This guide explains how to think about using AI to generate quizzes and learning paths. The most important rule: don't hand the whole job to AI.
- AI is an assistant, not the designer
- For quizzes, the quality of what goes in decides what comes out
- For paths, AI works best when you use it to suggest an order
- A person always reviews before anything is published
AI is an assistant, not the designer
AI is very handy, but it isn't responsible for your learning design. Start by being clear about who does what.
Decide the purpose, the goals and the assumptions
Drafts, ideas, and help with coverage
Generating quizzes with AI
Input is everything
When AI generates a quiz, the quality of the information it starts from decides the result.
Control the expected score
Left alone, AI drifts toward questions that are too easy or too hard.
Always prepare the reasoning
In a quiz, "why it's wrong" matters more than the answer itself.
POINT 1 Input is everything
At a minimum, be clear about:
- The target role (who it's for)
- The purpose (checking understanding, or assessing judgment)
- What prior knowledge learners have
In Manabu2, the AI option generation (AI による選択肢生成) panel in a quiz's question editor suggests 3 to 5 candidate questions based on the course, section and lesson titles and your organization's name. There's no field for instructions, so specific, descriptive titles are your input. Use the three points above as your criteria when you pick a candidate and edit it. If you have an AI agent create quizzes through MCP, put these points in your prompt.
POINT 2 Control the expected score
Left alone, AI tends to write questions that are too easy or too hard. So decide on an expected outcome like the ones below and use it as your yardstick when you choose and edit candidates (if you use an AI agent through MCP, you can put it in your prompt).
- "I want 80% of people to get this right"
- "I want more than half of people to get this wrong"
POINT 3 Always prepare the reasoning
In a quiz, why an answer is wrong matters more than the answer itself. Design every question so you have both of these:
- Why the correct answer is right
- Why the wrong answers are wrong
Note that the AI candidates in the quiz editor don't include this reasoning, and lesson quizzes have no field for storing an explanation. Work out the reasoning yourself and convey it in the lesson body or a supplementary lesson.
Generating paths with AI
POINT 1 Use AI to suggest an order
AI is good at proposing a structure, such as:
- Which order makes the material easiest to understand
- What can be skipped
In Manabu2, the AI learning path suggestions (AI による学習パス提案) panel on the paths list lets you choose a Target department (対象部署) and add Additional requests (optional) (追加の要望（任意）). It then shows suggested path names and descriptions based on your organization's information, departments and existing courses. The panel doesn't show a course order, so a person decides which courses go into the path and in what order.
Meanwhile, people make these calls:
- What's required and what's optional
- What gets assessed
POINT 2 A natural fit for role-based paths
AI is good at branching designs, such as separate versions for learners and for managers. It's most useful as an aid to decisions like "for this role, skip this lesson."
Rules for using AI generation safely
Common mistakes
- Publishing what AI produced without changes
- Building a test while the assessment criteria are still vague
- Not adjusting the difficulty
Every one of these comes down to not enough design.
Related guides
The next guide walks through, with concrete examples, where people tend to stumble when using AI.
