The observation
Google, Amazon, YouTube, TikTok run continuous, planetary-scale matching of queries to results. They spend billions of dollars a year doing it. That matching IS market segmentation — refreshed every hour.
Every keyword cluster (e.g. "noise cancelling headphones", "best retinol serum", "all-inclusive Barbados") is a demand pocket with a specific intent. Every domain or ASIN that shows up for those queries is a consumer voting, in real time, for a solution.
The structure of the market is already there. It's readable. It's refreshed every hour. The tools to read it properly have only recently caught up.
What this changes about market research
The old way of segmenting a market: hire a research firm, run attitudinal surveys, produce a 5-segment customer model.
The new way: read the segmentation search engines have already done at planetary scale. Eleven segments emerge for the German headphone market. Seven for the UK gift market. Each segment is buying right now.
Both methods have value. Only one is refreshed weekly. Only one shows you who's gaining and losing share in real time.
How Theia reads search-as-segmentation
The process:
- Build a bipartite graph of keywords × products (or keywords × URLs) for the target market
- Weight the edges by CTR — position-1 traffic counts more than position-10 traffic
- Run Leiden community detection with Surprise optimisation
- Validate the emergent clusters against retailer category trees, expert intuition, and continuity over time
- Name the clusters using HHI-weighted distinctiveness
The output: market segments that emerge from data, not from category trees.
Why this works
Search engines optimise for relevance. Hundreds of engineering teams, hundreds of millions of dollars in compute, hundreds of trillions of training signals.
You're not going to out-segment Google by building your own segmentation in Excel. You can, however, read what Google has already segmented — and find the strategic shape of your market faster than any panel survey could deliver.