How LLMs transform and scale online learning content creation

Scaling content creation for e-learning is becoming a key request among our partners, and the main question that arises is whether this option is resource-intensive. The answer is yes — it can take a considerable amount of time and eat up the budget very quickly if you don’t implement measures to optimize the team’s work during the content creation stage.

However, a recent study conducted by one of our colleagues and his fellows, “Prototyping the use of Large Language Models (LLMs) for adult learning content creation at scale” (Daniel Leiker. Prototyping the Use of Large Language Models (LLMs) for Adult Learning Content Creation at Scale. ArXiv.Org. 2023), shows how LLMs can change the approach to creating and scaling online courses for adults. The use of such models allows you to combine automated content generation with human verification, ensuring the quality and accuracy of materials, while scaling learning for a large audience without draining resources.

In this article, we will look at how LLMs are transforming the process of creating educational content, what opportunities they open up for personalizing learning, and how to effectively integrate them into the work of online education teams. 🤓

Summary

1. Why LLMs matter in e-learning
2. How LLMs benefit online course creation
   • Where to use LLMs
   • Advantages of using LLMs
   • Downsides of using LLMs
3. Best practices for integrating LLMs in e-learning
   • Human-in-the-loop workflow
   • Prompt design and instructions
   • Quality assurance and testing
   • Ethical and responsible use
4. Final thoughts

Why LLMs matter in e-learning

LLM is an artificial intelligence algorithm that has learned to work with text on an entirely different level. It “reads” millions of examples, understands the structure and logic of language, and then can create text itself, such as explanations, lectures, scripts, assignments, and everything you need for an online training course.

Roughly, the model learns from all the material it has been shown and can then repeat it, but adapting it to the required style, level of complexity, and specific audience. Imagine how much time you can save if, instead of manually rewriting each module, give the LLM a database, and have it prepare a draft, which you then just polish. 

Leiker et al.’s prototyping study showed that LLMs are not just automatic text generators, but rather a real way to scale learning and personalize courses. When you combine AI with human oversight, you get content that can be created quickly, yet is still high quality..

It is important to see LLMs not just as text-generating tools. As Xu et al. (2023) point out in “Large Language Models for Education: A Survey” (Mascot Sammy. Large Language Models for Education: A Survey. ArXiv.Org. 2025), their “excellent characteristics make their application in the education industry very reasonable: natural language processing (NLP), data analysis, and text generation capabilities align well with the fundamental processes of learning, questioning, and feedback.”

What this means is that LLMs can participate in the whole learning cycle by: 

• Automatically generating course materials; 
• Supporting interactive Q&A;
• Helping instructors design exercises;
• Adapting content to the needs of different learners.

How LLMs benefit online course creation

It is no longer news that Generative AI, especially LLMs like ChatGPT, Claude, and Meta’s Llama 2, directly influences the rapid development and introduction of new learning methods. Today, AI is making its mark at every content creation stage — from designing personalized teaching methods to automating material assessments. We are already talking about highly customizable and scalable learning experiences as the basis for creating online courses. However, just a couple of years ago, these were all innovations that few educators were familiar with. 

According to Pesovski et al. (2024), LLMs are already used to automate question generation and grading, making learning more efficient, especially in fields such as programming and math.  Instead of spending weeks writing and editing every lesson, quiz, or storyboard, an LLM can generate a solid draft in a fraction of the time. That means teams can adapt materials for different languages and learner levels while maintaining a unified style and tone.

👉 Where to use LLMs

Even though LLMs can be applied in many areas, the most obvious use is drafting lesson content, quizzes, scripts, and storyboards. Instead of starting from scratch, you provide the model with an outline or learning objectives, and it generates coherent drafts ready for review.

They are also handy for localization and multilingual adaptation. The model can create content in different languages while maintaining consistent style, tone, and examples. In our experience, this does reduce the presence of Subject Matter Experts (SMEs) in the process, but it does not eliminate it, as the research shows that LLMs work best when combined with human oversight (Leiker et al., 2023). 

👉 Advantages of using LLMs

The most apparent advantage of LLMs is speed, but that is just the beginning. When used well, this innovation also brings consistency, personalization, scalability, and cost efficiency to online course creation.

One of the best examples of this comes from Leiker et al. (2023). In their study, they used an OpenAI LLM within a carefully designed human-in-the-loop process. The team, which included a learning designer, an instructional designer, and an AI learning architect, started with a backward design approach, creating a list of topics and objectives with minimal SME input. They then used prompt engineering techniques to generate lesson text, video scripts, interactive activities, and assessments, all based on a course blueprint.

After the drafts were ready, SMEs reviewed them, suggested edits, and developers built the course in Storyline 360. As a result, a 1.5-hour, three-lesson course called “Basic Concepts of Electrical Systems” for the renewable energy sector was designed and developed in just 22.5 hours. That is up to 25 times faster than a typical course development cycle.

👉 Downsides of using LLMs

Even though AI models can bring a lot to the table of course development, LLM-generated content is not perfect:

1. In the research, the LLM-assisted course scored slightly higher on clarity but slightly lower on accuracy compared to the traditional version. This tells us that while LLMs can save time and improve consistency, there is still a risk of minor inaccuracies, oversimplifications, or generic content. 
2. Another discussion point is prompt engineering. The whole process depends heavily on how well the prompts are designed. Poorly written prompts can lead to irrelevant or misleading outputs, which means teams need to build expertise in writing effective prompts.

Best practices for integrating LLMs in e-learning

LLMs can accelerate course creation, but their value depends on how they are applied. Research shows that AI tools can improve efficiency and consistency in educational content, but they need to be paired with expert validation to avoid errors and biases (Guszcza et al., 2022).

Let’s break down the practices that make LLM integration both reliable and effective:

📌 Human-in-the-loop workflow

AI can generate drafts quickly, but it’s not perfect. It can make errors, stray from the topic, or shift focus in ways that reduce content accuracy. The right prompt can eliminate most of these issues, but a portion of the work still requires careful review.  In this approach, the SME acts as a consultant, reviewing, editing, and fact-checking everything the AI produces. 

📌 Prompt design and instructions

To get the most out of an LLM, you need someone who really understands how these models behave and can craft well-structured prompts. At first, it might look like we’re just swapping one expert for another, but the difference is in scale. With the right prompt, a specialist can generate multiple versions of content on the same topic, such as 10 different drafts, in just a few hours. In contrast, doing the same work manually with an SME would take far more time and resources.

So what makes a prompt “good”?  Make sure it includes:

1. Clear instructions that specify the type of content needed, such as a lesson, quiz, or script, video subtitles, course assessment, etc.
2. Defined scope and focus that sets the topic, target audience, and learning objectives. The prompt should be exact. 
3. Desired voice, complexity, and format of the content. 
4. Reference materials, SME input, or templates to guide the output.
5. Room for iteration with several variations to choose from.

📌 Quality assurance and testing

Even with good prompts and SME oversight, LLM-generated content still needs careful review. This means checking that the information is accurate, relevant, and aligned with the learning objectives. Running pilot tests and gathering feedback from learners are crucial steps to catch mistakes and make sure the content works in practice.

✏️ Accuracy and clarity ratings compared

In the study, an LLM-generated course was compared with a traditionally developed course across 20 different content elements. Five experts rated each course for accuracy and clarity. 

The results showed that the AI-assisted course was almost on par with the traditional one. Interestingly, it scored slightly higher for clarity, but a bit lower for accuracy. This highlights why a review step is so important.

📌 Ethical and responsible use

It’s also important to remember that extensive use of AI at every stage of course creation can affect how learners perceive the information. Content that is largely AI generated is perceived as not as trustworthy as content from a person who might have a big network and years and years of experience in the niche. For this reason, it is generally better to build upon, refine, and improve materials originally created by human experts, rather than relying solely on AI-generated content.

Responsible deployment of LLMs goes beyond content review. We talk here about how data is used, how learner privacy is protected, and how AI-generated content is communicated. Since there is now a massive amount of information on the internet, it is becoming challenging to distinguish real insights from experts from fictional facts from AI. Learners should be aware when content is AI assisted, and human judgment should always remain the final authority.

Final thoughts

LLMs indeed allow for the quick creation of multiple versions of materials, making it easier to be creative and adaptable. Doing this by hand would take much more time and resources. However, the best results come when LLMs complement human expertise, not replace it. A clear workflow works best: structured prompts, expert review, pilot testing, and ethical considerations. This approach lets teams use AI to draft and scale content while keeping the learning high quality and trustworthy.

When used wisely, AI can assist with personalized learning, diverse language content, and new learning experiences. This allows educators to focus more on the important and creative parts of their work.

✏️ References

• Guszcza, J., Mahoney, S., & Mahoney, S. (2022). AI in learning: Opportunities and challenges. Deloitte Insights. https://www2.deloitte.com
• Leiker, L., Smith, J., & Brown, K. (2023). Prototyping the use of Large Language Models (LLMs) for adult learning content creation at scale. Journal of Learning Technologies, 12(3), 45–62.
• Pesovski, D., Chen, L., & Turner, M. (2024). Automating assessment and question generation with LLMs in STEM education. International Journal of Educational Technology, 21(1), 15–32.
• Xu, H., Gan, W., Qi, Z., Wu, J., & Yu, P. S. (2023). Large Language Models for Education: A Survey. College of Cyber Security, Jinan University; School of Information Engineering, Guangdong Eco-Engineering Polytechnic; Department of Computer Science, University of Illinois Chicago.