Prompts, topics, and fanouts
Build a representative prompt set and keep it organized for reliable analysis.
Prompts define the questions Sight repeatedly observes, so prompt design determines the meaning of every GEO metric. A representative set should cover real audience decisions, category discovery, comparisons, problems, branded intent, and relevant market or language differences without over-weighting one theme.
Topics and tags keep the set interpretable as it grows. Maintain a stable core for trend analysis, use temporary tags for experiments, and archive prompts when they should stop running without deleting history. Fanouts and suggestions are research inputs that still require review before becoming tracked prompts.
Review the prompt inventory periodically for duplication, outdated products, ambiguous wording, and accidental brand bias. Prompts should sound like questions an audience could genuinely ask without knowing the desired answer. When a prompt changes materially, treat it as a population change and avoid presenting the new response history as a seamless continuation of the old wording.
Design the prompt set
Start with audience questions that represent real decisions, then balance branded, category, comparison, and problem-led intent.
Write prompts at the level of a real decision and avoid embedding the expected brand or answer unless branded intent is the subject. Balance discovery, comparison, evaluation, and problem-solving questions. Review the complete set for over-represented products, markets, or funnel stages.
- Group prompts into topics that map to products, services, audiences, or decisions.
- Use location and language where the answer can vary by market.
- Track a stable core set before adding experiments.
- Use tags for campaigns, funnel stage, intent, or business ownership.
- Archive prompts to stop future runs without deleting answer history.
Suggestions and fanouts
The brand profile can generate topic and prompt suggestions. Fanouts expose related search or shopping queries produced during AI research behavior.
Suggestions accelerate research but should not enter recurring tracking without review. Confirm that the wording is understandable, relevant to the selected market, and distinct from active prompts. Fanouts can reveal adjacent interests, but they may be transient or model-specific.
| State | Behavior | History |
|---|---|---|
| Suggested | Review, accept, or reject before activation. | No tracking history yet |
| Active | Runs in future scans within plan and credit limits. | Accumulates answer history |
| Archived | Stops running but remains available for analysis. | Preserved |
| Deleted | Removed with related data according to the endpoint behavior. | Not recoverable through the product UI |