AI visibility claims become unreliable when screenshots, implementation activity, business outcomes, and independent validation are mixed into one story. Undercover.co.id separates these evidence layers so that a reader can see what was observed, what was implemented, what changed, and what remains unknown.
Evidence layers
Observed evidence
A time-bound record of an AI output for an exact query, provider, mode, date, language, location context, session, and retrieval condition. It records whether the brand was mentioned, how it was described, which sources were used, and what limitations apply.
Implementation evidence
A record that a specific page, schema graph, relationship, source correction, evidence route, or governance control was implemented. Implementation evidence proves the work occurred, not that an AI system will respond in a particular way.
Outcome evidence
A comparable result measured against a baseline using a sufficiently consistent query set and methodology. Outcome evidence requires clear periods, denominators, failure handling, and limitations.
Independent validation
Evidence from a third party that is independent of Undercover.co.id and the subject organization. Independence, publication context, date, and claim relevance must be stated.
Methodology evidence
Documentation showing how queries, observations, classifications, scoring, source handling, and validation are performed.
Mandatory observation metadata
- Observation ID and exact query
- Query type and decision stage
- Engine, surface, model, or visible mode
- Date, time, timezone, language, and location context
- Session type and browsing or retrieval status
- Brand mention, description accuracy, and category accuracy
- Competitor and recommendation context
- Citation and owned-source status
- Raw reference, reviewer, interpretation, confidence, and limitation
- Observation status, including provider or retrieval failure
Case-study classification
Every public case study should identify its relationship status, such as verified client implementation, anonymized client implementation, Undercover internal implementation, public brand observation, hypothetical scenario, or methodology example. A public brand observation must never imply that the brand hired Undercover.co.id.
Claims we reject
- Permanent AI rankings
- Guaranteed recommendation
- Universal visibility across users and models
- Revenue attribution without comparable business data
- Provider failure counted as brand absence
- A favorable screenshot presented as a complete audit
Validation standard
For managed WordPress pages, evidence and visible claims must agree with the machine-readable graph. The rendered source must contain exactly one valid invisible JSON-LD graph, and every important relationship must identify its source, destination, anchor, insertion point, direction, and validation status.