Generative AI Guidelines for Brand and Marketing

These guidelines apply to UCSF, UCSF Health, UCSF Benioff Children’s Hospitals, and all associated entities (i.e. schools, departments, divisions, initiatives, etc.). For simplicity, references to “UCSF” throughout this document encompass all UCSF entities and apply to anyone creating marketing, web, design or similar materials on behalf of UCSF.

Purpose

Since UCSF marketing materials have the potential to influence decisions, shape perceptions, and impact trust, AI-assisted work must meet established brand standards and expectations for sound judgment, appropriate review, and accountability. If you choose to use AI for content and marketing work, remember that you remain responsible for meeting these standards.

Generative AI is one of the many tools that may be considered when developing creative, content, marketing, and branded materials. It can help people explore ideas, support creativity, and accomplish various tasks. These tools have been known to introduce risks related to accuracy, authenticity, bias, privacy, accessibility, copyright, and regulatory compliance; when choosing to use generative AI, human judgment and review are essential to navigate these risks. 

These guidelines are intended to help you understand appropriate uses of AI in marketing materials. They are intended to supplement UCSF brand guidelines and do not replace existing policies regarding branding, content review, patient privacy, accessibility, copyright, clinical review, or regulatory compliance. These guidelines will continue to expand to include categories such as editorial content in future.

Overview

Four Guiding Principles 

  1. Human accountability
    AI is a tool. Employees, not AI systems, are responsible for the accuracy, quality, legality, and appropriateness of all materials created or published on behalf of UCSF.
  2. Existing Standards Still Apply 
    The use of AI does not exempt content from adherence to:
  3. Protect Inputs
    Use only UCSF-approved AI applications (such as Enterprise ChatGPT), accessed through your UCSF account. A personal account on the same product is not an approved tool. Do not enter patient information or PHI, confidential, or embargoed information, or protected/licensed/non-consented assets unless the specific tool and AI usage are explicitly approved.
  4. AI Cannot Replace Professional Judgment
    UCSF is a human-centric organization. We take pride in the expertise of our brand, marketing, and content teams. AI may augment that expertise, helping employees to do their work more effectively, and not replace professional judgment. The introduction of AI for creative and marketing materials does not change existing roles, responsibilities, approval requirements, or governance, nor does it expand an employee’s authority to perform work requiring content, design, legal, clinical, or other professional expertise.

What to verify before publishing UCSF branded materials

  • Brand: Use only approved assets and standards (logos, typography, colors, imagery, messaging, editorial guidelines, and nomenclature) as outlined on identity.ucsf.edu. Consider the broader brand landscape and touchpoints that an audience may experience and avoid unnecessary duplication, inconsistent treatments, or conflicting messages. Do not rely on AI to enforce brand standards.
    NOTE: UCSF logos must never be generated by AI. AI may recommend logo placement but should not be used to place or reproduce logos, as it may alter them even when provided with approved files or references. Logos must always be added manually using approved logo files.
  • Factual and visual accuracy: Verify facts, statistics, citations, names, locations, labels, and other factual details against authoritative sources.
  • Clinical accuracy: Confirm health information, treatments, recommendations, and claims with the appropriate subject matter expert.
  • Representation and authenticity: Verify that people, anatomy, objects, settings, and context are devoid of inaccuracies, including distortions, stereotypes, or misleading depictions.
  • Compliance: Ensure all content complies with privacy, copyright, licensing, accessibility, and regulatory requirements.
  • Human craft: Edit AI-generated content for strategy, clarity, tone, originality, and usefulness. Remove generic, formulaic, or unnecessary content.

Risks and Considerations

An AI hallucination is an inaccurate, misleading, or completely fabricated response that an AI system may present as fact. Hallucinations can occur with AI-created content and creative. As a healthcare, research, and academic institution, the risks and considerations go far beyond errors and inaccuracies. Accordingly, it is each individual’s responsibility to understand the risks, perform due diligence, and adhere to UCSF’s guidelines for AI-generated branding and marketing materials outlined here. Anyone choosing to use generative AI remains accountable for the resulting work.

Accuracy and Representation Risks

AI-generated content commonly contains:

  • Factual inaccuracies such as incorrect statistics, locations, or healthcare information.
  • Visual inaccuracies such as distorted anatomy, body positions, physical features, incorrect geographic representations, distorted logos or brand marks, or incorrect numbers or labels. It may also introduce stereotypes, artificial diversity, or other artifacts that make depictions feel inauthentic or misleading.
  • Clinical inaccuracies such as incorrect care recommendations, misrepresentation of treatments, or unsupported claims.

Compliance Risks

Use of AI does not change UCSF’s responsibility to comply with existing requirements related to:

Brand Risks and Considerations

  • Brand Compliance: Even when prompted to use UCSF or UCSF Health brand guidelines, AI-generated content often fails to follow approved standards related to compliant and accurate logos, typography, colors, imagery, messaging, and nomenclature.
  • Brand Consistency: Even when individual assets appear acceptable, decentralized creation of  materials can result in multiple versions of the same asset, inconsistent visual treatments, conflicting messages, or fragmented brand experiences.

Do’s and Don’ts

The examples below illustrate uses of generative AI that may be appropriate, as well as uses that should be avoided. This list is illustrative, not exhaustive. Consider the purpose, context, potential risks, and level of human expertise or review needed before deciding whether to use AI.

Generally appropriate use of AI, with qualified human review

  • Ideation and exploration: To brainstorm themes, concepts, headlines, visual approaches, storytelling ideas, and other creative directions.
  • Early concepts and prototypes: To explore preliminary layouts, mockups, presentation structures, and other concepts before developing them with approved UCSF brand assets.
  • Copy adaptation: To adapt, consolidate, or summarize marketing content using approved source material and facts you provide.
  • Routine production support: For tasks where the output can be readily reviewed and verified, including organization, formatting, readability, and early accessibility considerations.

Treat AI-generated content as a starting point. Verify, refine, and evaluate the work for accuracy, appropriateness, accessibility, and brand alignment before use. Human review and refinement is needed when leveraging generative AI.

Inappropriate use of generative AI

  • Brand identity and representation: Don’t use AI to create or alter UCSF logos, protected brand elements, or applications such as merchandise, signage, uniforms, badges, or environmental graphics.
  • Authentic people, places, and experiences: Don’t create or publish AI-generated imagery that could be mistaken for actual UCSF people, places, facilities, events, or experiences.
  • High-stakes or accuracy-dependent visuals: Don’t rely on AI to represent clinical procedures, medical equipment, interfaces, data, instructional information, or other subjects where visual accuracy is important.
  • Institutional voice and meaning: Don’t rely on AI to determine UCSF’s voice, messaging, positioning, or meaning—or allow it to alter the intent or clinical accuracy of approved content.
  • Drafting content: Don't use AI to create complete articles, press releases, website content, or other written materials without substantial human input and review.
  • Final review or approval: Don’t treat AI output as finished, brand-approved, factually verified, clinically accurate, legally compliant, or accessible without appropriate human review.
  • High-risk applications: Don’t use AI when fabricated details, errors, or inaccuracies could create patient, safety, legal, regulatory, privacy, or reputational risks.

Resources

Related Resources