Agent Ready HTML

How docs sites get cited

In short

In the Vercel and Ora data, 82.5 percent of answers traced to a fetched page, 47 percent of those to a docs page and 14 percent to llms.txt. Docs get cited when the answer is in the first screen of the page, the code can be copied verbatim, and the page is reachable by a link.

Citation here means the agent's answer traced to a page it fetched. The data says docs pages are the main grounding source, and the guide below turns that into page-level rules.

What the evidence says

  • Of 1,023 runs with an answer-provenance result, 844 (82.5 percent) traced to a fetched page. [primary source] source, read 4 November 2026.
  • Of the 844 grounded answers, 394 (47 percent) traced to a docs page and 120 (14 percent) to llms.txt. [primary source] source, read 4 November 2026.
  • Of 119 runs whose answer included code, 73 (61 percent) contained lines copied verbatim from a fetched page. [primary source] source, read 4 November 2026.
  • Ora's lab found (2026-10-19) that answers written without reading the site were 3.7 times as likely to miss every fact asked for (25 percent against 7 percent). [vendor claim, 2026-10-19] source, read 4 November 2026.
  • Six sites account for 691 of the 1,033 runs (66.9 percent), so the figures describe that sample. [primary source] source, read 4 November 2026.
  • GitBook reports 51.8 percent of intentional documentation reads on GitBook are AI agents. [vendor claim] source, read 4 November 2026.

Steps

  1. Put the answer to the page's question in the first 200 words, in the initial HTML.
  2. Write code samples as fenced blocks with a language tag. In the data, 61 percent of answers that included code copied lines verbatim from a fetched page.
  3. Link each docs page from the docs index and from llms.txt so there is always a path to it.
  4. Give every page a distinct title and H1; an agent reading the top of a page should know what it answers.
  5. Read your own pages in the viewer, one at a time, and fix any that show an empty left pane or no H1.

Check the result

The test is the viewer on a docs page: H1 found, first 200 words found, and the fenced code intact in the left pane. Open the viewer to run it on your own URL.

Related: all guides, initial HTML, llms.txt.

Sources

  1. Vercel and Ora: Make your site readable by AI agents https://vercel.com/kb/guide/make-your-site-readable-by-ai-agents, read 4 November 2026
  2. Ora: AgentReady launch post, findings 01 to 04 https://ora.ai/blog/agentready-launch, read 4 November 2026
  3. GitBook agent-readiness checker https://gitbook.com/agent-score, read 4 November 2026

Last reviewed 4 November 2026