Artificial intelligence has quickly become one of the most requested capabilities in enterprise software. Content management systems are no exception.

Increasingly, organizations are asking their technology providers for more AI functionality. Sometimes it’s because they’ve seen an interesting demonstration or heard about something another institution is doing. Other times, the request comes from leadership, where there’s a broader mandate to adopt AI across the organization.

But here’s what concerns me: How often do we stop to ask what we’re actually trying to accomplish?

“We need more AI” isn’t a business requirement. Neither is “Our CMS needs to integrate with ChatGPT.”

Those are requests for technology, not descriptions of problems that need solving.

There’s nothing wrong with leadership encouraging AI adoption. Organizations absolutely should explore how emerging technologies can improve their operations. But a mandate to “use more AI” without clearly defined objectives can encourage teams to prioritize adoption over outcomes. Success becomes a question of how many AI tools we’ve implemented rather than what we’ve actually improved.

Take the growing interest in connecting content management systems directly to generative AI tools. The idea sounds appealing: Give contributors a prompt box where they can ask AI to generate new content, rewrite existing pages, or make updates without having to do all the work themselves.

But before we get too excited about what the technology can do, shouldn’t we be asking whether it’s going to help us do the right things?

I’m not at all opposed to AI. Quite the opposite. I believe it has tremendous potential to improve how colleges and universities manage their websites.

But I think we’re at risk of confusing the availability of AI with the value of AI.

And in our rush to put AI into everything, we may be creating new versions of problems we’ve spent years trying to solve.

We’ve Been Here Before

Let’s think about how university websites have evolved. For years, institutions have struggled with decentralized content management.

These symptoms may sound familiar: hundreds of contributors with varying levels of expertise, inconsistent messaging across departments, outdated information, duplicate pages, and content that doesn’t serve the needs of its intended audience.

We’ve worked hard to address those challenges through structured content, templates, editorial standards, permissions, workflows, content ownership, and governance policies.

The goal has always been the same: to make it easier for people across an institution to contribute content without sacrificing quality, consistency, or accountability.

Now imagine introducing a feature that lets every contributor type whatever they want into an AI prompt box to generate or rewrite institutional content.

One person asks AI to make a page sound more exciting. Another asks it to make the same type of content more conversational. Someone else asks it to rewrite a program description to appeal to prospective students, without specifying which students or what information matters most to them.

Each person uses different prompts, provides different context, and brings different expectations.

What could possibly go wrong? We may be reintroducing the very inconsistency that content governance was designed to prevent. Except now, we can produce it much faster.

The Output Is Only as Good as the Instructions

We’ve all heard that AI output is only as good as the prompt. That’s an oversimplification, of course. Model capabilities, source information, context, and other factors play an important role, too.

But the underlying point remains: what you ask AI to do, and the information you provide, significantly shapes what you get back.

Now consider the typical higher education website. You may have hundreds of contributors. Some work in marketing and communications. Others are faculty members, administrative assistants, program coordinators, or departmental staff who update their pages a few times a year.

Many institutions already struggle to train these contributors on writing for the web, accessibility, SEO, and institutional editorial standards. Are we now expecting all of them to become proficient at prompt engineering?

And if they aren’t, who is responsible for making sure the content they generate is accurate, appropriate, consistent, and useful?

An AI integration doesn’t automatically understand your institution’s voice, your target audiences, your academic programs, your editorial guidelines, or your strategic priorities. Those things have to be provided, enforced, and maintained. Simply connecting a CMS to a large language model and giving contributors an unrestricted prompt box doesn’t accomplish that.

“We’ll Just Put It Through a Workflow”

One response I’ve heard is that AI-generated content could simply trigger an approval workflow before publication. That’s certainly better than allowing unrestricted publishing.

But let’s not confuse approval with quality assurance.

A workflow can ensure that someone reviews a proposed change, but it doesn’t guarantee that the reviewer knows whether the information is accurate, the language is appropriate for the intended audience, or whether the content is consistent with institutional messaging.

In fact, AI-generated content may make the review process harder. And when generating content becomes almost effortless, what’s to prevent contributors from submitting significantly more of it? We could end up overwhelming the very people responsible for maintaining quality.

Poorly written content is often easy to spot. AI-generated content can sound polished, confident, and authoritative even when it’s inaccurate, misleading, or missing important context. A reviewer might spend more time verifying the information than it would have taken to write the content in the first place.

And if we’re asking already stretched web teams to review a growing volume of AI-generated content, have we really improved productivity?

Or have we simply moved the work somewhere else?

Does Your Institution Even Have an AI Content Policy?

Here’s another question worth asking. Before giving hundreds of contributors access to generative AI within your CMS, has your institution established clear policies for using it?

I don’t mean general guidelines about responsible AI use. I mean specific policies for official institutional website content, such as:

1. Who is allowed to generate content using AI?
2. What kinds of content can AI create or modify?
3. What information can be shared with external AI providers?
4. Who is accountable for verifying factual accuracy?
5. How should AI-generated content be documented?
6. Are there restrictions around sensitive information, student data, or unpublished institutional information?
7. What happens when AI-generated content conflicts with approved messaging or institutional policy?

These are important operational questions. And many institutions are still developing their answers.

Adding AI functionality before establishing the policies that govern its use risks putting technology ahead of institutional readiness.

Maybe You Don’t Need More Content

There’s another aspect of this conversation that I find particularly troubling. Most university websites don’t suffer from a shortage of content. They suffer from too much content that isn’t properly maintained, isn’t sufficiently differentiated, isn’t written for a clearly defined audience, or no longer serves a meaningful purpose.

We’ve spent years encouraging institutions to audit their websites, eliminate unnecessary pages, consolidate duplicate information, and focus on content effectiveness rather than content volume. Now we’re making it easier than ever to generate more.

But faster content creation isn’t necessarily progress.

If an institution has 10,000 pages and struggles to keep them accurate and current, the solution probably isn’t to make it easier to produce another 10,000.

A better use of AI might be to identify outdated content, detect inconsistencies, recommend improvements, or help contributors understand whether their content actually answers the questions visitors are asking.

That’s a very different proposition from unrestricted content generation.

What Should We Be Asking Instead?

When someone tells me they want more AI in their CMS, my first question is simple:

What are you trying to accomplish?

Before discussing specific AI models, integrations, or functionality, I want to understand the problem they’re actually trying to solve.

If they say they want AI to update pages, I want to understand why those pages need updating. Is the problem that:

• Contributors don’t have time to make changes?
• Information is scattered across multiple pages?
• Nobody knows which content is outdated?
• Content owners aren’t responding to review requests?
• The institution struggles to maintain a consistent voice across hundreds of departments?

Those are different problems that require different solutions.

I’d also ask a few other questions:

• What would success look like? Are you trying to save time, improve quality, reduce inconsistencies, or accomplish something else?
• What should AI actually do? Generate new content, rewrite existing content, identify problems, or recommend improvements?
• Who should control the instructions? Should individual contributors write their own prompts, or should the institution establish centrally managed standards?
• What guardrails are necessary? How will you ensure accuracy, appropriate tone, audience relevance, accessibility, and compliance with institutional policies?
• How will you measure whether AI is helping? Will it reduce work, improve content effectiveness, or simply increase the volume of material that needs reviewing?

These questions are meant to make AI adoption more purposeful, not to discourage it.

Technology providers have a responsibility to do more than simply respond to feature requests. We should help customers challenge their assumptions, identify the underlying problems, and evaluate whether the technology they’re asking for will actually deliver the outcomes they need.

When a customer says, “We want AI to generate content directly in our CMS,” the easiest response is to add that feature to a roadmap.

The more valuable response is to understand why they want it, what they hope to achieve, and whether there’s a better way to reach that outcome.

There’s nothing inherently wrong with leadership encouraging AI adoption. In fact, organizations should be exploring how emerging technology can improve their operations. But a mandate to ‘use more AI’ without clearly defined objectives can encourage teams to prioritize adoption over outcomes. Success becomes a question of how many AI tools we’ve implemented rather than what we’ve actually improved.

AI Should Reinforce Governance, Not Circumvent It

I’m certainly not saying AI-generated content has no place in a CMS. There are legitimate use cases for generating drafts, transforming existing content, producing variations, and helping contributors work more efficiently. But those capabilities should operate within institutional guardrails.

Imagine AI that understands your approved terminology, follows your editorial guidelines, considers the intended audience, works within your structured content models, and flags potential problems before content ever reaches an approver. Think about contributors receiving clear, actionable recommendations rather than being expected to figure out what to ask an AI model. Picture AI helping institutions maintain consistency across thousands of pages rather than introducing another source of variation.

That’s the direction I find more appealing, because it helps institutions do something measurably better.

The Question Isn’t Whether Your CMS Has AI

We’re going to see many more AI features in content management systems. Some will deliver meaningful improvements. Others may look impressive on a feature list without providing much practical value. And there will surely be use cases we haven’t even considered yet.

But I hope institutions resist the temptation to evaluate their technology by the number of AI features it offers. Instead, ask what those features actually accomplish. Do they make content more accurate? More consistent? More accessible? More relevant to the people you’re trying to reach? Do they reduce the burden on contributors and reviewers? Do they strengthen your institution’s ability to manage its digital presence responsibly? Or do they simply make it easier to generate more content?

Before you ask your CMS vendor for more AI, make sure you know what problem you’re asking AI to solve.

After all, the goal shouldn’t be to have more AI in your CMS, just to check a feature box.

Instead, the goal should be to have a more effective website.

Would you agree?