Why this matters

AI isn't just a shortcut—it's a powerful partner for learning, creativity, and growth. When you use AI thoughtfully, you build the skills that matter most: critical thinking, creativity, and judgment. When you use AI to ask better questions, understand multiple perspectives, and explore new ideas, you are preparing for a future with AI as a partner, not a proxy, in learning.

Key skills to use AI for learning

AI tools can support your learning—helping you understand complex topics, practice more effectively, and explore ideas more deeply. The key is using these tools thoughtfully and developing what we call appropriate reliance —finding the right balance between trusting AI when it performs well and maintaining healthy skepticism to catch mistakes.

Here are three complementary approaches that can help you get the most out of AI learning tools:

✅ Verify first: source check AI output

AI can get things wrong, so you should always double-check AI-generated information by looking at trusted sources. Determining that key information is correct isn't just about due diligence; it can reinforce the most important concepts for you as you study!

Why it works: Checking a claim against external sources can force your brain to retrieve what you know and assess it — and retrieval helps to make learning stick (Roediger & Karpicke, 2006).

Prompt: Which key points are most crucial that I get correct if I want to learn [topic] properly? For each, suggest a few trusted sources I can check (e.g., textbook chapters, official websites) to independently confirm that you've covered it accurately. If I identify differences or gaps, help me investigate.

✅ Beyond explanations: use AI to personalize practice and review

You can use AI to quiz yourself and create a custom plan for spacing out your practice, which has been shown to help remember things better over time.

Why it works: Retrieval practice and spaced repetition can produce more durable learning and transfer (Roediger & Karpicke, 2006; Cepeda et al., 2006).

Prompt: Be my retrieval coach for [topic]. Start by giving me a short answer question that I'll do my best to respond to. Score it 0–2 with a one sentence correction if needed. Repeat this 8 times, interleaving one or two items from prior topics. Based on my results, suggest a spaced plan for the next 7 days (1d, 3d, 7d) with varied formats (explain/compute/apply) if applicable.

✅ Teach to learn: explain to AI and let it push back

When you explain a concept in your own words, AI can ask questions to help you spot mistakes or gaps in your understanding.

Why it works: Self-explanations and elaborative "why/how" prompts can deepen understanding and support knowledge transfer (Chi et al., 1994; Pressley et al., 1987).

Prompt: Act like a curious student while I systematically teach [concept] over a series of messages. Interrupt whenever I skip reasoning with targeted 'why/how do you know?' questions - proceed as far as is reasonable. When I indicate that I'm done teaching, restate my explanation in your own words and list any misconceptions or gaps.

✅ Remove the training wheels: from guided to independent practice

You can use AI to learn by following examples, then trying problems with less help, and finally solving them on your own to build confidence and skill.

This process is especially powerful in STEM, quantitative social sciences, structured language tasks, and any domain where stepwise mastery is needed.

Why it works: Worked examples can reduce cognitive load; fading can build independence; varied/interleaved practice can enhance transfer ( Sweller , 1988; Renkl & Atkinson, 2003; Rohrer & Taylor, 2007).

Prompt: Teach me [problem type] using a 3 step progression, one message at at time. First one fully worked example with brief justifications that I'll confirm I understand. Then, one faded example with 1–2 steps with hints for me to try on my own (I'll respond with my answer, then you should give minimal, immediate feedback). Finally, an independent problem that I'll try on my own. Repeat with two more independent problems (one at a time) with mixed structures. Vary surface features but keep deep structure, then lightly interleave with a similar but different type. After each item, list the rule/formula/concept I used and suggest where to verify each definition or rule with an independent source. End with a 1 minute 'what changed?' reflection comparing two similar problems.

✅ Try, check, and reflect: use AI to learn in a cycle

When using AI to practice a concept or new skills, you can always give it your best attempt, get feedback from the AI system, and then keep track of what you got right or wrong to improve your learning.

Why it works: Prediction and confidence calibration correct illusions of knowing; "desirable difficulties" drive long-term retention ( Koriat , 1997; Bjork & Bjork, 2011).

Prompt: For [topic], use a predict commit check process for 6 items (ideas, questions, or tasks). For each one: (1) ask me for my answer and my confidence (0–100%) before you share the correct or model answer, (2) reveal the answer with a brief explanation or reasoning, (3) prompt me to write a one sentence reflection or lesson learned if I didn't get it right. Keep a running error/insight log with columns: Item → Pattern (error or insight) → Possible Cause → Next Step . (4) once I answer with a reflection, move on to the next. If I do well (get three questions in a row correctly), gently increase the challenge; if I struggle (three in a row missed), decrease the challenge.

Sources

Bjork, R. A., & Bjork, E. L. (2011). Making things hard on yourself, but in a good way . New Theory of Disuse / Desirable Difficulties.

Cepeda, N. J., et al. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis.Psychological Bulletin .

Chi, M. T. H., et al. (1994). Self-explanations: How students study and use examples in learning to solve problems.Cognitive Science .

Dunlosky, J., et al. (2013). Improving students’ learning with effective learning techniques.Psychological Science in the Public Interest .

Koriat , A. (1997). Monitoring one’s own knowledge: A cue-utilization approach.Psychological Review .

Kornell, N., & Bjork, R. A . (2009). A stability bias in human memory.Journal of Experimental Psychology: Learning, Memory, and Cognition .

Renkl, A., & Atkinson, R. K. (2003). Structuring the transition from worked examples to problem solving.Educational Psychologist .

Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning.Journal of Experimental Psychology: General .

Rohrer, D., & Taylor, K. (2007). The shuffling of mathematics problems improves learning.Applied Cognitive Psychology .

Sweller , J. (1988; 1994). Cognitive load theory and instructional design implications . Instructional Science ; Learning and Instruction .

Additional resources

Find your AI learning path with the AI Skills Navigator

Learn more about Microsoft'sresearch on appropriate reliance

Read the AETHER GenAI learning outcomes review

Facebook LinkedIn Email

Need more help?

Want more options?

Explore subscription benefits, browse training courses, learn how to secure your device, and more.

Microsoft 365 subscription benefits

Microsoft 365 training

Microsoft security

Accessibility center