Methodology and evidence standards
A finding should survive the customer’s product expert.
Evidiam uses a conservative evidence standard designed to reject ambiguous product claims, sibling-model transfers, stale authority, and circular AI validation.
Observation
Captured buyer answer + test conditionsControl
Exact-SKU authoritative recordGates
Identity · authority · materiality · repeatabilityOutput
Reviewed finding
The six-phase method
Truth first. Testing second.
The controlling product record is established independently before buyer-facing AI answers are evaluated. The tested system is not given the answer or source it is expected to find.
01
Scope
Select the products, markets, buyer journeys, competitors, platforms, and material claim classes in scope.
02
Establish truth
Build authoritative exact-SKU records, preserving source hierarchy, effective dates, and configuration boundaries.
03
Test
Run natural buyer questions without revealing expected values, while recording the interface, web-search state, prompt, date, and other observable conditions.
04
Classify
Separate correct answers, cautious answers, unsupported claims, source conflicts, verified AI errors, and visibility gaps.
05
Review
Apply exact-model, current-authority, materiality, and repeatability gates before a finding is customer-facing.
06
Retest
Repeat the same buyer journeys after corrections and monitor future changes in answers and sources.
Current test coverage
Know which answers we actually tested.
Customer reviews currently use the OpenAI and Anthropic APIs. We save the question, returned answer, available citations, model, and test conditions. API results can differ from the ChatGPT and Claude consumer apps. Google AI, Perplexity, and Copilot are not included in this automated review.
Potential issues receive independent AI verification against official evidence before publication. This is not a promise of human review. A missing or incomplete provider response is a coverage limitation, not evidence of a visibility gap.
A complimentary review samples up to three offerings with one buyer question each. It is not a full catalogue audit. You see where you appeared or were absent in captured answers; a review may find no verified issues. Repeating a question later shows what changed under those conditions, not why it changed or how much revenue was affected.
Repeatability standard
One captured answer is an observation, not a market claim.
AI answers can vary by wording, interface, retrieval mode, session, location, and model update. Evidiam escalates language only as the evidence becomes more repeatable.
- 01
Observed
A response was captured once under documented conditions. It is retained for review, not presented as representative.
- 02
Reproduced
The same material outcome appeared again under the same defined test conditions.
- 03
Cross-prompt
The outcome persisted across more than one realistic buyer phrasing without leading the tested system.
- 04
Cross-platform
The same issue or visibility pattern appeared on more than one buyer-facing AI platform.
- 05
Verified material finding
Exact-SKU authority, materiality, captured evidence, and the applicable repeatability gates all pass review.
The test record
Enough context to understand what was actually observed.
Each retained result connects the buyer question to the environment in which the answer appeared.
- Exact prompt and buyer-journey purpose
- AI platform, interface, retrieval or web-search state, and exposed model information
- Date, market or region when observable, response text, citations, and source URLs
- Product identity, controlling evidence, materiality decision, reviewer, and retest history
Finding classification
Precise language protects credibility.
Not every discrepancy is an AI error, and not every absence is a visibility failure. Each result is assigned the narrowest defensible classification.
| Classification | What it means | Customer action |
|---|---|---|
| Verified AI error | A captured affirmative claim contradicts controlling exact-SKU evidence. | Correct the source ecosystem and monitor the answer. |
| Source-data conflict | Two buyer-visible sources publish incompatible product claims. | Reconcile and remove ambiguity. |
| Visibility gap | A qualified product is absent from a defined buyer journey. | Improve product coverage and supporting evidence. |
| Citation-control gap | The answer relies on stale, indirect, or third-party sources. | Strengthen authoritative retrieval paths. |
| Insufficient evidence | The available record does not justify a definitive conclusion. | Escalate to the manufacturer or exclude. |
What we do not claim
Optimization without mythology.
AI answers are variable and no vendor controls every response. Evidiam measures observable outcomes, improves the information environment, and reports what changed.
- We do not describe a public-source discrepancy as an AI error without a captured response.
- We do not present a one-time answer as a repeatable pattern.
- We do not use facts from one model, sibling product, or configuration to correct another.
- We do not infer certification, safety, or compatibility from the absence of evidence.
- We do not promise a guaranteed ranking or recommendation from an AI platform.
See the answer your buyers get.
Send your website. We test three real buyer questions and show where your company or products disappear, drift, or lose control of the source.