A fluent answer can still contain the wrong date, an invented condition, or a missing warning. In this course, you learn a practical routine for deciding whether an AI answer is ready to use.
You help a fictional local library review an assistant that answers visitor questions. You define what a useful answer must do, check its claims, compare two drafts, and test a change. All source notes and example outputs are written for teaching. They are not real library policies or results from a live model.
Your 25-minute route
| Chapter | What you practice | Lesson time |
|---|---|---|
| 1. Define a good answer | Write requirements you can check | 4 minutes |
| 2. Check the evidence | Match claims to source passages | 5 minutes |
| 3. Compare answers fairly | Apply the same checklist to two drafts | 5 minutes |
| 4. Test an improvement | Use varied cases and check for new mistakes | 4 minutes |
Plan about 18 minutes for lessons and exercises, 4 minutes for chapter checks, and 3 minutes for the final test. These are estimates. Take more time where you need it.
Each chapter has one lesson and two questions. The final test has four new questions, one per chapter. You must answer every question correctly to pass a chapter check or the final test. Explanations appear after grading. For multiple-choice questions, select all correct options.
How to practice
You can complete every exercise on paper using the supplied notes and answers. No setup, coding, or paid tools are required. Generative AI Essentials and Practical Prompting are useful starting points, but neither is required.
This course focuses on evaluating answers, not building an automated evaluation system. A small test set gives evidence about the cases you tried, not a guarantee about every future answer. For consequential decisions, involve someone qualified to review the result.
What you will learn
- Turn a task into a short checklist with clear pass conditions
- Check individual claims against sources and distinguish unsupported claims from contradictions
- Compare answers using the same requirements without letting style hide important errors
- Create a small test set and check that an improvement preserves answers that already worked
Before you start
- You know how to use everyday web applications
- You do not need programming experience, an AI account, or paid tools
Course modules
Define a good answer
Turn a visitor's question into clear requirements and identify errors that must block publication.
Check the evidence
Separate an answer into claims and check what the sources actually support.
Compare answers fairly
Use fixed requirements and the same inputs to compare drafts without rewarding style over correctness.
Test an improvement
Test varied inputs, preserve working behavior, and keep conclusions within the evidence.