Skip to content

AI companies

How does Microsoft source training data?

By SourceX Editorial · Updated

Short answer

Microsoft primarily deploys models built by OpenAI alongside its own, and says it does not use customer data from services like Exchange or OneDrive to train Copilot or other models. Microsoft has also launched a Publisher Content Marketplace so publishers can license content for use in AI products such as Copilot. This page reflects only Microsoft's own disclosures and reported deals.

Microsoft and customer data#

Microsoft states in its enterprise data protection documentation that Microsoft Copilot and Microsoft Copilot Chat do not use customer prompts or responses to train underlying models, and that organizational data is protected under Microsoft's Data Protection Addendum.

OpenAI as a subprocessor#

Microsoft's documentation describes OpenAI as a subprocessor for certain Copilot experiences, operating under Microsoft's contractual oversight, in addition to Microsoft's own Azure OpenAI deployments, following Microsoft's long-running multi-billion dollar partnership with OpenAI.

  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

Publisher Content Marketplace#

In February 2026, Microsoft Advertising launched the Publisher Content Marketplace, letting publishers set licensing and usage terms for AI builders and get paid based on how their content is used, co-designed with publishers including the Associated Press, Business Insider, Condé Nast, Hearst and Vox Media, with early use feeding Microsoft Copilot.

Microsoft's own foundation model efforts#

Microsoft has not published a detailed line-item list of training sources for its own smaller in-house models; most of its public disclosure has focused on data protection commitments for customer data rather than model training composition.

What this means for a business#

A U.S. company using Microsoft 365 Copilot should check whether a given workload uses the Azure OpenAI deployment or the OpenAI-subprocessor option, since both carry different contractual data handling terms, and should confirm with its Microsoft agreement that prompts and files are excluded from model training.

SourceX and Microsoft#

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

Frequently asked questions

Does Microsoft Copilot train on my company's files?

Microsoft states that Microsoft Copilot and Copilot Chat do not use customer prompts or responses to train underlying models, under its enterprise data protection terms.

Is OpenAI involved in training Microsoft Copilot?

Yes, Microsoft documentation describes OpenAI as both a long-term partner providing models through Azure OpenAI and, more recently, as a subprocessor for certain Copilot experiences.

What is Microsoft's Publisher Content Marketplace?

It is a licensing platform Microsoft launched in February 2026 so publishers can set terms and get paid for content used to ground AI answers, including in Copilot.

Sources

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

  1. Microsoft Copilot does not use customer prompts/responses to train underlying models learn.microsoft.com
  2. OpenAI operating as a subprocessor for certain Copilot experiences under Microsoft oversight learn.microsoft.com
  3. Microsoft's multi-billion dollar, multi-year investment and partnership with OpenAI openai.com
  4. Microsoft Publisher Content Marketplace launch details, Feb 2026 about.ads.microsoft.com
  5. List of publisher partners co-designing PCM including AP, Business Insider, Condé Nast, Hearst, Vox Media searchengineland.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