# AI discoverability

> Make a site readable and citable by AI assistants and agents — markdown versions of pages, an llms.txt index, plain-language summaries, and crawl permissions that say yes.

Source: https://gaitro.com/skills/ai-discoverability

Assistants and agents read the web differently from people: they prefer clean text, clear structure, and an index they can start from. A few cheap moves make your site first-class for them.

## Offer a markdown version of every page

Serve each important page as plain markdown at the same address plus `.md` (`/docs/x` → `/docs/x.md`). Markdown is what an agent would rather read than parse HTML. Keep the two in sync from one source.

## Publish an llms.txt

At `/llms.txt`, a short index: what the site is in a sentence, then a linked list of the pages worth reading, each with a one-line description (point at the `.md` versions). Add `/llms-full.txt` with everything concatenated for assistants that want it all at once. See llmstxt.org for the convention.

## Lead with plain-language answers

State what the thing is, and answer the common questions, in plain words near the top — the same text your structured data repeats. Assistants quote the clear sentence, not the clever one.

## Say yes to AI crawling (if you want to be cited)

In robots.txt, allow the AI crawlers and set a content signal that permits AI input/training if that's your choice. If you don't allow it, don't expect to be cited.

## Keep facts machine-checkable

Structured data (JSON-LD) and consistent, specific wording help an assistant get your facts right rather than guessing. Contradictions between pages become wrong answers.

## Verify

Fetch `/llms.txt`, a page's `.md`, and robots.txt from the live site and confirm they're correct and in sync — this is the part that silently drifts.