I own and can query—not a number trapped in someone else's dashboard. Each site has an autonomous daily routine that reads its own stats, evolution log, and git history, decides on exactly one high-leverage move (fix a dead link, deepen a winning page, fix a weak converter), implements and verifies it, then commits and logs the decision. A daily scraper keeps prices honest, and a headless-Chrome pipeline renders branded Pinterest pins and publishes them through the official API with no external image hosting.

The Outcome

Two self-maintaining content sites that grow their own catalogs, improve their own SEO, and distribute themselves every day—behind a verification gate that blocks any change that would leak a non-compliant price or ship an empty page. Built to compound while I sleep.

Key Details

  • Astro static sites on Cloudflare Pages; first-party traffic logged to Workers + D1
  • Autonomous daily "evolution" routine: read stats → pick one move → ship → log
  • Browser-free Pinterest auto-pin pipeline (headless-Chrome render + API publish)
  • Daily price-and-stock scraper keeps product data fresh and honest
  • SEO: year-keyed pages targeting "best X [year]" query families, deep facet coverage
  • eBay Partner Network + Amazon Associates monetization, commission-tier ranking
  • Pre-push verification gate catches price-leak and empty-page regressions