Method

Every claim traces back to raw model evidence.

A useful AI visibility result should survive a simple question: “Show me how you know.” Clarivy keeps the question, model context, answer, sources, and comparison rules together so your team can check the work.

Read full methodology Inspect sample deliverables

The short version. We ask a controlled set of buyer questions, record each answer with its measurement context, and compare results only when the conditions match. Facts are scored only against an approved, current source of truth.

Updated 24 August 2026.

How a Clarivy run becomes usable evidence

Define the questions before the run.
Buyer questions receive stable scenario and prompt identifiers so the panel cannot quietly change later. Scenario Panel v2
Record the answer and its context together.
Every observation identifies the surface, model, search state, question lineage, time, provenance, and raw evidence. Observation Contract v2
Keep different kinds of evidence separate.
The answer, native citations, model-emitted links, claim support, errors, and raw record remain distinct instead of being blended into one confidence label.
Check whether two runs are truly comparable.
A change is reported only when the question panel and observation conditions are compatible. Otherwise the delta is blocked or qualified.
Verify facts against approved sources.
If a source is missing, expired, conflicting, or unapproved, Clarivy returns “unavailable” instead of inventing an accuracy score. Fact Registry v1
Deliver evidence people and agents can both inspect.
Human and machine-readable files share stable IDs, while every downstream action still requires a current-evidence check.

Keep execution and acceptance separate

Clarivy does not ask leadership to treat the vendor's activity report as proof of improvement. The action record and the outcome test remain separate.

The customer keeps the acceptance rule.Questions, surfaces, device or account conditions, repetitions, and the meaning of “recommended” are agreed before implementation.
The receipt proves the action happened.It records approval, implementation, publication, independent checking, and rollback availability. It does not prove that the action caused an uplift.
The retest decides the next move.Compatible evidence supports a scale, revise, hold, or rollback decision. Missing and inconclusive outcomes stay visible.

Consumer-surface boundary. Customer-held blind questions and real-device acceptance apply only when consumer-surface verification is included in the approved scope. Model API, consumer Web, and native App results are separate cohorts and are never merged into one engine score.

Who Clarivy is—and is not—for

A good fit

You care about recommendation outcomes, not activity reports.
You will agree on a baseline and acceptance rule before implementation.
Your team can approve factual claims and provide official sources.
You want misses, uncertainty, and competitor displacement reported honestly.

Not a fit

You require a guarantee that one model will recommend the brand in one specific answer.
You want fabricated reviews, disguised endorsements, fake sources, or platform-incompatible automation.
You want only successful screenshots reported or expect test rules to change after the work begins.