Buy an evidence process, not a promise
A credible GEO provider should be able to explain what it will observe, what it can change and which outcomes remain controlled by third-party systems. “Guaranteed citations” and unexplained visibility scores remove the very uncertainty the engagement is supposed to measure.
Five questions for the first call
- What is the unit of measurement? Ask whether results refer to questions, answers, citations, source URLs, sessions or leads.
- How is the query set frozen? Require markets, languages, platforms, dates and inclusion rules.
- What evidence accompanies a finding? A recommendation should point to a captured response, URL, log event or first-party record.
- What work is excluded? Separate audit, implementation, authority work and analytics.
- How are limitations reported? The provider should describe variability, missing referrers, anonymization and sample size.
Use a buyer scorecard
| Area | Strong answer | Warning sign |
|---|---|---|
| Method | Fixed, repeatable and documented. | “Our proprietary AI does everything.” |
| Metrics | Counts and denominators are visible. | A composite score with no formula. |
| Cases | Scope, period, evidence and limits are stated. | Logos and percentages without attribution. |
| Technical work | Changes are reviewed against production URLs. | Schema or files filled with hidden claims. |
| Commercial link | Referral and lead rules use first-party data. | Crawler requests presented as prospects. |
| Ownership | The client retains pages, data and documentation. | Access or evidence disappears after cancellation. |
Ask to inspect a sample deliverable
A useful sample does not need client secrets. It can show the query register, capture fields, page-level finding, severity rule, implementation note and limitations. This reveals more about delivery quality than a polished dashboard screenshot.
Match the engagement to the decision
Use a diagnostic when the team does not know where visibility breaks. Use a focused pilot when one service cluster is ready for implementation. Use ongoing monitoring only when there is enough search, content and commercial activity to justify repeated measurement.
The provider can guarantee the agreed work, documentation, QA and evidence pack. It cannot guarantee that an external answer system will mention or cite a page.
Before signing
- Named deliverables and page/query limits.
- Access requirements and data responsibilities.
- Approval path for public claims and schema.
- Measurement dates and comparison rules.
- Change ownership, handover and cancellation terms.
- Privacy treatment for prompts, analytics and lead records.
Editorial reference
Google's people-first content guidance recommends original value, clear authorship, demonstrable expertise and a satisfying answer for the intended audience. Those are useful procurement tests for a provider's own website as well as for its client work.
Where to go next
Use the Nexus AI Visibility Framework to place this topic inside a complete measurement system, or review the AI Visibility Audit scope.
13 Jul 2026 — Expanded with an independent structure, examples, implementation guidance and primary references.