Independently test the claims behind AI marketing performance.

The party selling the intervention should not be the only party proving it worked.

AI-centric marketing creates a measurement conflict when agencies report their own performance, platforms report their own delivery, GEO vendors define their own scores and PR programs select favorable evidence. Verify answers a narrower question: what actually changed, how was it measured, and does the evidence support the claim?

1. Scope before conclusion.
2. Observed, modeled and independently verified evidence stay distinct.
3. Methodology is versioned and conclusions can be supported, partially supported, unsupported, inconclusive or not measurable.
4. Correlation is not automatically presented as causation.
5. If InnMedia also executed Ads or Distribution, that relationship is disclosed and permissions/evidence remain separated.

Audit products

Audit workflow

1. Scope.
   Define the claim, owner, campaign/activity, baseline/evaluation periods, markets, platforms, metrics, sources, confidence needs and exclusions.
2. Acquire evidence.
   Use SIGNAL Intelligence, Analytics, customer/platform exports, campaign logs, publisher evidence, CRM and crawler tests as authorized.
3. Normalize.
   Align dates, scopes, identities, metrics and missing data.
4. Measure.
   Run the approved methodology pinned to a version.
5. Review.
   Test sensitivity, confounders, discrepancies and evidence completeness.
6. Conclude.
   Assign claim support status and quantify change/uncertainty when the evidence permits.
7. Package evidence.
   Deliver definitions, methodology, limitations and references with the result.

Claimed vs measured is the core visual language.

Vendor claim: +300% AI visibility
Verify finding: +17.4% measured uplift
Status: Partially supported
Confidence / uncertainty: defined in methodology
Primary issue: denominator and competitor set changed

The values above are demo data, not customer results. The important distinction is that the vendor claim and the independently measured finding are retained separately.

A serious audit ends with an evidence package.

The package can include an executive summary, audit question, scope, data sources, evidence class, methodology version, exclusions, baseline, findings, uncertainty, support status, limitations, appendices and a machine-readable result where contracted.

Every Verify engagement links to the public Methodology and Metrics standards. Verify should not sell a black-box proprietary trust score with no interpretable method.

Independence is architectural, not just branding.

• Separate roles and permissions for activation and verification.
• Immutable evidence snapshot for the final report.
• Ads/Distribution cannot edit audit findings.
• Methodology lock and report versioning.
• Conflict disclosure and reviewer workflow for enterprise engagements.
• Inconclusive is a valid outcome when evidence is insufficient.

Use SIGNAL evidence without turning the audit into self-reporting.