Google Dataset Search: the alternatives that list commercial data

Google describes Dataset Search as a search engine that lets people discover datasets hosted in thousands of repositories across the web. It does not store the files. A result is a summary built from markup on someone else's page, with buttons that send you to that page to read the terms and download.

How pages get in

Pages get in by describing themselves. Google's structured data guide says a dataset page can use schema.org Dataset markup or the W3C DCAT format, with JSON-LD preferred. Two properties are required: a name and a description of 50 to 5,000 characters.

A license property, ideally a URL for a specific licence version, and an isAccessibleForFree flag are recommended. Dataset Search reads only the first 5,000 characters of any text property, and a page owner who wants out uses a robots meta tag.

What it covers, and what it misses

The result is a catalogue of whatever repositories bother to mark up their pages: government portals, university archives, research repositories and data hubs. It is very good at open and academic data. It sees little of the data companies sell privately, because that data is not published on public pages at all.

• Last updated: the past month, year or three years.
• Download format: table, text, image or other.
• Usage rights: whether the dataset allows commercial or only non-commercial use.
• Topic: broad disciplines such as social sciences or life sciences.
• Free: a toggle for datasets that cost nothing to use.

If Dataset Search doesn't turn up what you need, the next place to look depends on whether you want open data or are prepared to pay. The table separates tools that only point to data, sites that host it, and marketplaces that sell it.

• Academic or government statistics: start with data.gov or data.europa.eu, then Dataset Search for university repositories.
• Machine learning practice data: Kaggle and Hugging Face host the files, so you can load them in a few lines of code.
• Large raw corpora such as satellite images or genomes: the Registry of Open Data on AWS, where the files already sit next to compute.
• Business data your team will query: BigQuery sharing, Snowflake Marketplace or Databricks Marketplace, depending on the warehouse you already run.
• A commercial data product with a sample and a price: a data marketplace such as Datarade or AWS Data Exchange.
• Data that has never been published: none of the above. You need to find the company that holds it.

Sellers sometimes ask whether listing a paid dataset in Dataset Search brings buyers. You can mark up a product page with name, description, license and isAccessibleForFree set to false, check it in the Rich Results Test and submit a sitemap; Google's help page says re-crawling takes a few days.

What it won't do for a seller

That makes the page findable. It does not handle samples, contracts or payment, so it is a supplement to a marketplace listing or direct sales, not a replacement, and our guide on where to sell data covers those routes.

DataBroker is not a search engine or a data marketplace, and it hosts no datasets, so for finding open data Google Dataset Search and the repositories above will serve you better.

What we do instead

Our paid course and app are for brokering data that is not public: a lead finder that ranks companies likely to hold sellable data, value ranges for codebases, tasks and raw data, contract templates and a CRM. The downside is effort. A brokered deal means finding the owner, checking rights and negotiating, and many conversations end without a sale.