Audit answer visibility

Audit Answer Visibility

Produce an evidence audit, not an invented probability of being cited. Separate live observation, inference, and recommendation.

Inputs

Require:

• target URLs or files;
• target audience, questions, locale, and competitors;
• observation date;
• access and authentication constraints;
• claim and source policy.

Use current authoritative sources when platform behavior may have changed.

Workflow

1. Record the exact surface, date, locale, and access state.
2. Check response status, canonical, robots, structured data, heading hierarchy, answer blocks, and cited sources.
3. Test target questions only on surfaces the user authorized.
4. Record what was directly observed without interpreting it.
5. Add inference with confidence and alternative explanations.
6. Add recommendations linked to observed gaps.
7. Record inaccessible or unverified surfaces as unknown.
8. If using a diagnostic score, expose weights and label the score as a heuristic.

Boundaries

• Let the model organize questions, gaps, and recommendations.
• Use deterministic checks for HTTP and markup facts where available.
• Require a person to interpret competitive importance and approve investment.
• Do not claim ranking, citation, or answer inclusion without a dated observation.
• Do not treat crawler accessibility as proof that an answer system used the page.
• Do not present a proprietary score as a calibrated probability.

Outputs

Create:

• visibility-observations.jsonl
• citation-gaps.md
• technical-findings.md
• recommendations.md

Read references/output-contract.md before writing them.

Stop conditions

Stop when the target cannot be accessed, observation context is missing, a requested claim needs unavailable live verification, or a source cannot be attributed.

Verification

Confirm every finding has one of these states:

• observed
• reported
• inferred
• recommended
• unknown

Rerun time-sensitive observations before publication.