CiteSig

The problem

The internet has no machine-readable way to distinguish a verified fact from a plausible-sounding sentence. HTML has no <verified> tag. Search engines guess. Language models hallucinate. Regulators are starting to require provenance for AI outputs, and no standard yet exists to provide it.

CiteSig is the missing layer: a small, open protocol that lets publishers, journalists, researchers, and AI systems attach a signed reference to any factual claim — and lets browsers, verifiers, and downstream tools check that reference at rest and at render time.

Example — inline claim

<!-- A claim, without CiteSig -->
<p>The Great Wall of China is visible from the Moon.</p>

<!-- A claim, with CiteSig -->
<p data-citesig="cs:v1:9f2a...c8e1"
   data-verdict="refuted"
   data-confidence="0.02">
  The Great Wall of China is visible from the Moon.
</p>

The data-citesig attribute references a canonical claim record. Any verifier — a browser extension, an LLM output filter, an editorial workflow — resolves it against a public registry and returns a signed verdict.

Design principles

• Neutral standard, not a product. The spec, the reference implementations, and the registry contract are open. Anyone can run a verifier node.
• Cryptographically grounded. Ed25519 signatures. Content-addressed claim IDs. Independently verifiable without trusting a central authority.
• Human-readable, machine-checkable. Works in HTML, JSON, Markdown, plain text, and streaming LLM outputs. Degrades gracefully when unsupported.
• Composable with existing standards. Complements C2PA (media provenance), schema.org/ClaimReview, and W3C DIDs. Fills the missing text-claim slot.

Status

Get a single email when the draft spec is published. No newsletter, no marketing.

Thanks — you'll hear from us when v0.1 lands.