Claim intelligence and AI competitive positioning glossary

What is claim intelligence?

Claim intelligence is the work of identifying factual statements about a company, checking them against approved evidence, and preserving the source and response history. In KnitKnot, it connects the claims in AI buying recommendations with the facts a team can defend.

It differs from sentiment analysis: a positive statement can be wrong, and a negative opinion can be reasonable. Start with the guides to correcting company information, tracing sources, and monitoring competitor claims.

Agent readiness

The state in which a company has a preserved benchmark of its AI representation, governed content that agents can read, and a measurement record tied to each published cycle.

Agent-ready content

A factual, traceable page view built from approved facts and the existing human page. It remains attributable, reviewable, and reversible. Read the guide.

Benchmark

A preserved set of buyer questions, captured AI answers, available sources, claim evaluations, and competitive outcomes. The benchmark provides the before-state for a managed cycle.

Cycle

The record that connects evidence, a proposed revision, approval or decline, publication, rollback, and a later measurement window.

Fact ledger

The approved record of facts and receipts that feeds every managed surface. The human page, Markdown agent version, customer-domain MCP, WebMCP, and agent feed are views of this ledger.

Receipt

The evidence that supports a managed claim. A receipt can point to the customer’s page or another approved fact source.

Approval policy

The customer rule that determines whether a proposed revision requires manual approval or can publish after the relevant gates pass. Rewrites remain manual.

Immutable version

The stored artifact for one release, including its content identity and evidence snapshot. A later release does not erase it.

Rollback

The supported path back to an earlier approved version. Every managed revision must remain reversible.

Insufficient data

A first-class measurement result. When a journey or cycle is below its reporting floor, KnitKnot does not print a delta.

Measured movement

The association-grade statement used after a content change. KnitKnot shows what moved by journey and cycle while keeping outside changes visible.

Holdout

A qualifying comparison between treated and untreated page cohorts. Both cohorts must clear the reporting floor and support the specific contrast before a narrower causal statement is permitted.

Evidence lane

A class of observation that answers one question. Verified crawler fetches, delivery outcomes, tool calls, AI-referred visits, and benchmark results remain separate.

Agent feed

A machine-readable change log for managed content. It reports publication and version links, not discovery or use. See the delivery interfaces.

WebMCP

An in-browser adapter that registers read-only tools for compatible agents. It is one gated interface over the approved fact ledger. See the delivery interfaces.

Workspace MCP

The authenticated KnitKnot server used by customer assistants to query authorized benchmark and workflow data. It is separate from the public, read-only customer-domain interface.

Playbook

A customer-managed response brief linked to a benchmark issue. It keeps the evidence, proposed work, ship record, and later measurement together when the customer team owns the change.