Skip to content

Product case · 13 / 13

Sites that have to be found, and then believed

Product note. Organizations, dates, private data and implementation details are omitted.

Abstract illustration: one page read from the left by a dot and from the right by rings

The problem and the decisions

For small businesses I build the website and then keep it running: in two languages, with the hidden notes that tell Google and the AI assistants what the page is about, the numbers that say whether any of it worked, and something that keeps publishing without me typing every week.

Two audiences now read the same page and they do not read it the same way. A person skims for whether this business is real; a model looks for a claim it can quote with a source attached. A page that only serves one of them is only half built.

The boring half is the checks: the site refuses to go live if a link dies, if one language falls behind the other, or if a number quietly starts reading zero. A check that can never fail is not checking anything, it is decoration.

What it was built with

  • Next.js
  • Vercel
  • Supabase
  • Schema.org
  • llms.txt
  • GA4
  • MCP

What I did

  • SEO
  • GEO
  • Content
  • Instrumentation
  • Bilingual
  • Automated QA
  • Social automation
  • AI-assisted analysis
  • Experimentation

What I learned

Getting found and being believed are different problems. Most sites solve the first and assume the second.

More context · password required →