Skip to content
University Marketing and Communications

Marcom AI Committee

Generative AI Policy/Guidelines Recommendations for Marketing, Communications, and Digital Teams

Explore and learn; Apply with integrity.

The generative AI landscape is changing and evolving rapidly. These guidelines exist to provide guardrails and guidance as you leverage these tools. This content will change and be updated regularly to reflect the changes in capabilities of this technology.

The University of Rochester has policies in place to protect institutional data, which all university members must follow. When using generative AI, consider the data being used and whether it complies with university policy. Inputting data into AI tools could potentially expose it to the world. Only “low-risk” data (as defined by University data classifications) should be entered into generative AI tools.

The University of Rochester currently has no agreements with AI providers, and no commercial AI tool meets the university’s standards for handling non-public data. Therefore, only publicly available information may be used with generative AI tools. Always adhere to all existing policies set by the New York State Board of Regents and the University of Rochester when using generative AI.

Guidelines

  • Generative AI is more than just writing tools, they can summarize data, write code, and create and alter images, video, and audio, among other uses
  • For the most part, LLMs (large language models) learn and generate content via supervised learning and by analyzing data and patterns pulled from publicly available information on the web. Therefore:
    • Content isn’t fully original, and may overlook the newest information
    • Reflects historical bias, tends to propagate stereotypes
    • Tools can make up information (“hallucinate”)
  • Little or no security/data protections within most publicly available generative AI tools (there are tools that do provide data confidentiality but many that do not)
  • Many generative AI platforms pattern match against the Internet causing copyright/intellectual property concerns
  • More new regulations and legal precedents are being created for this technology regularly

Do not…

Publish generative AI outputs verbatim or near-verbatim

  • Content derived from generative AI must be edited for originality, grammar, and clarity, and checked for accuracy, plagiarism, bias, and brand consistency. Content should not be copied and pasted directly from generative AI tools and all content sources should be confirmed.
  • Content from generative AI can contain infringed Intellectual Property and you will be responsible for the content you use from these tools.
  • In most cases, content created by generative AI cannot be copyrighted.

Never enter…

  • Confidential, proprietary, or competitive information
  • PHI, PII, patient, or personal data
  • Never use generative AI as a replacement for our expert sources
  • Purchasing/downloading software
  • New AI services/platforms must be vetted by leadership before moving through our standard purchasing processes and agreements.

Do…

Use generative AI to assist with the types of tasks documented in the Use Cases and Prompts Guide.

Transparency and Disclosures

*Whenever possible, rely on our credible University of Rochester experts and use original imagery to avoid the usage of disclosures. These disclosure recommendations apply to internal and external content.

Written Content

  • If you use a full sentence or more verbatim from an AI-assisted technology, use this disclosure:
    • Disclosure: This piece was edited by (subject matter expert) with editorial assistance from AI technology.

Multimodal Content (Imagery, Audio, Video, 3D, etc.)

  • If you create or fundamentally alter images or video via AI technology, use this disclosure:
    • Disclosure: (Original credit when applicable). A portion of this (image, audio or video) has been digitally enhanced using AI technology to (description of alterations).

Editing and improving content using generative AI (using writing assistants like Grammarly or Hemmingway or leveraging AI features within tools like Photoshop/Firefly or Premiere) does not require disclosure. The specific amount of editing that is allowable without disclosure is not currently codified so err on the side of transparency.

Please reach out to a supervisor with any questions on disclosures.

Specific Data, Document Upload Policies, Restrictions

Do not put any data or information into these tools that fall outside the “low risk” category of University data classifications.

Ethical AI Usage / Responsible AI Guidelines

Do not use generative AI to create content that requires a very human and authentic level of engagement (crisis communications, sensitive or personal information, etc.)

Do not use these tools to create any malicious content or deepfakes (falsely creating the likeness of a person’s physical appearance or voice without their approval)