What it does
GetGeoAgent expands a seed topic into the real query space around it, classifies every query by search intent, groups them into clusters that can be owned by a single page, and maps each cluster against what you have already published.
A list of keywords is not research
Most keyword tools return a spreadsheet: thousands of rows sorted by volume, most of which are variations of the same question. That output is not a plan. Deciding what to build from it is the actual work, and it is the part that gets skipped when a quarter gets busy.
The agent does that deciding step. Expansion, classification, clustering and gap analysis run as one process, and the output is a set of decisions — build this page, extend that one, leave this alone — rather than raw data you still have to interpret.
How the research runs
- 01Expand the seedHead terms, long-tail variants, questions, comparisons and the conversational phrasings people type into assistants — because those phrasings increasingly matter as much as classic keywords.
- 02Classify intentEvery query is labelled informational, commercial, transactional or navigational. Intent decides the page type; without it you get one generic article per keyword and none of them satisfy anyone.
- 03Cluster by answer, not by stringQueries a single strong page can satisfy are grouped together. Two phrasings of the same question belong on one page, not on two pages competing with each other.
- 04Map against your siteEach cluster is matched to an existing page where one exists. That immediately exposes cannibalisation — two pages chasing the same cluster — and genuine gaps.
- 05Score the opportunityClusters are ranked on how well you could answer them relative to what already ranks, so effort goes where it can realistically win.
Why intent classification changes the output
Two queries can share a topic and need completely different pages. "What is generative engine optimization" wants a definition and an explanation. "GEO software" wants a product page with a comparison. Publishing an article for the second one is a reliable way to rank for nothing.
| Intent | What the searcher wants | What to build |
|---|---|---|
| Informational | To understand something | A definitive explainer with a direct opening answer |
| Commercial | To evaluate options | A comparison, an alternatives page or a category overview |
| Transactional | To act now | A product, pricing or signup page |
| Navigational | To reach a specific place | A clear route to that page, not a new article |
Clusters, not keywords
A cluster is the unit the agent plans around: one primary question, the supporting questions that belong with it, and the single page best placed to answer them all. Clustering this way prevents the most common self-inflicted SEO problem — a library of near-duplicate posts splitting their own signals.
- Every cluster maps to exactly one primary page.
- No two pages compete for the same cluster.
- Supporting questions live on the page they belong to, or on a linked page that earns its own existence.
- Gaps are visible as clusters with no page attached.
Question phrasings, not just keywords
People ask assistants full questions. "What software can automate SEO for a small marketing team?" is not a keyword, but it is exactly the phrasing an answer engine has to satisfy. The agent collects those phrasings alongside conventional queries, because the same page usually has to serve both. See AI search optimization for how that feeds GEO.
What happens to the research
Research is only useful if something is built from it. Clusters flow into the content planner, which turns them into a sequenced plan, and into the content optimizer, which checks whether pages you already have cover their cluster properly.
- Seed
- Queries
- Intent
- Clusters
- Opportunities
- Plan