Write a falsifiable question
“Does clearer service information improve the accuracy of descriptions in our recorded responses?” is more useful than “Does GEO work?” The first names an intervention and an observable outcome. You can decide what would count as improvement, no change or a worse result.
Use a small prompt bank that reflects real user tasks. Include prompts where the brand currently appears and where it does not. Explain how you selected them so the report does not become a collection of favourable examples.
Example plan
This is a hypothetical educational exercise. Baseline: run a fixed set of service-selection prompts several times and record whether the response describes the service correctly. Intervention: clarify the service scope on one source page. Follow-up: repeat the same prompt set over a documented period and preserve the output.
Do not promise an immediate change. Publication, crawling, retrieval, product updates and ordinary response variation can all complicate the interpretation.
Keep an evidence log
- Research question and expected outcome
- Prompt set, selection method and run conditions
- Baseline outputs and scoring criteria
- Exact intervention and publication date
- Follow-up outputs, unchanged cases and adverse results
- Alternative explanations and the next test
Separate observation from causation
If descriptions become more accurate, report the observed change and the circumstances. Without a suitable control, you cannot be certain that your page edit caused it. If the result does not change, preserve that finding; it can still help the next researcher avoid a weak assumption.
Use the measurement guide for consistent scoring. Send a documented experiment to World GEO Association with the evidence you can share publicly and a clear statement of any restrictions.
References: The original GEO paper.