AI is a significant technological advancement with rapid adoption, providing substantial opportunities for partners and users. However, AI in MSN also poses risks, among them the challenge to maintain uniformly high content-quality standards and avoid misinformation and spam.
This policy creates a framework and recommended best practices to handle AI content in MSN. As AI capabilities continue to transform, we expect to further evolve this policy collaboratively, working with partners and users to produce world class AI-assisted content experiences.
In this article
Definitions of unreviewed AIGC and AIAC
Summary of customer-facing actions
Policy overview
Our audience and user community trust us to clearly distinguish AI content from human-created content. To maintain trust, AI content on MSN must remain AI-assisted content (AIAC) and unreviewed AI-generated content (Unreviewed AIGC) will be prohibited with very limited exceptions.
The MSN AI content policy maps to the Microsoft AI principles and Responsible AI Standard to provide accountability and transparency via human oversight and disclosure, and incorporates all current MSN content standards, such as prohibiting violating content and preventing plagiarism.
Definitions of unreviewed AIGC and AIAC
Unreviewed AI-generated content (Unreviewed AIGC) and “generated autonomously by AI” refer to the generation of an output by an AI system or model without downstream human review or intervention. AI-assisted content (AIAC) refers to outputs that are generated with human review, material human intervention and/or direction. In this case, “output” means text, image, audio, video, designs, and other works.
For clarity, we understand “material human intervention” to mean humans providing input or feedback. This can mean changing the AI-generated content or automating certain basic tasks, such as word or phrase replacement, to improve the efficiency of the process, such as spell check, transcription, or translation. This could also include basic newsroom tasks to assist with content production, such as research, newsgathering, or the use of structured data or data analysis.
For example, if this policy document was generated from a single human prompt and published without any further human intervention, it would be classified as unreviewed AIGC. If the outcome of the prompt was reviewed, approved, and/or edited by a human, it would be classified AIAC.
Principles
Adherence to core professional publishing and journalistic standards is an essential part of the evaluation process for any entity entering and remaining in the MSN ecosystem. Our content standards apply to all content, whether a product of pure human creation or AI. AI tools to help create content (text, imagery, audio, video) can be used in MSN if partners and posters abide by the following three guiding principles:
Human oversight and accountability
Unreviewed AI-generated content (AIGC) must be clearly labeled or otherwise tagged as AI-generated content. All final content should be prepared with the same care and attention used in publishing content today. Content creators are responsible for ensuring the quality of the content they provide is aligned with the standards they have agreed to, even when they used AI to create the content.
AI-assisted content must:
Have a clearly accountable human owner/editor.
Use an accountable byline where appropriate.
Originality
All existing content standards apply to AI content, including plagiarism. Any prompt submitted into AI tools to create content (video, text, audio or in combination) that plagiarizes current published works is prohibited. Content cannot be copied from multiple articles and spliced together without change. If content is built leveraging other articles, more than one should be utilized, and no more than 80% of the content should be leveraged from one source to help avoid republication/duplication of the same content. All articles created by leveraging other articles must credit or otherwise provide attribution for transparency. The following actions are also not permitted:
Using AI to rephrase/remodel existing published content and republishing them under a different by-line and/or brand.
Using AI to create singular or high-volume rewriting of content. This is specifically aimed at prohibiting news sites using AI to rewrite or rephrase genuine news content and republish large volumes of thin, unoriginal news. This is done with the aim of changing the angle of the original story to focus on a partisan political angle, racial slant, or other divisive topic, with the goal of manipulation by clogging the feed with its appearance in multiple places.
Using AI to impersonate others or create content in the style of a specific author or artist to deceive the audience as to provenance of the work (e.g., avoid prompts such as, “Write a paragraph on this topic in the style of Stephen King or Margaret Atwood” or “Draw a Warhol-style T-shirt”).
Disclosure and transparency
MSN requires all partners and their content to comply with all applicable laws, including transparency provisions in regulations such as the EU AI Act and the California AI Transparency Act, if applicable. This includes not intentionally removing any legally required AI transparency data for content provided to MSN, and mandatory disclosure of content constituting a deep fake.
Best practices for AI-assisted content (AIAC)
Disclose AIAC to MSN at ingestion and tag it where possible. If tagging isn’t possible, disclose prominently within the article, video, or gallery.
Provide user-facing disclosures for AI-generated images, audio, or video, with examples such as:
- "This content was created with the help of AI"
- "AI generated, human customized"
- "Part of this content was created with the help of AI, including human input and review."
Require human accountability through bylines such as:
- "Jane Doe (AI-assisted)"
- "By Jane Doe (with AI)"
Summary of customer-facing actions
Our customers trust us to inform them if the content they are consuming was created using AI. It is a best practice that all partners self-report AI metadata as follows:
Tagging and Badging: All content created with AI, including any user-generated content created with the help of AI, be disclosed at the time of publishing content, either at the feed or content level. Additionally, publishers may not strip legally required AI provenance metadata from images, video, or other content. MSN systems will preserve transparency metadata where feasible.
Enforcement
All partners must abide by the AI Guidance listed in this policy. They must also adhere to MSN’s publishing standards and guidelines and will be subject to our prohibited content takedown policies and guidelines.
FAQs
If a headline is modified or generated by AI to create a better headline, does it need to be labeled?
Yes. As mentioned above, it is a best practice that anything produced with the assistance of AI be disclosed, ideally but not necessarily as an intro or in the footer of the document, with the wording: “This content was created with the help of AI”. Partners have the ability to mark incoming content as AIAC at either the feed or content level from our partner tools UI.
If the research for content comes from AI but the article is written by a human, is this AI content?
No, using AI for research when a human writes, rewrites, or reshapes an article entirely is not the same as using AI to modify or create the content.
If a story is written by a human but then the AI revises it, is this AI content?
Yes, and it will need to be appropriately labeled.
How will MSN enforce the policy?
Suspected violations of the policy will be escalated for review and potential removal of the content. If your content is determined to be in violation of the policy, we reserve the right to suspend or terminate your partnership.
How should AIAC content be labeled?
As above, in terms of language for the disclosure, the recommended wording is “This content was created with the help of AI.” This will make it clear it’s assistive. We recommend the wording be visible as an intro or in the footer of the article. You also have the ability to mark incoming content as AIAC at either the feed or content level from our partner tools UI.
How will MSN show to users that my content is AIAC?
If you disclose that your content is AIAC as part of the content body, then that will be shown along with the content. As above, we strongly recommend as a best practice that anything produced with the assistance of AI be disclosed, ideally as an intro or in the footer of the document with the wording: “This content was created with the help of AI”.
Will MSN selection be different for AIAC content?
MSN’s selection of content to show to users will not be different based on whether it is AIAC or human-generated content, as long as the content meets our quality bar.