AI in Content Management: From Hype to Reality

By Janus Boye

AI can generate the best content, proofread text, write code, and even conduct code reviews. AI is the best and will put us all out of our jobs.

Is this hype grounded in reality? Is AI truly helping people or are we just diverting our attention to becoming experts in prompt engineering which then consumes our hard-earned time?

I chaired the Tech Forum at the recently held Web Summer Camp in Croatia, which was also the Q3 meeting for our European CMS Expert members. Ondrej Polesny from Kontent.ai kicked us off with a deep dive into the real-life narratives, practical insights, and the potential of AI in content management.

In this post, I’ll share my notes from the talk. Towards the end you can find additional links on the emerging topic of AI and CMS and I also encourage you to be a part of the conversation.

As with any good speaker trying to untangle the almost deafening hype, Ondrej started with setting the stage and making us think.

Have you read an article written by AI without noticing? This was the opening question by Ondrej Polesny at Kontent.ai. Editorial note: This article is written by a human.

Is the AI hype grounded in reality?

Moving quickly through what we’ve learned this year when it comes to prompt engineering, the new meaning to hallucinations and how the ‘intelligence’ part of AI tends to take the conversation in all different directions, Ondrej moved specifically to large language models and explaining the generative part.

For content managers, whether the content is created in the CMS (which it rarely is) or outside the CMS, the generative part has a huge impact. Now you can use AI as a smart assistant to flesh out a brief in almost all languages. You can train it in your tone of your voice, and you can naturally do much more than just working with text. It can also help you create compelling visuals that make the text stand out and why only do A/B testing, when the machine can help you find the headline that really performs the best.

What is typical for generative AI? Ondrej shared the characteristics of generative AI. If you look carefully, you’ll notice Deane Barker sitting in the front. He joined us from the US and gave a keynote later at the Web Summer Camp.

What do headless CMS users expect from AI?

Ondrej nailed it when he said:

“Users believe the current hype and are afraid AI is going to take their job, but at the same time, they want AI to actually do it all for them. In reality, they don’t know what they expect.”

In his session, he in particular focused on these two questions

  • What saves content editor’s time?

  • What operations does the content editor do that are repetitive?

So, it’s all about removing the manual and boring work that the content contributor must do, which is not adding much value like copy-pasting, duplicating items and editing images. If vendors can optimise that, the editor can focus on actually adding value.

Do headless CMS users actually use AI?

Sharing from behind the scenes at the public beta of AI features in Kontent.ai, Ondrej revealed that their usage statistics show that people keep experimenting with the implemented AI features. However, the most used actions show that the users keep reiterating over the results.

A comparison here could be the early days of search, where many used Google and found it to be so good, while they couldn’t really make search work on their own website. In other words: People are very demanding, especially when their private interaction with chatGPT seems to be so good even though they don’t understand why and how.

According to Ondrej, product improvements in particular need to happen on three fronts:

  • We need to provide a more tailored experience for editors and make the AI features more available and easy to use. Right now, editors are happy to jump through multiple hoops to actually see AI in action, as they are excited to see the results, but this will soon end and we will face the hard reality - if it’s not trivial to use and gets us real useful results, editors will quickly develop resistance to the word AI and connect it with “will waste my time with unusable results”

  • We must improve prompts - the more context we pass to AI, the better results we’ll get. This is still a challenging game as there are restrictions on input length and the know-how on what type of prompting actually works is emerging. As vendors we need to identify ways to teach AI about each project’s specifics and then explaining very specifically what is expected and needed as a result.

  • We require better models - looking at the serious applications within a CMS, GPT3.5 results were mostly unusable. GPT4 is much better and this will likely improve. The big question here will be - how can we adjust the model to give us results that the editor expects. There are so many variables that we’ll likely need to train to model on customer specific data, which comes with challenges on its own – data privacy/security, the amount of data being too small, and of course pricing. Right now, it is still unclear what the usage of AI is going to cost and we’re already talking about the next level.

In closing Ondrej shared ideas for future work, but also shared his sense of urgency working at a vendor in a competitive marketplace:

“We don’t have much time - the industry is moving forward at an exceptional pace, so we need to start delivering AI-powered great and useful tools yesterday.”

Finally, you can download the slides (PDF) from the session.

Learn more about AI and CMS

There’s already much written about this topic, and many CMS vendors are rushing to add AI-powered features, but keep in mind that it’s still early days.

Here’s just a few recent posts on the topic from our blog:

The conversation continues in our CMS Expert group and naturally also at our upcoming conferences:

You can meet Kontent.ai in both our groups and at the conferences. Join us and help shape the future and if you can’t make it, do leave a comment below and share your perspective.