Question sensitivity in GEO measurement advice

Helen Bird documented the prompts, preserved the selected answers, and published the analysis.

This study tested whether asking how to measure AI visibility or how to validate a visibility score changes the advice independent builders receive.

The test used two intents with two wording variants each. OpenAI Responses API with web search and Grok provided three selected answers per wording, for 24 selected answers. The archive records 27 submissions, including three earlier consumer-interface failures or partial captures. The API condition and consumer ChatGPT are separate environments.

The coding separates tool suggestions, information-source categories, and citations. The study compares repetition and wording differences, while retaining missing and empty sets separately. Its small matched panels describe the tested prompts and do not estimate a general visibility rate. Exact questions, selected answers, coding, and the full submission history are available in the public record.