Tool behind the study
Query Fan-out Explorer
I built Query Fan-out Explorer to record the searches Gemini reports and the sources it cites with Google Search grounding. I used saved runs to compare query wording, source overlap and local context.
Explore the toolLaunch appInside the research
The intent held while the wording moved
Across the completed runs, only 3 of 30 distinct searches repeated word-for-word, ignoring case and punctuation. This measures exact wording, not semantic similarity: several searches expressed similar needs differently. For content planning, the useful hypothesis is to cover recurring intents rather than optimize around one run’s exact sub-queries.
Price and place changed the search pattern
The under-$300-in-Canada question triggered 4–7 searches per answer, often checking individual product prices. The report counted .ca sites as 53% of sources for that question and 46% for the BC packing question. A .ca domain is the study’s geographic proxy, not proof of a source’s ownership or audience.
Repeat the question before drawing a conclusion
Average pairwise overlap in cited sites ranged from 17% to 62% across the four questions. Reddit appeared in 9 of 11 answers. These observations support repeated testing; they do not establish that any specific content change causes an AI citation.