AI-generated learning content without losing quality

What the AI in CoVince concretely does

You give a topic or a document. From it the AI makes the structure, chapters, slides and texts. Images and video are created and placed, and you can search stock photos. You translate a text with one click per field. Per slide you can have the AI make a change ("make this question harder"), and the editor has an AI chat that thinks along.

Large courses can pause and continue later, so a long design does not have to happen in one go. Before you publish, a preflight check goes through the learning journey.

Where it goes wrong without editing

AI writes fluently, and fluent is not the same as correct. Three risks keep coming back:

• Plausible but wrong. A sentence sounds good and is not right. In learning content about procedures, safety or rules that is serious.
• Generic. The examples fit "a company", not your company. People drop off when they do not recognise themselves.
• The wrong goal. The AI makes nice knowledge material, while the learning goal was a behaviour.

You solve all three not with a better prompt alone. You solve them with a way of working.

A way of working that guards quality

1. Write the learning goal yourself. One or two sentences about what someone does differently afterwards. That is the anchor for everything the AI makes.
2. Give the AI real material. A document with your procedure or policy yields better results than a bare topic.
3. Read everything as the learner. Go through the learning journey with the eyes of someone who does not know it. Where does it lose people, where is it wrong?
4. Check the facts at the source. Every rule, standard or number goes past an expert.
5. Make it your own. Replace generic examples with situations from the work. Use the AI per slide to do that quickly.
6. Run the preflight check and let a small group test the learning journey before everyone gets it.

Where your time shifts to

The beauty of AI design is not that it is faster, but that the work shifts from formatting to judging. The expert spends less time typing questions and more on whether they are the right questions. That time yields quality: a second and third round, a better case, a sharper goal.

Good use is iterative. Make a first version, have a few people do it, look at the reports and adjust. Because adjusting is cheap, learning from the results is feasible. See measuring learning.

Do not throw away existing material

You rarely start from zero. A Moodle backup you can import one to one or have upgraded with AI. A PDF becomes a learning journey, and an Excel or CSV file with questions (also multilingual) becomes a series of questions. That way AI is not a reason to throw out the old but a way to bring it to life.

What you give the AI determines what you get back

A learning journey is only as good as the input. A document with your real procedure, your language and your examples yields something that looks like you. A bare topic yields a general story. So give the AI three things: the learning goal in your words, the source material and a description of the learner (what do they already know, what do they do daily?).

Also ask for what you do not want: no jargon you do not use, no examples from other sectors, no questions that only test facts. The clearer the assignment, the less you have to rewrite afterwards.

A checklist for the final editor

Before you publish an AI learning journey, go through these points:

• Is the learning goal clearly there, and can every slide be tied to it?
• Is every fact, number and rule correct? Has the source been checked?
• Do the examples fit our work and our language?
• Does each question test something that matters, and are the wrong answers plausible?
• Is the difficulty at the right level, and are there branches for those who already know more?
• Has a second person gone through the learning journey, and what did they say?

The CoVince preflight check helps with what is missing, but the judgement on content stays with you.

Large courses and large documents

With a long course or a large document it is tempting to generate everything in one go. It works better in rounds. First let the AI make only the structure, and review that: are the chapters right, is anything missing, is the order good? Only then have the slides worked out per chapter. Large courses can pause and continue, which fits that way of working.

This keeps control small and quality high: an error in the structure you fix in a minute, an error in fifty worked-out slides you do not. It also produces a better course, because you can see the whole before you dive into the details.

Being honest about what AI is

Tell learners what was made by AI and what was checked by people. That is also the line of the European AI Act: transparency about AI-generated content. For your learning journeys it mainly means that responsibility lies with a person, by name. A learning journey is only finished when an expert has approved it.

Want to see the whole design process? Read designing the complete learning journey.