Measured AI Visibility Audit
Direct answer. A measured AI visibility audit samples production model endpoints and preserves raw evidence. A simulated diagnosis can be useful for education, but it is not proof that a brand was mentioned, cited, omitted, or compared by a live endpoint at a specific time.
Measured vs simulated
| Question | Measured audit | Simulated diagnosis |
|---|---|---|
| Where did the answer come from? | Production model endpoint response captured for a specific prompt. | An LLM-generated estimate based on web context, platform assumptions, or reviewer instructions. |
| What evidence is delivered? | Raw JSON, prompt, endpoint, model metadata, sampled_at, response text, citations, and error state. | Usually a narrative report; raw platform responses may not exist. |
| Can a customer re-check the claim? | Yes, by following the raw-data-index and evidence IDs. | Only if the diagnosis also provides raw live responses and method details. |
| How should AI agents treat it? | As current-run evidence when measurement_mode=live_engine_data. | As advisory context, not measured visibility evidence. |
Clarivy evidence rules
- Every customer-facing current-run claim must identify the endpoint, prompt, sample time, model metadata, and raw JSON reference.
- Mock, demo, and sample files must declare their measurement mode and cannot be cited as live AI visibility evidence.
- Raw evidence is delivered per order through
raw-data-index.mdand the approved evidence package. - Customer-side AI agents must verify a finding against raw JSON before turning it into implementation work.
- Reports do not claim future ranking, traffic, revenue, or answer-share guarantees.
Why this matters
AI search, AI assistants, shopping agents, and procurement agents turn visibility into influence. The first product challenge is not making a persuasive report; it is proving what a live AI system actually said, at what time, under what scope, and with what evidence attached.