AI for Content Development: Do’s and Don’ts

Lately, we’ve been getting a lot of questions about generative AI and content. Questions like, "I’ve got 50 more web pages to write before next week’s launch. Could I use AI to get that done faster?" Or, "Could I use AI to help keep up with all these social media posts?"

You won’t be surprised to know that the answer to these questions is "Yes. But…" Yes, AI can help you create high-quality content and speed up content creation. But you’ll need to think through a few things if you’re going to use AI effectively and strategically to meet your content and marketing goals.

Here’s a list of do’s and don’t’s that can help you figure out how you want to use AI for content creation. just to avoid the repeat of "think through" from the previous sentence.

Do: Consider the Negative Aspects of Generative AI

The magic of generative AI comes at a price. The environmental impact of the data centers needed to make AI work has been well documented. So too have issues with gender, racial, and cultural bias, along with concerns about working conditions for the annotators who label images and other inputs for consumption by AI models. 

It’s — a lot. While work to mitigate these problems is ongoing, it’s still just getting started. If your institution has commitments around any of these issues, you should consider how your use of AI fits within those commitments. Make sure you have any needed internal discussions before moving to an AI-dependent workflow.

Do: Have a Content Strategy in Place

As a content strategist, I am bound by law and custom to remind you that you need a content strategy. That advice is doubly important when using AI for content development. You need clear answers to questions like:

• What kind of content are you trying to make?
• Who will be reading or watching this content?
• How will making that content for that audience help you accomplish your (non-content, offline) goals?

Answering those questions helps you evaluate content made with help from AI. Let’s say you want to create a sequence of emails for admitted students and your goal is convincing them to choose your school. What you get back will probably sound great at first glance. But you need to evaluate that content and decide if it would, in fact, convince your audience to choose your school.

And that doesn’t just mean verifying that it didn’t hallucinate any false or misleading information about your institution. You need to evaluate the content from an admitted student perspective and decide if sending it out will increase — or decrease — your yield.

Don’t: Let AI Do the Writing for You

Which brings us to our first don’t: don’t have AI write your content directly.

I know, I know, that’s not what you were planning to do. But it might be what your boss, VP, or president thinks AI can do. So, in the interest of preparing you to explain to your higher-ups why you can’t just have AI write the entire site, let’s look at some of the reasons that would be a bad idea.

The main issue is that AI writing just isn’t there yet. Typing in a prompt and getting back several paragraphs of human-sounding prose feels magical. But read that text carefully. AI-generated writing often feels generic or "off" and may exhibit bias. While better prompting can improve things, at some point you need to consider the tradeoff between how much time you spend on prompt engineering and how much time you could have spent simply writing the content. Your time and effort savings may be elusive.

Posting content with that generic AI feel also affects your brand. You’ve worked hard to define your university brand and differentiate it from the competition. After that investment, the last thing you want to do is post generic content that could be from any school. Again, prompt engineering can help, but AI-generated content will still need careful review to ensure it conforms to your brand guidelines.

On a practical level, AI copy raises search engine optimization issues. Because AI-generated websites are great for generating online scams at scale, Google is updating its algorithm to detect AI-generated copy and downrank it. If you’ve been working on your SEO, you don’t want to undo that effort with AI content.

Do: Use AI as a Brainstorming Partner

So if it’s not going to write the content for you — what’s AI good for?

A useful model is to approach generative AI like a partner. Try these ideas:

• Ask it for three (or ten!) ways to phrase a headline.
• Have it summarize interview transcripts or long-form survey answers.
• Use AI to create templates for different types of content, like event descriptions or reminder emails.
• Write something, then have your chatbot suggest how to work in your SEO keywords. Or feed it content and ask it to tell you what the most frequent keywords are.
• Have the chatbot suggest how to rewrite something to reach a different audience. For instance, adapting something written for a student audience to a parent audience.
• Ask your chatbot to adjust the tone of content you’ve already written—making it friendlier or more authoritative.
• Use AI to personalize repetitive content for different situations or audiences. For example, use AI to generate natural-sounding alert messages about your school's 50+ different types of account holds—avoiding the formulaic results you would get by using merge tags.
• You can also try using AI to write drafts of meta descriptions or alt text for images.

There are many, many other possibilities. But whatever you try, review the results carefully. For instance, I asked ChatGPT to create a meta description for the undergraduate accounting program at a major university. It seemed fine:

"Build a strong foundation in finance with [university]’s Accounting program. Gain practical skills, hands-on experience, and prepare for success in today’s business world."

Until I asked it to generate meta descriptions for competing programs at other universities:

"Excel in business with [another university]’s Accounting program. Develop critical financial skills, gain real-world experience, and prepare for a successful career."

"Achieve your career goals with [yet another university]’s Accounting program. Gain essential skills, real-world experience, and prepare for success in the business world."

It turns out that accepting the meta description as written would present this accounting program in an entirely generic way. And that’s exactly the opposite of what you want in a meta description, which needs to make you stand out on the search engine results page. The AI version could be a start, but you would have to revise it to showcase more unique aspects of your school’s accounting major.

Do: Experiment and Encourage Experimentation

Nothing about AI is set in stone. You will find many more ways to use it in the months and years ahead. And you will get there faster if your team (and your web editors) are experimenting with AI and letting others know what they have been trying.

You will get there more slowly if you simply tell people at your institution not to use AI, or to use it only in certain ways. Let’s be real: people at your university are already using generative AI. Forbidding its use will only drive that usage underground — and deprive you of influence over how they use it.

A better approach is to allow your team to use AI while also providing guidance. |n other words, a governance plan.

Do: Have an AI Governance Plan

Your campus may already have or be considering a campus-wide generative AI policy. (Wherever you are in that process, Educause’s 2023 guide to Cross-Campus Approaches to Building a Generative AI Policy remains highly useful.)

But you should also consider the need for a more marketing-specific policy or governance framework. This document could cover, among other topics:

• Best practices for using generative AI tools
• Examples of appropriate and prohibited usage
• Guidance on controlling for bias in AI-generated content and on data privacy
• Links to recommended AI tools and resources
• A requirement for human review of AI-generated content

It should also encourage experimenting with generative AI and sharing the results of that experimentation. For a smaller team, this might be as simple as a Slack or Teams channel. But if you have communicators and marketers all across campus, your policy might call for an AI working group where new approaches and discoveries can be shared.

For additional guidance on AI governance, take a look at Creating an AI Policy for Your Marketing Communications Team, written by our vice president of digital strategy for Inside Higher Ed last year.

Do: Educate Your Campus About the Capabilities of AI

Ensure that, however you use generative AI to develop content, your campus knows what you’re doing and the results you’re seeing.

You are sure to have people on campus who want to know that you are using AI to its full potential as well as people who would prefer you not use AI at all. Keeping them up to date and explaining why you’ve made the choices you have is important for making those conversations useful. This might include:

Helping people on campus understand the difference between what generative AI can create and well-crafted, on-brand copy.

Sharing your AI governance plan and policies so people understand how AI fits into your overall content strategy.

Providing training on the effective use of generative AI for content creation.

The use of AI for content creation is still very much a developing area, and the months and years ahead are sure to require further thought and decisions as new capabilities become available. Keep learning, keep experimenting, and be sure to evaluate each potential use of AI against your content and marketing goals.