1. Define the question
Your question is sent to the selected result mode. Opinion map looks for nuance and recurring themes. Quick poll asks the model to classify the direction as yes, no, or can't say.
2. Gather context
Selected public-source categories guide the research scope. Coverage depends on availability, indexing, language, recency, and platform access. A source count is not a count of unique people.
3. Synthesize
Gemini produces structured fields for the answer, confidence, themes, tensions, and—when requested—directional percentages. SurvA validates the response shape before showing it.
4. Show limitations
- Platform and sampling bias can shape the signal.
- Public discussion overrepresents people who choose to post.
- AI can misread sarcasm, context, or cultural nuance.
- Percentages are modeled estimates unless explicitly labeled otherwise.