Back to case studies
BUILD//Own project//2026-04-17//5 min

CiteThis: Evidence-Based Research Platform

→Working product in 3 days, zero infrastructure cost
CiteThis: Evidence-Based Research Platform
Visit website

Starting point

April experiment: build something where every technical detail is intentional. Topic - evidence-based medicine (supplementation, sleep, ADHD, postpartum). Goal - not to write "another blog," but a research platform that AI engines can and will cite.

What I built

16 evidence-based protocols, each ~2500 words. Structured frontmatter, key definitions, methodology notes, FAQ sections, comparison tables, safety considerations, DOI links to primary studies.

33 tag landing pages with dedicated explainers (~2100 words each) and DefinedTerm schema markup. When someone searches "what is myo-inositol for PCOS" in ChatGPT, citethis.site is directly an answer, not just a list of articles.

Tech stack

  • Framework: Astro 6 (SSG mode) — no client-side JavaScript, clean HTML for AI crawlers
  • Content: Markdown content collections with TypeScript validation
  • Search: Pagefind (static search, 60KB JSON)
  • Deployment: Vercel (free tier, TTFB under 100ms)
  • Design: Tailwind CSS, brutalist dark-mode, WCAG AAA

GEO-first architecture

6 layers of structured data on every page:

  1. ScholarlyArticle + MedicalWebPage schema with Person entity
  2. Dataset schema with explicit links to .md and .json versions
  3. FAQPage schema auto-generated from H3 questions in articles
  4. DefinedTerm schema for every tag
  5. CC-BY 4.0 license (explicit AI citation permission)
  6. Cite this protocol box with formatted citation + raw markdown endpoint

Plus: llms.txt manifest, robots.txt with AI crawlers whitelisted (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, Applebot), /api/protocols.json RAG endpoint.

Self-audit and iteration

After launch, a GEO audit of my own site found 4 critical bugs: llms.txt generated 404s (wrong slug field in Astro template), author schema said "jroh.cz" instead of "Jakub Roh", duplicate tags (ADHD/adhd), missing FAQPage schema.

Fixes within an hour, then adding methodology notes to all articles and writing 33 tag explainers (~70K words of new content).

Average citability score: 42 → 48-49, B-grade passages 1 → 4-5 per article.

Results

  • 16 protocols + 33 tag landing pages = 117 static pages
  • ~110K words of GEO-optimized evidence-based content
  • 6 layers of structured data, functional llms.txt, auto-FAQ schema
  • Cost: $0 infrastructure (Vercel free tier, Namecheap domain for a few dollars/year)
  • Time: 3 days from idea to production

What this demonstrates

Most Czech companies have a blog without FAQPage schema, without Author entity, without .md endpoints, without llms.txt — and ask why ChatGPT doesn't cite them. The answer: because they haven't done anything to allow it.

CiteThis is proof that GEO-first architecture can be built fast and without budget — if you know what you're doing. And the detailed article about building it is here.

Project: citethis.site · Stack: Astro, TypeScript, Tailwind · License: CC-BY 4.0 (content), MIT (code)