Markdown twins are plain-text or markdown copies of an HTML page served at a parallel URL (typically {url}.md) so retrieval systems can ingest the content without an HTML parser — a practical convention endorsed by the llms.txt proposal but not yet formally consumed by any major answer engine.
The idea is simple: an HTML page carries navigation, ads, and layout that a retrieval crawler has to strip. A markdown twin is the same content minus the chrome, with the URL and provenance footer at the bottom. If a retrieval crawler prefers a clean copy, the twin is easier to parse; if it prefers HTML, the twin costs nothing.
The convention is inspired by llms.txt, which suggests llms.txt and llms-full.txt at the root plus optional per-page markdown files. No answer engine has publicly confirmed that it fetches twins, but the pattern has zero downside beyond disk space — the twin can be generated from the same source that renders the HTML page.
Two implementation notes. First, ship a provenance footer on every twin: canonical URL, author, published/modified dates, license. This gives any retrieval system a machine-readable attribution block. Second, do not put the twin in your sitemap — the HTML page is the canonical, and a duplicate entry in the sitemap muddles the URL structure the search engine sees.
The twin is insurance. It is not a substitute for server-rendered HTML, valid schema, or an IndexNow ping.