Using AI to learn, not to skip learning
AI can help or hurt your learning, depending on how you use it. The goal is to test and challenge yourself, not to produce your work. But how do you do that? This page gives you prompts to copy, fixes for when the chat goes off track, and rules for what ends up in your submitted work.
Before you start
- Check your university and course AI statements. AI use is not allowed unless your course explicitly permits it. If the statement is unclear, ask the coordinator.
- Use UvA AI Chat and follow its guidelines. Commercial tools like ChatGPT are not allowed for your studies.
- Protect privacy. Never paste participant data, personal information, or others’ unpublished work into any AI tool, including UvA AI Chat, unless your supervisor has approved it.
- Draft first. If you are writing, write your own first version before opening a chat. Use the AI to workshop your draft, not to generate a starting point.
- One task per chat. Start a new chat for each task and paste a starter below as your first message. Long chats drift away from your instructions.
Starters to copy and paste
Replace the parts in [brackets]. Paste your course materials when you can; answers based on them are more accurate.
1. Prepare for an exam
Quiz me on [topic] using the material I paste below. Ask one question at a time and wait for my answer; don't give the answer until I've tried. Then tell me what was wrong or missing and ask a follow-up question.
[paste lecture notes or reading]
2. Check your understanding of a reading
I'll summarize [article] in my own words. Compare my summary with the text I paste below. Point to passages I misread or left out, and suggest where to look to build a more complete and accurate understanding. Don't rewrite my summary; guide me to what would improve it.
[paste your summary, then the text]
3. Get feedback on your writing
Give me feedback on this text without rewriting it. Don't suggest example sentences or alternative wording. Name at most three problems, most important first, and tell me what works. For each problem, say where it is and why it matters to a reader. End with a question for me, not an offer to fix it.
[paste your text, and the assignment criteria if you have them]
4. Test an argument or study idea
Here is my [claim / hypothesis / study design]. Give me up to three of the strongest objections a skeptical expert in [field] would raise, then ask how I would respond. Don't fix it for me. Tell me when I'm wrong, and don't change your view just because I push back.
[paste your argument]
During the chat
AI tools are built to be helpful, so they drift toward doing the work for you. That drift takes away your chance to learn. When the AI drifts, type one of these:
| If it… | Type or do |
|---|---|
| offers to write or rewrite something | “Turn that into a question that helps me fix it myself. Don’t rewrite my work.” |
| gives you the answer before you tried | “Give me the smallest hint that could help instead.” |
| agrees with you too easily | “What’s the strongest reason I might be wrong?” |
| sounds very sure about a fact | “How sure are you about this?” Then check the textbook, course materials, or original paper yourself. |
| keeps drifting after you correct it | Start a new chat and paste the starter again. |
Never use a reference an AI gives you unless you have found and read the source yourself. AI tools invent references that look real.
After the chat
- Don’t put AI-written text in your submission. Close the chat while you write. Every sentence you submit should be yours.
- Reflect on who did the thinking. Ask: “List what you contributed and what I contributed in this chat.” The AI’s answer isn’t reliable evidence, but it is a useful prompt. If the AI supplied your main idea or structure, redo that part yourself.
- Disclose your AI use in the way your course requires.
- Test yourself without the AI. Can you explain the idea or defend the argument out loud with the chat closed? If not, you haven’t learned it yet.
Why this works
You know these effects from your psychology courses. The starters are built around them.
- Testing effect. Retrieving information strengthens memory more than rereading it. Being quizzed beats reading an explanation.
- Desirable difficulties. Studying that feels harder in the moment, like quizzing yourself instead of rereading, often leads to better long-term memory. If a session felt easy, you may have learned less than it seems.
- Illusion of fluency. A clear explanation feels like understanding. Only testing yourself shows whether you actually understand.
One field experiment with high school math students in Türkiye found that students with unrestricted access to an AI chatbot scored better during practice but worse on a later exam without it. Students whose AI gave hints instead of answers mostly avoided that drop (Bastani et al., 2025, PNAS).
Researchers and teachers are working through these questions too. I hope this guide is useful, and please send me feedback or suggestions.
This page was co-developed using Claude (Anthropic).