AI Guidelines for Communicators 

These guidelines provide clarity on how to use AI with fluency and in keeping with UC Berkeley’s Principles of Community, the Appropriate Use of Generative AI Tools advisory, the UC Responsible AI Principles, and the Berkeley Brand.

The guidelines apply to anyone creating communications on behalf of UC Berkeley, including professional communicators, marketers, designers and creative teams, campus partners, leaders, faculty, staff, students, and vendors. They apply to written, visual, audio, video and other public or internal communications. In practice, they are most relevant when AI is used to create, edit, summarize, illustrate, translate or publish communications in UC Berkeley’s voice.

UC Berkeley communications depend on human judgment: the discernment to know what’s true, the expertise to know what’s good, and the responsibility to know what should carry the university’s name.

• AI can be helpful for written work as a thinking, drafting and editing partner, but we do not just pass along AI output “as is” without first evaluating it as unvetted source material. We as communicators are ultimately responsible for discerning whether the output is useful, biased, incomplete, plagiarized or a hallucination – and then using diligence to judge the accuracy, tone, sourcing, voice and revising accordingly.
• AI can also be useful for creative, illustrative and production work, but it requires discernment and taste. The question is not just whether we can make something with AI, but whether it is good enough, accurate enough and Berkeley enough to use. Our visual style is light, natural, human and authentic, not generic, overproduced, or recognizably AI-generated in the worst sense (“slop”). That means AI-generated work should meet the same brand standards we apply to photography, including avoiding unnatural effects, poor quality, overly posed images or visuals that lack context.

What we put into an AI tool can matter as much as what comes out. UC Berkeley communicators are trusted with confidential information, embargoed research, and early knowledge of major announcements before they are ready for public distribution. That trust is the foundation of our work, and putting sensitive information into the wrong tool is a breach of that trust, regardless of intent.

Before entering material into an AI tool, ask: Would this information be appropriate to share outside the university or before publication? If not, use only an approved tool at the appropriate data-protection level – or do not use AI for that material.

Lucky for us, the Berkeley AI Hub maintains the current list of Licensed AI Tools by data protection level, along with an advisory guide on how to use these tools to support innovation without putting institutional, personal, or proprietary information at risk.

New AI tools emerge constantly, and the temptation to use whatever is newest is real. UC Berkeley-licensed tools exist because the pace of change makes vetting more important, not less. They have been reviewed for privacy, security, and institutional risk – data entered into licensed tools is not used to train models and is subject to enterprise privacy agreements. Consumer AI tools, including free accounts and personal subscriptions, do not provide it.

Note that some information is fully off limits for AI systems. There is no AI system rated safe enough for information requiring the highest level of confidentiality and integrity.

Be aware of Allowable Use by Data Classification, and always use discretion when it comes to any potential adverse impact to UC Berkeley’s reputation and business continuity.

We disclose AI’s role when the production method is meaningful to how the audience (whether that be the public or internal) understands the content.

For example:

• For written communications, routine AI help with brainstorming, outlining, editing, proofreading or summarizing generally does not need disclosure. Disclosure is appropriate when AI shapes the substance in a way the audience would reasonably care about. A public-facing report that uses AI to analyze survey responses, generate findings, or create summaries of community feedback should say so, because readers need to understand how the conclusions were produced.
• AI-generated visual assets should be clearly distinguished from documentary, journalistic or artist-made work, but an obviously fictional video of the Chancellor traveling through space with Oski may not need a detailed label.
• An AI-generated version of a professor speaking in another language should note that AI was used, that the speaker consented and that the translation was reviewed for accuracy.

• Appropriate Use of Generative AI Tools
• UC Responsible AI Principles
• UC Berkeley AI Hub, approved tools
• Legal guidance on AI use
• Data Protection Levels
• Data and privacy questions

Questions? Contact [email protected]. This guidance will be updated as tools, standards and resources evolve.