For qualitative city features — food scene, nightlife, expat community, cultural vibe — structured datasets don't exist. No government agency publishes a "restaurant quality index." So we use large language models to extract structured signals from open text sources, then validate against ground truth.
How it works
LLMs read and extract structured data from travel guides, local media, and community sources. The output is a set of scored dimensions per city — not free-form text, but structured fields that integrate into the same scoring pipeline as satellite and statistical data.
What we flag
Every AI-derived field is explicitly marked on the city page. We don't mix AI-extracted data with direct measurements without telling you. If a score comes from LLM extraction rather than a government dataset, you'll see that in the source citation.
What it doesn't capture
- LLM extraction reflects what's written about a city online, which skews toward popular destinations. Less-discussed cities may have thinner or less accurate lifestyle data.
- Qualitative signals are inherently subjective. "Great food scene" means different things to different people.
Why we use it
The alternative is not having lifestyle data at all. Climate and cost are measurable from satellites and statistics. But whether a city has good coffee shops, a welcoming expat community, or safe streets at night — those require a different kind of signal. AI extraction gives us a structured, comparable baseline where none existed.
We'd rather show you a clearly-labeled AI signal than leave the question blank.