Google Dataset Search

Google Dataset Search is a search engine for datasets rather than a dataset in its own right. It reads the structured metadata that publishers embed in their pages using the schema.org `Dataset` vocabulary, then folds it into one index that reaches across millions of datasets from thousands of sources: government open-data portals, university repositories, Kaggle, Zenodo, Figshare, national statistics offices, and many more.

Using it is deliberately simple. You type a topic at datasetsearch.research.google.com and get a ranked list, each result carrying a short description, the publishing organisation, a last-updated date, and the available formats where the publisher supplied them. There is no API and no bulk export: this is a discovery layer built for people, so you click through to the original host to download or query anything.

For RAG, treat it as the fastest first move when you know the subject but not the source. Hunting down a niche corpus, say a regional air-quality series or a particular disease registry, is far quicker here than trawling portals one by one. Once you have a shortlist, the real assessment happens on each dataset's own page: licence, access terms, file formats, update cadence, and whether anyone still maintains it.

The caveat to hold onto is that Google indexes whatever metadata a publisher chose to expose, and it does not vet quality, freshness, or licensing. A tidy-looking result can lead to a broken download or a dataset with murky reuse terms, which is why we mark licensing here as 'check source': the licence always lives with the individual dataset, never with the search layer.

Think of it as complementary to a curated catalogue like this one. Dataset Search casts the widest net across the open web; a curated directory trades some of that breadth for vetting, context, and RAG-specific guidance. The two work best together.