Generative AI

Guidelines for Use of Generative AI in Marketing and Communications at Michigan State University

University Communications and Marketing > Resources >

Guidelines for Use of Generative AI

Exploring the Use of AI

University Communications and Marketing and campus communicators have explored a variety of use cases for integrating generative artificial intelligence (generative AI) into marketing and communications work. We evaluated different uses and different tools and have crowdsourced our learnings. 

This document is meant to serve as a source of information and guidelines for University Communications and Marketing staff as they explore the use of AI.

Explored Use Cases

The following are example use cases that showed potential for mid-to-high value:

  • Brainstorming and idea generation for various projects
  • Content suggestions for social media posts
  • Content versioning, including for A/B testing, such as alternate email subject lines
  • Headline suggestions for articles
  • SEO suggestions for optimizing content
  • Administrative tasks, such as drafting project briefs
  • Transcript summarization for meetings
  • Research support, such as summarizing capabilities of tools, comparing/contrasting programs and defining terms
  • Generating alt text for images

Additional use cases were explored, including coding, writing stories, editing and image generation. These required more human intervention to deliver value, had higher levels of risk and/or delivered lower value. University Communications is continuing to evaluate use cases as tools progress.

AI Landscape

Generative AI tools present exciting opportunities for those working in communications and marketing, offering innovative solutions to enhance our strategies and achieve greater success. These tools can assist in various aspects of our work, including content creation, data analysis, customer engagement, and more. By harnessing the power of AI, we can streamline our processes, gain deeper insights into audience behavior and deliver more personalized and effective communications to our target audiences. Embracing AI in our work will enable us to adapt to the rapidly evolving landscape of communications and marketing.

Things to Keep in Mind

AI tools offer numerous opportunities for communications and marketing professionals, but it's essential to be aware of certain concerns and considerations when incorporating them into our work. Here are some of the key points to keep in mind:

  1. Data Privacy and Ethics: There is a need to handle all data with the utmost care. Ensure that your AI solutions adhere to data protection regulations, maintain transparency in data usage and follow ethical guidelines to respect user privacy.
  2. Bias and Fairness: AI systems can inherit biases from the data they are trained on. It's crucial to be vigilant about potential biases in AI-generated content or decision-making processes to avoid unintended discrimination or misrepresentation.
  3. Quality Control: While AI can automate many tasks, human oversight is still crucial. Always review and edit AI-generated content to ensure it aligns with the brand's voice, values and quality standards.
  4. Over-Reliance on AI: AI tools are supportive but should not replace human creativity, strategic thinking and intuition. It's important to strike a balance and use AI as a tool to augment, not replace, human expertise.
  5. Flexibility and Adaptability: AI tools and technologies are continually evolving. Stay adaptable and be prepared to adjust strategies as new AI innovations emerge.
  6. User Agreements and Terms of Service: It’s important to be aware of specific tools’ terms of service and user agreements prior to their use. This clarifies how generated output can be used and helps avoid potential violation of the user agreement. In some cases, this will mean a paid version of a tool is necessary to obtain the user rights necessary. 

By being mindful of these considerations and taking appropriate measures, communications and marketing professionals can harness the power of AI tools while mitigating potential risks. 

Useful Terminology

  • Large Language Model (LLM): A large language model is a sophisticated computer program that can understand and generate human-like text. It's trained on vast amounts of written language to assist with tasks like answering questions, writing content or having conversations with people. Examples of LLMs include GPT-3.5, GPT-4 and LAMDA.
  • ChatGPT: ChatGPT is a chatbot interface for the GPT LLM. 

Disclaimers

Tools such as ChatGPT have limitations. For one, GPT-3.5 was trained on data available prior to September 2021 and GPT-4 was released in March 2023, with an updated knowledge base extending to September 2023. ChatGPT cannot access the internet to report on live events. 

LLM models also inherit biases from the datasets on which they were trained. In addition, they will tend to fill in information gaps with information that may not be accurate, and they don’t always get math correct. This means the tools can’t be relied upon to generate information that doesn’t have inaccuracies. When using the tools, take that into consideration and don't assume the information they present is entirely accurate. Users also need to consider audience expectations and ethics regarding content creation and presentation using AI tools.

Use caution when asking ChatGPT to do something a person doesn’t understand or can’t verify. A person who doesn’t have development skills may not know the risks of generating code, for example. Or someone who isn’t familiar with SEO may not know the output will not make SEO better.

AI is a fast-moving, fast-changing area of technology. Government officials are still determining how best to regulate the technology. The European Union is working on regulations for use. An executive order was issued by President Biden regarding platform testing with implications on use by federal employees in October 2023, and future legislation is possible.

MSU will continue to monitor developments and will update recommendations, as necessary. 

Guidelines for Use

General AI Usage

MSU IT has posted Interim Guidance on Data Uses and Risks of Generative AI on the tech.msu.edu website. These cover MSU’s expectations regarding use of AI platforms from an IT perspective.

IMPORTANT: Please take specific note of the guidelines section regarding data stewardship. There are specific instructions regarding institutional regulations and requirements pertaining to use of data with AI tools.

Creating successful prompts in chat interfaces with AI tools is a learned skill. Better prompts yield better responses. There are tricks and nuances involved in how you instruct the tool using your input to supply a quality output. Many online articles provide tips on how to craft good prompts. You can also ask the tool for tips on how to improve your prompts or combine tools to craft quality prompts. 

For example, ask ChatGPT how to improve your prompts for Midjourney. Ask ChatGPT to tell you what it needs to know to create a quality output. Suggest the tool take on a persona, such as a prospective college student, in its response. 

Practice and learn through experience. Over time you may develop a standard set of prompt language to “train” the tool to answer in the way you would like.

Read the terms of service for any tools you’re using to determine whether your output is protected. By default, treat any AI-generated content as not eligible for copyright. There is still debate as to who can “own” the copyright on AI content, due in part to uncertainty about whether the data used to train the LLMs included copyrighted material and whether a computer can be considered an “author” under the Copyright Act.

Meetings and Operations Tasks

AI meeting assistants often take notes and summarize meetings. This can be very helpful. However, they do not always get things correct. Additionally, any documentation these tools generate is subject to Freedom of Information Act (FOIA) requests. Do not use AI tools that generate meeting notes and summaries in meetings that cover confidential or sensitive topics. For meetings when the tool is used, do not disseminate notes or summaries without first reviewing. If the tool automatically shares notes or summaries with attendees, review these and provide clarifications or corrections as soon as possible via email to all recipients.
If using an AI tool in a meeting, start the meeting by notifying attendees that you’d like to use the tool. Explain what the tool does and request consent from all attendees to use the tool. Do not rely on chatbot notifications to attendees as a form of consent. If anyone objects, do not use the AI tool in the meeting. 
Some AI meeting tools take notes and summarize meetings. Some go further, providing insights to the tool user into meeting attendees’ behavior (e.g., camera on/off) and engagement (e.g., eye contact, body language). Make people aware of the tools being used and use your best judgement on which are appropriate for your circumstances.

Writing and Editing

Use an AI tool as a first pass at editing, and then review any recommendations in the context of MSU’s brand voice and style requirements. AI tools for editing, such as Grammarly or the grammar check built into word processing tools, default to their own editorial standards. These guidelines do not always match the MSU Editorial Style Guide. For example, many use the Oxford comma, which is not used for MSU nonacademic content.

For more information on the editorial style guide, see the University Communications website.

AI tools, such as ChatGPT, can help with initial drafts of written content. Any material written or drafted using an AI tool must be reviewed for factual accuracy. For stories involving research work, ask the involved researcher(s) to review the final draft to ensure accuracy. For AI-generated bios and profiles, the subject of the bio should review the content.
Public attitudes towards AI vary and are not always favorable. There have been instances of negative public reaction to its use in some circumstances, such as using AI to generate condolence messaging after crisis situations. Additionally, some studies show the reliance on AI for messaging leads to a decrease in trust in the messaging. Consider your audience and context before using AI for messages.

Graphics, Photo and Video

AI tools can create some amazing imagery that can accompany stories and designs, but the images are not perfect. You can expect to need multiple rounds of edits to fine-tune your image using prompts, and possible further editing using other traditional tools, such as Photoshop.

Provide high quality prompts that may include various forms of visual material. Do not publish without review and adjustment to fix the kinds of imperfections often found in AI images. Care should be taken to use only professional versions of AI tools such as Midjourney or Dall-E, so that terms of service allow for MSU to use the image in publications.

We want to maintain the trust of our audience, so we need to be judicious with how AI-generated images are presented. Any image created using AI should not be used as illustration or art within a context that implies the image represents reality. AI images should not be positioned to represent a real location, person, etc. If an image is used for illustrative purposes within a context that could cause confusion (e.g., with a news article or informational content), labeling should be considered. Avoid using AI to generate images of individuals without their expressed consent.
AI-generated transcription and caption tools can drastically speed up the process of captioning videos. However, these tools are not perfect. The transcript and caption files should be reviewed for accuracy before adding to the video file.

While certain alterations of the approved Sparty mascot illustration have been permitted to represent our various athletics teams and for use on licensed products, development of these altered marks is carefully overseen by University Licensing Programs, which must approve their development. Sparty the mascot (and images thereof) is a trademark of the university and should not be altered in any way without permission. Sparty is one of the most recognizable collegiate symbols in the world, and while generative AI may make it easy to place him in various settings with custom apparel augmentations or other props, this is not permitted.

University units may book the Sparty mascot for photo shoots and the resulting images may be used for communications and promotional purposes. Units wishing to use Sparty in this way should first consider if this imagery will best support the messaging and tone of their communication. Information about booking Sparty may be found at https://alumni.msu.edu/sparty/appearance-request.

Summary

Do:

  • Experiment and try out different AI tools. It’s important to stay informed on industry trends and to learn how to use the tools.
  • Read about generating quality prompts and practice to build this skill.
  • Use AI to save time or increase productivity.
  • Use AI when it will generate better results than you can do on your own.
  • Use AI for supporting and accelerating creative workflows (e.g., writing project briefs, generating transcripts, generating social media post copy to accompany an article).

Do Not:

  • Upload confidential or proprietary data to a public AI.
  • Use AI in confidential or sensitive meetings.
  • Use AI to generate finished images for publication or distribution, without review.
  • Use AI to generate images of Sparty or other university trademarks.
  • Include AI images in a context implying reality.
  • Publish anything created with AI without review for accuracy, brand voice and editorial style.

Tools & Resources

There are hundreds of AI-based tools on the market. Here is a sample of those that have been reviewed by the MSU communicator network:

Bing Chat

Bing Chat is the Microsoft chat-based AI, built on GPT-4. It is a part of Bing search and available via the web or in the Microsoft Edge browser sidebar. 

ChatGPT and ChatGPT Plus

ChatGPT is free to all users. It includes access to the GPT-3.5 LLM, with data up to 2021.

ChatGPT Pro has a monthly cost ($20 as of October 2023). This includes access to GPT-4, with enhanced capabilities, faster response speed, visual input options and beta features (e.g., web browsing, plugins, advanced data analysis).

There are iOS and Android mobile apps available. The parent company also provides a best practices guide.

Note: ChatGPT has been added to MSU Procurement’s list of approved software purchases as Tier 1, meaning it can be purchased if it does not use, process or store confidential information.

Claude.ai

Claude is another AI chatbot, operating an LLM by Anthropic. The newest version, Claude 2, is free but comes with limitations. Claude Pro was released in September 2023 with a paid subscription fee of $20/month. They also have a “token” model that allows you to pay for tokens that correspond to words, subwords and characters. One campus user shared that they use both Claude and ChatGPT to generate social media copy.

Google Bard

Bard is a conversational AI tool by Google, operating on the LAMDA LLM. It is free, but currently has limited capabilities in its experiment state. Learn more about Bard on the Google Bard FAQ page.

Grammarly

Grammarly is a writing improvement and editing tool. It can help generate writing, rewrite text, and suggest corrections and enhancements to grammar, tone, clarity and consistency. There are apps, plugins, and browser extensions to aid in adding into existing workflows. There are free and premium plans. 

WordTune

WordTune is a tool that enables rewrites of content, such as social media copy. There is a Chrome browser extension to add the tool to your web workflow. A campus user suggested this tool for obtaining rewrites of social media copy for different uses such as including formal, shortened, casual, and lengthened versions. There are paid and free (with a daily limit) plans.  

Gender Decoder

Gender Decoder reviews your job postings for gender bias by searching for masculine or feminine-coded words. It does not suggest replacements, but it is helpful, particularly if your job descriptions are old. Free to use, no subscriptions or restrictions.

Adobe Creative Suite (built-in tools)

Adobe is adding AI tools into its suite of technologies. These can aid creators in such tasks as generating video transcripts and text-based video editing improvements (Premiere), object selection (After Effects) and image editing tools (Photoshop, Lightroom). 

Dall-E

Dall-E is a text-to-image generator. The most recent version is DALL-E 3. This is available to ChatGPT Plus users. DALL-E 2 operates on a credit basis, with each request costing one credit. 

Midjourney

Midjourney is a bot used to generate images from text. It works on the Discord platform. There are several tiers of subscription plans with varying limitations, starting at $10/month (as of October 2023). 

Stable Diffusion

Stable Diffusion XL is a text-to-image diffusion model that can create images from text. The tool is available online for free.

Read.ai

Read provides AI generated meeting summaries, topics, action items and questions from meetings. It can also provide speaker feedback to help users become better communicators. Some campus users were impressed by the meeting notes, which one user rated as higher quality than another tool tested (Otter.ai). However, some users expressed concern about the privacy standards for data captured about meeting attendees, so be sure to review the terms of service. There are several tiers of pricing available, starting with a free basic plan (5 meetings per month). 

Sprinklr AI capabilities 

Sprinklr (MSU’s enterprise social media management platform) includes AI capabilities designed to provide insights into account audience, inbound message sentiment, themes in listening and time to publish. These are available to campus Sprinklr users via their annual paid subscription.

ChatGPT FAQ for MSU Educators

The #ITeachMSU Commons website includes a roundup of resources and FAQs specific to educational applications. See the post on #ITeachMSU Commons.
Version 1. Updated: December 4, 2023