Discovery and documentation
Help agents find and understand your product
For teams whose website, docs or sign-up an AI agent has to find and read before it can help anyone.
Before an agent can do anything with your product, it has to find the right page and make sense of what it says. This group covers that first stretch: websites and documentation an agent can read, how AX sits next to UX, DX and GEO, what llms.txt does and doesn't do, and the newer files that tell an agent what a domain offers. The last guide is about letting an agent start a task before anyone has signed up. Start here if agents keep landing on the wrong page or guessing at your docs.
The guides, in order
Build websites agents can understand
Clear structure, stable interactions, and meaningful labels help both people and agents navigate a site.
11 min read
Write agent-readable documentation
What AI agents read, how to shape a page around a single decision, and how to keep the text true as the product moves.
8 min read
AX, UX, DX, and GEO: where each fits
How the related disciplines connect, and the different product problem each one describes.
14 min read
Does llms.txt actually work?
What adoption and server-log studies show, and where the evidence is limited.
11 min read
The state of agent-web standards
Compare agent-facing standards by who reads them and what they help an agent do.
15 min read
Agentic Resource Discovery, implemented
What ARD is, how ard.json relates to MCP's ai-catalog.json, and a copy-pasteable manifest for a content site.
19 min read
Let agents start before anyone signs up
Claim-later onboarding: the agent provisions a bounded resource now, and a person claims it later through a link.
17 min read
Rules on these topics
Claimed URL is broken
A URL in the file returned 404 or 410, or its host did not resolve.
Claimed CLI command does not exist
The file shows a subcommand that the CLI's --help does not list.
Claimed CLI flag does not exist
The file uses a flag that neither the CLI's --help nor the subcommand's --help documents.
Linked local file is missing
A relative link or manifest path points at a file that does not exist.
Claimed MCP tool is not served
The file names an MCP tool that the live server's tools/list does not return.
Help is not reachable
None of --help, -h or help printed usage text with exit status 0.
Evidence on these topics
Vendor claim: agents never invoked a docs skill in 56% of eval cases
Vercel states that in its Next.js 16 evals the docs skill was never invoked in 56% of cases, so the skill matched the 53% pass rate of no docs, while an 8KB docs index in AGENTS.md reached 100%.
Coding agents mostly read instruction files, not docs sites
An observational study of 557 agentic coding sessions (preprint) found that instruction files and working notes made up 60.5% of agents' documentation interactions, against 10.6% for classical technical documentation and 1.3% for API references.
A capable agent skipped the index and guessed the page
A preregistered ablation on a 709-page Markdown wiki (preprint) found that a capable tool-using agent never loaded the compact catalogue index, inferring page paths from the question instead, while retrieval-based access kept answer quality non-inferior and cut cost by about a third to over half.
Adding the context7 MCP server gave no lift; its tools went unused
Microsoft reports that adding the context7 MCP server to an anti-hallucination skill gave no meaningful lift on an SPFx upgrade: its tools did not load in 3 of 5 runs and were not called in the other 2, while telling the agent to use CLI for Microsoft 365 raised configuration correctness from 30/80 to 75/80.
Agent skills recommended a package that does not exist
Merged pull requests in Vercel's agent plugin repository corrected skill instructions that recommended an npm package that is not published, and plugin guidance that advertised deployment cards the production MCP server does not expose.
Skill selection precision collapses as the skill pool grows
A preprint reports that as the pool of available skills grew from 5 to 100, the precision with which agents actually used the right skill fell from 29.6% to 3.3%.