Build notes3 sections · 2 min readBy Dezső Mező · Published October 3, 2026.mdAEOAI SEOllms.txtStructured data
Search engines index documents; answer engines lift claims. The difference changes what you publish: a definitional paragraph that stands alone, a list of steps, an honest FAQ — and machine-readable twins of all of it so a crawler that prefers plain text does not have to parse a canvas. This is a description of the actual setup on this site, including the parts we are not sure work yet.
01Proven: the structured-data graph
JSON-LD restating what every page says — articles with authors and dates, a defined-term set for the glossary, FAQ and HowTo where the page genuinely is one, LocalBusiness identity tying the studio to a place. Honest markup earns rich results and entity reconciliation; markup that says more than the page earns neither.
02Probably useful: markdown twins
Every route answers .md — same content, plain markdown — for the growing set of tools that read text better than DOM. Cheap because it is generated from the same content files, which is also what keeps it honest: there is no second source to drift.
03A bet: llms.txt
A curated index of the site's best content in the format models prefer — adopted early by docs-heavy sites, unproven in actual crawler behavior. It costs one file and the same effort produced the twins and the schema, which definitely get read; the bet is cheap enough to place.
What to take away
- Structured data restates the page — never argues with it.
- Markdown twins are generated from the same source, so they cannot drift.
- llms.txt is a cheap bet, not a proven channel.
- Write claims that stand alone — answer engines lift sentences, not pages.
More from the lab
Browse all entries
The AI-search visibility checklist we run on this site
Structured data an answer engine can use
The EU accessibility deadline passed — what still applies
Core Web Vitals on a real Next.js site
Want this looked at on your own system?Start a conversation
