Skip to content

AI companies

How does OpenAI source training data?

By SourceX Editorial · Updated

Short answer

OpenAI says its models are built from three main sources: information publicly available on the internet, data it accesses through partnerships with third parties, and data that users, human trainers, and researchers provide or generate. OpenAI also uses synthetic data generated by its own systems. This page summarizes what OpenAI has published about this; it is not a review of OpenAI's actual training set.

What OpenAI says publicly#

OpenAI's help center describes foundation model development as relying on three primary sources: information publicly available on the internet, information accessed through third-party partnerships, and information provided or generated by users, human trainers, and researchers.

OpenAI's California Civil Code Section 3111 training data summary adds that it has used public data, partnered third-party data, user and human-trainer data, and synthetic data since it began collecting data around 2018, with training data in the trillions of tokens across text, image, audio and audiovisual content.

Web crawling and licensing#

OpenAI operates a web crawler as part of collecting publicly available internet content, and it has signed commercial content licensing deals with publishers, including Axel Springer and News Corp, to access archives and premium content for training and for grounding ChatGPT answers.

The Associated Press was an early licensing partner, agreeing in 2023 to let OpenAI use part of its news archive in exchange for access to OpenAI's technology.

  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 summaries#

OpenAI publishes EU AI Act 'Public Summary of Training Content' documents for newer models such as GPT-5.5 and GPT-6 Astra, describing training text as a multilingual mixture of publicly available data, data from partnerships, synthetic data, and human-generated text, including web content, reference material, technical documentation and source code.

User data and opt-outs#

OpenAI states that users can opt out of having their ChatGPT conversations used for model training, and that it reduces personal information in training data where it can, though some incidental personal information may remain.

What this means for a business#

A U.S. business considering any kind of data licensing relationship with a major AI lab should expect to go through a negotiated agreement rather than an automated signup, and should plan time for reviewing what rights it is granting, removing personal or confidential details before anything leaves the building, and deciding who internally signs off on what data is shared.

SourceX and OpenAI#

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

Frequently asked questions

Does OpenAI buy data directly from businesses?

OpenAI has publicly disclosed licensing agreements with large publishers such as Axel Springer, News Corp and the Associated Press, but it has not published a general program for buying data from individual businesses.

Does OpenAI use ChatGPT conversations to train models?

OpenAI says it may use conversation data to improve its models unless a user opts out, according to its help center.

Is OpenAI's full training dataset public?

No. OpenAI publishes high-level summaries, including EU AI Act training content disclosures, but not an itemized list of sources.

Sources

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

  1. OpenAI's three primary data sources: public internet info, third-party partnerships, user/human trainer data help.openai.com
  2. OpenAI training data summary including synthetic data, trillions of tokens, data collection since 2018 help.openai.com
  3. GPT-5.5 EU AI Act training content summary describing public web, partnership, synthetic and human-generated data cdn.openai.com
  4. Axel Springer partnership with OpenAI for content licensing, reported 2023 reuters.com
  5. News Corp content licensing agreement with OpenAI, reported 2024 reuters.com
  6. Associated Press licensing part of its news archive to OpenAI, reported 2023 reuters.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