GEO keyword research
Traditional keyword tools optimize for Google ranking. GEO keyword research optimizes for the actual buying questions humans type into AI models. They produce different lists.
What the pipeline does
Drop a single seed keyword into /dashboard/research/keywords and Outrigger runs:
- Question expansion: Sonnet generates ~25 buying-intent question variants from the seed.
- AI probing: every question goes to Perplexity (and any other models you've wired in) to capture how AI actually answers it today.
- De-duplication and clustering: Sonnet groups semantically similar questions, picks a canonical representative for each cluster, and tags intent.
- Ranked output: a clustered taxonomy with intent (informational / transactional / commercial / navigational), an estimated volume score, and the competitor brands the AI mentioned for each question.
You walk away with a real GEO map of the category, not just a Google-ranked-list of phrases.
Why competitor mentions matter
Each cluster shows which competitors AI models reference today. That's a gap map: clusters where Brand X is mentioned and your client is not are the highest-leverage citation seeding targets. Run them straight into Citations.
Promote to monitor
Every cluster has a one-click "Promote to monitor prompts" button. Click it and the canonical question becomes a live monitor prompt for that client: appearing in your weekly Share of Model scans automatically.
Limits
The endpoint is rate-limited per agency at 10 runs/hour. A run typically takes a few minutes since each question hits a model. Today only Perplexity (sonar) is wired by default; if your operator has connected ChatGPT or Gemini, those probe in parallel.