Skip to content

AI companies

How does Google DeepMind source training data?

By SourceX Editorial · Updated

Short answer

Google says it relies on publicly available data to train models like Gemini, drawing on web documents, code, and other text, image, audio and video content, filtered for quality and safety. Google's EU AI Act training content summaries add that data acquisition spans public web content across educational, government, legal and research sectors. This page reflects only what Google has published.

Google's general position#

Google's public policy pages state that it relies on publicly available data to train models and that it applies data protection practices developed over more than two decades to AI training.

Gemini model cards#

DeepMind's Gemini 3 Pro model card says the card aims to give more information about the training dataset, distribution and intended uses than prior model cards, though it does not itemize specific named sources.

  1. 01Your work records
  2. 02Personal details removed
  3. 03You approve the release
  4. 04Licensed to AI developers
How it works with SourceX: Your work records → Personal details removed → You approve the release → Licensed to AI developers.Illustrative

EU AI Act training content disclosure#

Google's EU AI Act public training content summary for the Gemini 3 Pro family describes text training data as including publicly available web documents across educational, government, legal and research sectors, plus datasets representing images, audio and video content.

Google Cloud-specific training#

Google's documentation for Gemini in Google Cloud states that Gemini models used in Cloud products are trained on publicly available code and Cloud-specific material in addition to the base Gemini foundation model datasets, and that it does not use customer prompts for training those products.

What this means for a business#

A business reviewing whether its records could end up feeding an AI system run by a major cloud or search provider should check the specific product's data-use terms, since policies differ between consumer products, enterprise cloud products, and foundation model training, and should remove personal data before any content is shared.

SourceX and Google DeepMind#

SourceX is not affiliated with Google or Google DeepMind. This page summarizes public reports; it does not mean Google DeepMind buys data through SourceX.

Frequently asked questions

Does Google disclose specific sources used to train Gemini?

Google's public EU AI Act summaries describe broad categories, such as web documents across sectors like education and government, but do not name specific websites or datasets.

Does Google use Cloud customer prompts to train Gemini?

Google's Cloud documentation states that Gemini for Google Cloud does not use customer prompts or generated responses to train or fine-tune its underlying models.

Is the Gemini model card the same as a training data audit?

No. DeepMind's model cards provide capability, safety and high-level training information but are not an itemized list of training sources.

Sources

Public sources, checked 2026-10-09. SourceX is not affiliated with the organizations named.

  1. Google states it relies on publicly available data to train models publicpolicy.google
  2. Gemini 3 Pro model card description and stated aim to disclose more training dataset information deepmind.google
  3. Gemini 3 Pro family EU AI Act public training content summary describing web document sectors gtt.harshp.com
  4. Google Cloud Gemini does not use customer prompts for training underlying models docs.cloud.google.com

Related resources

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies Talk to SourceX
See if your company qualifies