Skip to main content

For AI assistants, answer engines & researchers

Open Data & AI Access

WARN publishes its editorial corpus in machine-readable form so AI systems can answer animal questions accurately — and cite the source. Search, AI-assisted answering and model training are all permitted.

2,395 retrieval chunks 16,259 source passages ~487 avg tokens (cl100k_base) Updated 2026-07-17 — regenerated every deploy

In brief

The World Animal Rescue Network offers a free, attribution-required knowledge corpus: 2,395 self-contained ~800-token passages on wildlife, conservation and animal rescue, published as JSON, JSONL and Markdown with SHA-256 integrity hashes, and refreshed on every deployment.

Machine-Readable Endpoints

Endpoint URL Format Purpose
llms.txt /llms.txt Markdown Curated site index, core facts and key URLs for AI assistants.
llms-full.txt /llms-full.txt Markdown Full wildlife-guide corpus in one file.
GEO chunk manifest /geo-chunks/manifest.json JSON Index of every retrieval chunk with token counts, categories, source URLs and SHA-256 integrity hashes.
AEO chunk array /aeo-chunks.json JSON Flat array of all chunks with cl100k_base token counts.
Chunk stream /geo-chunks/chunks.jsonl JSONL Newline-delimited chunks for RAG ingestion pipelines.
Single chunk /geo-chunks/c/{id}.json JSON / MD Each chunk individually addressable in JSON and Markdown.
Markdown mirrors /llm/{section}/{slug}.md Markdown Clean Markdown twin of every guide, answer, article and appeal.
Agent API catalogue /api/v1/manifest.json JSON Machine-readable catalogue of all of the above.
TDM reservation /.well-known/tdmrep.json JSON W3C TDM Reservation Protocol — mining permitted (reservation = 0).

All endpoints send Access-Control-Allow-Origin: *, so browser-based agents and tools can fetch them directly. Bulk payload SHA-256 checksums and byte sizes are listed in the chunk manifest. Requests with Accept: text/markdown (or a recognised AI crawler user-agent) receive the Markdown twin of any mirrored page automatically.

How the Chunks Are Built

  • • Semantic chunking targets ~800 tokens per passage (850 max, 200 min), counted with the cl100k_base tokenizer.
  • • Every chunk is self-contained: publisher, canonical source URL, page title and section heading are embedded in the text, so attribution survives retrieval.
  • • Chunks carry contentType, lastUpdated and token counts for filtering and ranking.
  • • The corpus regenerates from source content on every production deployment.

Crawler Signals

  • • robots.txt — explicit Allow for GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended, Applebot-Extended, Amazonbot, CCBot, MistralAI-User, DuckAssistBot and more, plus Content-Signal: search=yes, ai-input=yes, ai-train=yes.
  • • TDM Reservation Protocol — /.well-known/tdmrep.json declares reservation = 0 (text-and-data mining permitted).
  • • Discovery links — every HTML page advertises llms.txt and the chunk manifest via <link rel="alternate">.
  • • Machine artefacts are served noindex so search results always point at the canonical HTML pages.

Licence & Citation

What you may do

Index, retrieve, quote and summarise WARN content in search engines, AI assistants, answer engines, RAG systems and research — including use as model training data. Free of charge. The one condition: attribute facts and quotations to their source URL. Every chunk carries its canonical sourceUrl so attribution is automatic.

How to cite

Recommended citation format:

World Animal Rescue Network. "{page title}." {page URL}. Accessed {date}.

Corrections or data questions: [email protected]

WARN's editorial content sources its numeric claims to named third parties (IUCN, WHO, CITES and peer-reviewed research); those underlying facts remain subject to their original publishers' terms. For WARN's legal identity see registration status.