llms.txt + Agent Discovery: Making WordPress Readable to AI Systems
Search engines are no longer the only systems reading your site. AI assistants, agent frameworks, and crawlers increasingly look for structured signals that explain what your site is about, what content matters, and how machines can interact with it safely.
For WordPress sites in 2026, two pieces matter most: a clear llms.txt policy file, and proper agent discovery (agent card, api-catalog, and related well-known paths). Together they make your site readable to people and to AI systems without turning every page into noise.
Why This Matters
Discovery
Agents Need a Map
Without clear entry points, external systems guess from random pages. A map beats scraping.
Trust
Scope and Limits
Declare what you cover, what you do not, and where humans should go for real work.
SEO / E-E-A-T
Consistent Source of Truth
Stable facts, contact paths, and topic lists reinforce authority better than thin FAQ pages alone.
Control
Less Hallucinated Navigation
When agents have verified URLs and skills, they are less likely to invent fake /docs/ or /help/ paths.
Operations
Separates People from Bots
Humans use the site and chatbot. Agents use discovery files and a defined message endpoint.
Future-Proofing
Agent-Ready Architecture
As more tools support A2A-style discovery, sites that publish clear cards and policies are easier to cite.
What Belongs in llms.txt
Identity
Who You Are
Business name, location, what you actually do, and primary contact paths.
Content Map
Important Sections
Forum, key service pages, and high-value guides — not every archive URL.
Boundaries
What You Do Not Cover
Legal, financial, and medical advice exclusions. No invented services or pricing claims.
Good llms.txt content
Short site summary, preferred citation pages, forum index, contact URL, phone, and a note that technical answers should prefer published guides over guesses.
Avoid stuffing
Do not dump the entire sitemap, private admin paths, or marketing fluff. Agents need signal, not a second homepage essay.
Point to real URLs only
Every link in llms.txt should resolve. Broken or invented paths destroy trust faster than having no file at all.
Agent Discovery Layer
Card
/.well-known/agent-card.json
Describes the assistant name, skills, provider, and message endpoint.
Catalog
/.well-known/api-catalog
Machine-readable index of related services and discovery links.
Auth Notes
/auth.md
States whether agent endpoints require login and where humans authenticate.
Message endpoint
A public POST route (for example under your REST namespace) lets agents ask scoped questions without scraping the whole site.
Skills should be honest
List only topics you can support with real content: firm info, self-hosting, VoIP, security, forum guidance — not every keyword on the internet.
Rate limits and REST lockdown
Keep WordPress REST locked for anonymous users, whitelist only the agent routes you intend to expose, and share rate limits with the on-site chatbot.
Practical WordPress Setup
Step 1
Publish llms.txt
Serve it at the site root. Keep it short, accurate, and updated when major sections change.
Step 2
Expose discovery files
Agent card, api-catalog, and auth notes under well-known or documented public paths.
Step 3
Ground the assistant
Use site search/context so answers prefer real pages, posts, and forum topics over invention.
Step 4
Align crawl policy
Allow search/user agents that help with answers and citations; be selective with bulk training scrapers. Keep robots.txt, llms.txt, and well-known readable where appropriate.
Step 5
Link headers help
Optional Link headers and HTML alternates make the agent card easier to discover from normal page loads.
Step 6
Test like an agent
Fetch the card, post a simple message to the endpoint, and confirm answers only use verified internal URLs.
What Not to Do
Do not open all of REST
Whitelist only the agent namespace you need. Full anonymous REST access is unnecessary risk.
Do not invent documentation paths
If a guide does not exist, say so and point to the forum or contact page. Fake /kb/ URLs destroy credibility.
Do not treat agents as a content strategy replacement
Discovery files amplify good content. They cannot fix a thin site.
Readable to AI systems means: clear policy, honest skills, verified links, and a defined way to ask questions. The rest is still quality publishing.
Questions for the Community
Have you published an llms.txt yet?
What did you include, and what did you deliberately leave out?
Are you exposing an agent card or message endpoint?
Widget only, full A2A discovery, or still evaluating?
What broke first when you tested?
Crawl policy, REST lockdown, bad URLs, or rate limits?
Disclaimer
This content is for educational and informational purposes only. It is not technical or SEO advice. AI platforms and discovery conventions change quickly. Always test configurations in your own environment and treat citations and rankings as outcomes of overall site quality, not any single file or endpoint.
