01
Human pages stay canonical
The public page is still the source of meaning. Machine surfaces mirror it; they do not replace it.
PointCast renders every Block at /b/{id} for people and /b/{id}.json for tools.
SEO · GEO · LLM DISCOVERY
Agent-native publishing.
A website is agent-native when its human pages and machine surfaces are designed as one system: readable HTML, stable JSON, discovery manifests, feeds, sitemaps, curated LLM context, and citation-ready permalinks.
Short Definition
Agent-native publishing is not SEO with extra meta tags.
It is a publishing contract. Humans get a designed page. Agents get structured retrieval paths. Search engines and LLMs get consistent entity identity, canonical URLs, and a clear answer to “what should I cite?” The page still matters, but the page is no longer the only interface.
The Pattern
Six practical rules.
02
Discovery is explicit
Agents should not infer your API map from navigation. Give them one manifest with every useful endpoint.
/agents.json lists feeds, contracts, schemas, CORS policy, citation format, and agent-mode behavior.
03
Content has stable IDs
Citations need permanent addresses. If a thing can be cited, it should have an immutable URL and a machine mirror.
A PointCast Block keeps its zero-padded ID forever: /b/0205 and /b/0205.json.
04
Feeds are first-class
Use standards where they exist. JSON Feed, RSS, and sitemaps make indexing and syndication boring in the best way.
/feed.json and /feed.xml cover the full PointCast archive, while /c/{channel}.rss slices it by channel.
05
LLM context is curated
A model needs a short answer, a deeper brief, and a retrieval order. Do not make it summarize the whole site cold.
/llms.txt is the short orientation; /llms-full.txt is the expanded version.
06
Citations are part of the contract
Tell agents exactly how to cite your work and which URL should be treated as evidence.
PointCast asks models to cite the Block permalink, not scraped fragments.
Reference Implementation
How PointCast does it.
PointCast treats every publishable thing as a Block: a stable JSON object with a permanent ID, channel, type, title, timestamp, and optional media or external context. The human page is the editorial surface. The JSON mirror is the retrieval surface. The manifest tells agents where everything lives.
- /agents.json canonical machine discovery manifest
- /llms.txt short LLM orientation and retrieval order
- /llms-full.txt long-form model context
- /blocks.json native full archive
- /b/{id}.json citation-level evidence for one Block
- /feed.json JSON Feed v1.1 standards feed
- /feed.xml RSS 2.0 standards feed
- /sitemap-discovery.xml priority discovery sitemap
- /BLOCKS.md public schema documentation
Reusable Brief
Copy this pattern.
1. Keep canonical human HTML pages.
2. Add JSON mirrors for every citeable object.
3. Publish /agents.json with endpoints, schemas, feeds, and citation rules.
4. Publish /llms.txt for the short model-facing summary.
5. Publish RSS, JSON Feed, and sitemaps.
6. Give every item a stable permalink and preferred citation format.
7. Monitor crawler and LLM-agent requests separately from human visits.Next Reads
