What is agent-ready content?

Agent-ready content is a factual, traceable version of a page that people and software agents can read. It starts from the company’s approved facts and existing human page. It does not hide the human page, invent a claim, or assume that a particular format will change an AI answer.

The content is one view of a fact ledger. A managed page can have a normal HTML view and a Markdown agent version built from the same approved record. Other interfaces can expose the same facts without creating separate sources of truth.

What makes the content ready for an agent?

The page states the subject, claim, evidence, and relevant date plainly. Its structure makes those facts easy to retrieve without relying on a screenshot, an interactive widget, or vague product language. Every managed claim traces to the customer’s page or to an approved fact with a receipt.

Readability is only one part of the standard. The publishing path also needs an approval policy, an immutable version record, and a rollback path. If a claim is revoked or refuted, it must be removed from the active content rather than covered with fallback copy.

Is Markdown enough?

No. A Markdown view can make a page easier for an agent to read, but serving Markdown does not prove that a provider used it in an answer. KnitKnot does not treat llms.txt as an inference contract and does not promise a citation from a format change.

The useful question is whether the full loop is controlled: what the benchmark recorded, which fact changed, which revision was published, and what moved in later journey-level measurement.

How does KnitKnot manage it?

KnitKnot preserves the before-state, proposes an evidence-backed revision, applies the customer’s approval policy, publishes an attributable version, and opens a measurement window. The cycle record keeps approvals, declines, receipts, and rollback information beside the revision.

The default outcome statement is simple: the content changed, then KnitKnot measured movement. Outside changes and thin samples remain visible. A qualifying holdout experiment is required before any narrower causal statement is allowed.

What interfaces can expose the same facts?

KnitKnot’s built interface set includes Markdown agent versions, a read-only /mcp surface on the customer’s domain, an in-browser WebMCP adapter, and a change feed. These interfaces are rolling out behind deployment gates. They are views of the same fact ledger, not separate content programs.

Next: how the agent feed works, what WebMCP does, or the measurement methodology.