Generative AI in learning and development

• Generative AI in learning and development works best for clearly defined use cases like drafting learning content, translations, quiz questions, metadata, and learning path recommendations.
• The greatest impact rarely comes from simply creating online courses more quickly. It’s when AI effectively connects skill gaps, workflows, and business objectives.
• Data protection, quality assurance, employee participation, and the EU AI Act have to be addressed early on before you start integrating AI into productive learning processes.
• L&D needs governance: Who’s allowed to use which tools? What data is off-limits? Who approves the content? How will results be documented?
• AI works better when your learning platforms, content systems, HR data, and reporting are all seamlessly integrated.
• The best way to get started is with a prioritized pilot project, clear success metrics, and a scalable operating model.

AI has become a part of life for L&D teams. The L&D Global Sentiment Survey 2025 once again saw AI ranked as the top industry trend. More than 3,000 people from nearly 100 countries participated in the survey, the results of which showed that interest in AI has continued to grow year-on-year (Source: Donald H. Taylor, Global Sentiment Survey 2025). 

For L&D, AI is an opportunity, albeit with some side effects. On the one hand, it can speed up tasks that take teams a lot of time, like developing scripts, learning objectives, knowledge-based questions, translations, summaries, scenario drafts, and versions tailored to different target audiences.

On the other, its management is becoming more onerous. If every department uses its own AI tools, they can quickly become siloed solutions. Content is created faster, but is harder to control. Learning paths might appear personalized, but the skills haven’t been clearly defined. The reports provide figures, but no reliable indication of the business impact.

This is exactly where L&D needs a strategic framework. AI should always address a specific question: Which skill gap is being closed? Which process is being made easier? Which compliance requirement is being met more reliably? Which metric is improving?

So it’s important to set priorities to ensure you get off on the right foot. Companies that use generative AI in learning and development don’t need a large AI program that does everything. A well-defined use case that quickly demonstrates its benefit and can be expanded later is much more valuable.

The most pragmatic approach is to start where AI can reduce repetitive work while still allowing for human approval. Content creation is an obvious example. An AI system can generate initial drafts of training texts, case studies, checklists, summaries, and knowledge tests.

This saves time, but isn’t a substitute for a well-thought-out didactic concept. An online course doesn’t automatically become effective when you use AI. That only happens if the target audience, learning objectives, use case context, and knowledge transfer have all been clearly defined.

Another strong use case is localization. International companies often need to roll out content across multiple countries, languages, and roles. GenAI can generate translations, variations, and tone adjustments. This can significantly speed up rollouts, especially for global compliance/product training courses.

The third starting point is metadata. The mmb Trendmonitor 2024/2025 describes AI-supported metadata assignment for learning content as a key factor in reducing the workload for training providers. The study also highlights the growing importance of large language models and text- and image-generating AI tools for educational purposes (Source: mmb Trendmonitor 2024/2025).

Good metadata might sound somewhat dull as use cases go. In practice, however, it determines whether learning content can be identified, recommended, evaluated, and incorporated into learning paths. Without this foundation, personalization is often nothing more than an empty promise.

A good first use case has to meet three criteria: It has to be relevant, manageable, and measurable. Relevant means that it solves a real problem for your L&D team. Manageable means that data protection, quality, and approval can all be clearly controlled. Measurable means that the criteria for determining success are defined in advance.

Typical starting points include:

1. Creating quiz questions from approved texts
2. Translation and linguistic adaptation of existing training content
3. Automatic suggestions for keywords and learning objective tags
4. Role-playing scripts for leadership/sales trainings
5. Suggested learning paths based on defined role profiles

These use cases typically deliver results faster than large-scale AI projects with numerous interdependencies. They also build trust because L&D, IT, data protection, and specialist departments learn how to use AI responsibly together.
 

Many generative AI projects fail for a very simple reason: The process was unclear from the outset. When this happens, AI often just makes things even more chaotic. When roles, approvals, data sources, and quality criteria are lacking, it leads to more content, more coordination, and more risk.

Before it starts using AI, your L&D team needs to carry out a thorough review of its processes. Who creates content? Who reviews it for accuracy? Who reviews its didactics? Who provides legal or regulatory approval? Where are versions documented? What content is audit-relevant?

The data the AI uses also has to be accurate. Personalized learning paths only work if roles, skills, training history, and content metadata are maintained consistently. Without these fundamentals, AI generates recommendations that sound plausible but don’t actually hit the mark.

A clear business case is particularly helpful when facing budget constraints, as it highlights time savings, quality improvements, reduced reworking, and faster rollouts. There’s also a clear link to the learning budget here: L&D has to be able to justify its investments. AI projects therefore need metrics that go beyond mere production speed. These include completion rates, knowledge transfer indicators, assessment quality, fewer support requests, and a shorter time-to-competence.

Quality assurance shouldn’t be an after-the-fact check. It has to be part of the workflow. Otherwise, reviewing the AI results will quickly eat up any time saved.

When it comes to generative AI in learning and development, companies should establish at least four rules:

• First: AI results are never published without being reviewed by a human.
• Second: Technical claims require a verified source.
• Third: AI must not process personal data unless explicit authorization has been granted.
• Fourth: Critical learning content is subject to documented approval.

Exams, certifications, compliance training, and competency assessments are particularly sensitive. The EU AI Act classifies certain AI systems used in education and vocational training as high-risk applications, for example when AI is used to assess learning outcomes or to guide learning processes (Source: EU AI Act )

For businesses, this means that when AI goes beyond mere support and starts to influence evaluations or learning paths, the requirements become more stringent. L&D should therefore consult with IT, data protection, information security, and the works council at an early stage.
 

The true value of AI rarely comes from a single, isolated tool. It comes when AI is integrated into the learning environment: Your learning platform, authoring tool, HR system, skill data, content library, reporting, and support processes.

But this is where things get technically challenging. Many companies are still using legacy systems. An LMS here, an authoring tool there, Excel spreadsheets for in-person training, manual compliance reports, plus custom AI tools in various departments. The result? Media discontinuities, duplicate maintenance, and unclear responsibilities.
Anyone choosing a new learning platform or switching to a different one should incorporate AI requirements directly into the system selection process.
This includes questions such as: What interfaces are available? How are roles and skills mapped? Can content be properly versioned? How does rights management work? What data can be used for AI functions? How can reporting be automated?

chemmedia AG supports companies precisely at this point where consulting, system selection, implementation, integration, administration, managed training services, and content come together. The advantage lies in the consistent project logic. L&D has to be able to explain the technical objectives. IT needs technical security. Data protection and the works council demand transparency. Specialist departments expect effective learning opportunities.

These perspectives need to be harmonized. Otherwise, you’ll end up with an AI solution that impresses people during the pilot phase but falters once in general use.
If you want to delve deeper into system-related issues, you can find further guidance in our articles on LMS integration and selecting a suitable learning platform. In terms of content, it’s also worth taking a look at AI in eLearning.