Decision Guide
How Can You Verify a GEO, AEO or AI Visibility Claim?
Assess a GEO, AEO or AI visibility claim by its evidence, comparison basis and attribution. Separate an observed increase from proof that an intervention caused it.
This Decision Guide explains how to assess a provider’s performance claim. It applies the Reference and Method standards; it does not assess a named provider or redefine those standards.
Decision Guide · Applied Interpretation · v1.2 · Last reviewed 2026-09-25
Applies to: Public interpretation of observed AI/search readings
Status: Applied guide — uses Reference definitions, does not redefine them
Quick answer
A credible GEO, AEO or AI visibility claim identifies what changed, how it was measured and which evidence supports the comparison. Check the baseline, Reading Basis and missing observations before interpreting an increase. Then separate the observed change from any claim that an intervention caused it. Accept only the conclusion the evidence supports, with its scope, period and remaining uncertainty intact.
The core principle
An increase and an explanation of that increase are separate claims. A before-and-after comparison can support a finding that a defined measure changed. Attributing that change to a provider’s work requires additional evidence that addresses other plausible explanations. A broad promise such as “GEO worked” must first become a specific claim that can be checked.
The conclusion should be no broader than the evidence. These are tests of claims, not findings about any provider.
The reading grid
| Claim presented | Evidence to inspect | Conclusion boundary |
|---|---|---|
| “Visibility increased.” | The counted object, baseline, denominator and comparable observations. | An increase in that measure, on that basis; not all forms of visibility. |
| “The brand is cited more.” | Captured source attributions, entity-assignment rules and the citation denominator. | A citation difference; not automatically more mentions or recommendations. |
| “Our work caused the increase.” | The intervention record, timing and a comparison design that addresses competing explanations. | Attribution only to the extent supported; a sequence in time is insufficient. |
| “The brand now leads the market.” | A clear distinction between answer-surface position and evidence about the business market. | A surface result does not establish commercial leadership. |
| “The result is verified.” | Who checked which evidence, using what procedure and with what access. | Traceability, review and independent assurance are different claims. |
Make the claim specific. Identify the entity, the result claimed and the period. Ask what “visibility” counts: a mention, an attributed source, an opening role or a position are different observations. Ask whether the claimed change is an absolute count, a rate or a share, and what its denominator contains. A statement that a brand was named is not a statement that it was recommended.
Establish the baseline and Reading Basis. A baseline needs dated observations, not just a starting percentage. Identify the surface and mode, the question or query panel and its version, and the conditions behind both periods. Check whether the entity rules or denominator changed. Ask how missing captures, unavailable answers and repeated observations were handled. Apply the complete comparability test before treating two figures as a like-for-like increase. If only part of the evidence is comparable, narrow the comparison accordingly.
Follow the evidence behind the result. A chart or case-study summary is a claim presentation, not the full basis for checking it. Look for dated Observations, source references, relevant Capture provenance and a clear account of how the published measure was derived. Ask which supporting material can be inspected and which remains restricted. A Capture is not automatically public raw data. The Attestation Standard distinguishes traceable publication from public reproducibility or independent assurance. Restricted access limits what an outside reader can verify; it is not, by itself, evidence that a claim is false.
Separate the intervention from other changes. Establish what the provider changed and when. Then examine other changes that could affect the comparison: the query mix, measurement instrument, external answer service, the brand’s wider activity or competing sources. Repeated observations can show whether a pattern persisted; they do not automatically isolate its cause. Stronger attribution requires a comparison design that can address those alternatives and explain its remaining limits. If that evidence is absent, an observed increase after the work remains an association.
Keep the conclusion at the supported level. The evidence may support a scoped increase, a qualified comparison, or no reliable comparison at all. State which. Benchmark Governance explains why stable rules and disclosed changes matter. Business Market vs Positioning Surface sets a separate boundary: a result about mentions, citations or rankings is not evidence of revenue, market share or commercial leadership. A claim can be useful without proving every stronger claim attached to it.
Decision rule
Accept the narrowest performance claim that the available evidence supports. Assess causal attribution as a separate decision.
- The claim, measured object, baseline and period are explicit.
- The Reading Basis and missing observations permit the comparison being made.
- The published evidence and available access support the reported result.
- Any claim of causation addresses timing and competing explanations.
- The conclusion preserves its scope and limitations, including what remains unresolved.
If the basis cannot be compared, leave the increase unverified. If the increase is supported but attribution is not, accept the scoped observation and leave the causal claim unproven.
FAQ
- Is a before-and-after visibility chart enough to show that GEO worked?
- It can support an observed change if the measure and comparison basis are clear and compatible. It does not, by itself, show that the provider’s work caused the change. Attribution needs additional evidence addressing other changes during the period.
- Must a provider publish every prompt or raw capture for its claim to be credible?
- No. Evidence access can be limited by rights, confidentiality or proprietary methods. The provider still needs to state the basis of the claim, identify what can be checked and explain the limits of outside verification. Restricted access is not a substitute for supporting evidence.
- What should I conclude when an increase is supported but its cause is uncertain?
- Report the increase within its stated measure, scope and period. Keep the causal explanation separate and identify what remains unresolved. Do not turn an observed association into a claim of proven effectiveness.
Intelligence by AAI Kernel. Published by The Observatory.
The Observatory does not claim to measure real-world revenue, market share, or business leadership unless explicitly stated. It publishes observed AI and search positioning within a defined scope and time window.
Related references
- Reading Basis — because the meaning of a figure depends on its stated scope and observation basis..
- Questions of Method — comparability — because two figures can share a label while measuring different things..
- Attestation Standard — because evidence access determines what can be traced and checked..
- When to Trust AI Answers About a Market — because assessing an answer’s claims is distinct from evaluating a provider’s performance claim..
- Platform, Monitoring, Audit or Managed Service? — because the service being bought determines which deliverables and responsibilities to expect..