Immutable methodology · ai-content-rubric-v1

FutureContent AI Content

AI Content asks whether an AI reader can identify the page's primary information need or task, extract a supported answer, summarize it without distortion, answer relevant questions with evidence, and assess observable accountability signals without inventing trust claims.

Pinned identities

Evaluation Version
page-scan-2026-08-18-v16
Manifest
assessment-e79ae98ad2b433fc
Rubric
ai-content-rubric-v1
Specification
page-scan-v1-specification-v1
Ownership Registry
page-scan-v1-ownership-v1
Provider Registry
page-scan-v1-provider-registry-v2
Weights
page-scan-157-check-scoring-v1
Prompt / schema
page-understanding-prompt-v2 / page-understanding-schema-v2

The stable AI Content methodology page points here while this is the active version.

Supported page types and applicability

Semantic scoring supports articles/news, documentation, product/service pages, landing pages, and policies/legal pages. Profiles, directories, forums, user-generated feeds, search results, dashboards, and unvalidated types receive applicable deterministic checks while semantic checks are unavailable.

Ambiguous page type, purpose, or applicability preserves competing interpretations and bounded evidence but makes affected checks unavailable. A missing-information penalty requires an explicit page-type or claim-type expectation; a generic preference for more content does not affect scoring.

Shared semantic artifacts

One bounded page-understanding run produces neutral artifacts for SEO and AI Content: page type, audience, primary information need or task, direct answer or offered action; a three-bullet evidence-backed summary; representative direct, composite, and relevant-unanswerable Q&A; a claim-and-evidence map; observable credibility signals; and contradiction, omission, terminology, context, attribution, quantitative-integrity, and temporal observations.

Artifacts validate independently. An invalid Q&A set does not erase a valid summary or claim map. Relevant unanswerable questions require explicit abstention rather than an invented answer. The default report shows the primary information need and three-bullet summary; deeper artifacts remain expandable.

Ownership and scoring

Deterministic rubrics—not the model—assign applicability, Defect Family, severity, and arithmetic. Summary, Q&A, claim, credibility, contradiction, and omission views cannot multiply one defect's score effect. Credibility describes observable signals as strong, partial, or limited with evidence and limitations; it never declares a person or organization trustworthy or untrustworthy.

ai_content.actionability
8 points maximum
ai_content.answerability
15 points maximum
ai_content.claim_support
20 points maximum
ai_content.completeness_and_qualifications
15 points maximum
ai_content.comprehension
15 points maximum
ai_content.integrity_and_scope
7 points maximum
ai_content.observable_credibility
5 points maximum
ai_content.summary_fidelity
15 points maximum

These configured maximum deductions were validated by the approved fifty-page V1 benchmark. Interpretation stability and instructional-content safety are visible non-scoring diagnostics; v1 runs the evaluator once and makes no stability claim.

AI Content defect families (8)

  1. Primary information need and comprehensibility

    Primary need/task, direct answer/action, terminology, context, and bounded supporting excerpts.

  2. Evidence-backed summary fidelity

    Three-bullet summary mapped to page evidence, qualifications, attribution, and temporal scope.

  3. Answerability and explicit abstention

    Representative direct, composite, and relevant-unanswerable Q&A with evidence links.

  4. Claim support, relevance, and attribution

    Claim-and-evidence map with claim type, attribution, qualification, scope, and support status.

  5. Answer completeness and qualifications

    Page-type and claim-type expectations, omissions, contradictions, and attached qualifications.

  6. Actionability and offered next step

    Primary task, offered action, prerequisites, constraints, and stated next step.

  7. Observable source-use and accountability signals

    Identity, expertise/review, evidence, editorial accountability, disclosure, and maintenance signals.

  8. Contradiction, quantitative, and temporal integrity

    Contradiction, terminology, context, attribution, quantitative-integrity, and temporal-applicability observations.

Evidence and retention

Only the application-owned normalized bounded page document enters the semantic evaluator. Retained evidence is capped at 240 characters per excerpt, three references per item, and 12,000 aggregate excerpt characters per scan after deduplication. Higher-severity and score-affecting evidence is preserved first; omissions caused by the cap are disclosed.

Copied excerpts expire after 90 days or earlier with the Page Scan or Workspace. Source field and location remain after excerpt expiry. Full raw HTML, the full normalized page, unrestricted prompts, raw model/provider responses, credentials, and unbounded diagnostics are prohibited retained data.

Results and limitations

AI Content contributes 25% to a complete Page Optimization Score. Unavailable checks create a non-reweighted Assessment Score Range; an unavailable area withholds the overall score. Moderation that prevents semantic analysis makes SEO and AI Content unavailable while deterministic areas continue.

This single-page assessment does not guarantee inclusion, citation, recommendation, ranking, factual truth, reputation, trustworthiness, or end-to-end agent completion. External citation checks establish bounded technical resolvability only.