Use AI to develop skills through practice

Before you start

This lesson is designed for students in Years 10 to 12. It will take approximately 25 minutes to complete. Before you begin, you should understand the difference between passive recognition (like re-reading notes or looking directly at solutions) and active retrieval (effortfully recalling information from your own memory). You will need a notebook, a pen, and access to any AI chat tool to use as a study partner.

Related workflows: browse every supported practice test type, AI writing feedback on your work, Mistake Book spaced review.

What you'll learn

• Explain how unrestricted AI use leads to "borrowed competence" and a drop in unassisted exam performance.
• Apply the "Produce First, Prompt Second" rule to both STEM and humanities study sessions.
• Configure an AI tool to act as a Socratic critic rather than an answer generator.
• Design a 40-minute active study session that leverages AI feedback without triggering cognitive offloading.

Core concepts

Many students fall into the "Homework Trap." You paste a complex prompt into an AI tool, receive a beautifully written response, and submit it. Your homework marks look brilliant, but your exam scores plummet. This is the performance paradox of modern study.

Recent research highlights this danger. A study by Bastani et al. (PNAS 2025) found that unrestricted GPT-4 access boosted practice math scores by 48% but led to a 17% drop in subsequent unassisted exam performance compared to no-AI controls. Similarly, Barcaui (2025) showed that undergraduates using unrestricted ChatGPT scored 11 percentage points lower (57.5% vs 68.5%) on a surprise retention test 45 days later than traditional study groups. These students spent 45% less time studying, which directly correlated with fragile memory formation. Furthermore, Strömberg, Lei, and Wu (CEPR 2026) conducted a 30-month study of 26,811 secondary students and found that AI adoption raised homework scores by 18% but lowered monthly closed-book exam scores by 20% within six months.

The Cognitive Mechanisms of AI Over-Reliance

Why does this happen? When you let AI do the work, you trigger several negative cognitive mechanisms:

• Cognitive Offloading: Outsourcing mental effort to an external tool, which prevents your brain from building the neural pathways required for long-term memory.
• Metacognitive Laziness: As identified by Fan et al. (2025), this is the tendency to confuse the ease of reading an AI's fluent response with your own actual understanding.
• Cognitive Surrender: Shaw and Nave (Wharton 2026) found that students stop independent reasoning more than 50% of the time when AI is available, even when the AI provides incorrect answers.

To combat this, we must preserve "desirable difficulties"—the effortful mental processes that make learning stick. The Australian Curriculum (ACARA) emphasizes critical and creative thinking capabilities, which require you to actively analyze, synthesize, and evaluate information yourself.

The Rules of Learn-Better-with-AI

To build true skills, you must follow these guidelines:

1. Never let the AI generate the first draft or initial solution. Always attempt the task solo first to trigger retrieval.
2. Use AI as a Socratic critic rather than an answer oracle. Ask it to identify logical gaps in your completed work.
3. Treat AI-assisted fluency as a warning sign of "borrowed competence" and schedule a surprise self-test 48 hours later.
4. Prompt the model to withhold direct answers and instead provide progressive hints or scaffolded questions.

To measure your unassisted performance across various exam formats, you can practice with past exam papers under timed conditions.

Worked examples

Example 1: Year 11 History Essay Outline

A Year 11 student is studying the causes of World War I. Instead of asking AI to write the outline, they use an active loop.

1. The student writes their own outline from memory, noting causes like militarism, alliances, and the assassination of Archduke Franz Ferdinand.
2. They open the AI tool and paste their outline.
3. They prompt the AI: "Review my essay outline on the causes of World War I. What historical perspective or structural link did I miss? Do not rewrite the outline for me."
4. The AI points out that the outline lacks focus on the domestic political pressures within Austria-Hungary.
5. The student closes the AI tab and rewrites their outline from memory, incorporating the new perspective.

Example 2: Year 12 Math Methods Calculus Problem

A Year 12 student gets stuck on a complex optimization problem involving derivatives.

1. The student attempts the problem on paper, writing down the volume formula V = x(10-2x)(16-2x) and taking the derivative dV/dx, but gets an incorrect critical value.
2. They copy their step-by-step working into the AI tool.
3. They prompt the AI: "Identify the specific step where my algebraic expansion or derivative logic breaks, but do not give me the correct numbers or the final answer."
4. The AI responds: "In step 2, when expanding the quadratic term, you multiplied -2x by -2x and got -4x^2 instead of +4x^2. Check your signs there."
5. The student closes the AI tool, corrects the sign error on their paper, and completes the calculation independently.

If you want to refine your writing style after drafting, getting feedback on your work can help you spot structural weaknesses without giving away the answers.

In tests and exams

In school assessments, you do not have access to AI. Exams test your unassisted retrieval pathways. Closed-book essay writing under timed conditions (e.g., NSW HSC English, VCE English) requires you to recall quotes, structure arguments, and write fluently without external prompts. Unassisted multi-step problem solving in STEM exams (e.g., QCE Math Methods, HSC Physics) tests your ability to recognize which formula to apply and execute algebraic steps without a digital safety net. Finally, source analysis and critical evaluation in humanities (e.g., VCE History, HSC Modern History) rewards deep, authentic reasoning and personal voice—skills that are severely eroded by metacognitive laziness.

To keep track of the errors you make during these unassisted sessions, maintain a digital log or mistake book for spaced review to ensure you do not repeat them under exam pressure.

Practice

1. Warm-up 1: Diagnose this student's study diary entry: "I had a history essay due. I pasted the prompt into ChatGPT, read the essay it generated, highlighted the key terms to help me remember them, and submitted it." Identify two passive study habits in this entry.
2. Warm-up 2: Rewrite this passive prompt to make it an active, Socratic prompt: "Write a paragraph explaining photosynthesis for my Year 11 Biology class."
3. Standard 1: Design a 40-minute active study session script using these four timeline segments: Solo Retrieval Try (15 min), AI Critique & Gap Analysis (10 min), Active Re-drafting (10 min), and Socratic Flashcard Prep (5 min).
4. Standard 2: A student is preparing for a chemistry exam on stoichiometry. They plan to read the textbook chapter and then ask AI to solve five practice problems so they can study the solutions. Convert this into an active retrieval-based plan.
5. Challenge 1: Write a comprehensive Socratic prompt template that you can use when you get stuck on a multi-step physics problem (e.g., involving F = ma). The prompt must force the AI to act as a strict tutor who only provides one hint at a time.

Answers

1. Warm-up 1 Solution: The two passive habits are: (1) Cognitive offloading—letting the AI generate the essay draft instead of writing it independently; (2) Passive recognition—highlighting the generated text instead of actively retrieving or testing memory. This creates an illusion of competence.
2. Warm-up 2 Solution: An active, Socratic prompt would be: "I am studying photosynthesis for Year 11 Biology. I will write my own explanation first. Please act as a Socratic tutor. Once I paste my explanation, point out any missing steps or misconceptions, but do not write the explanation for me."
3. Standard 1 Solution: A highly effective 40-minute script:
   • Minutes 0-15 (Solo Retrieval Try): Attempt a past exam question under timed, closed-book conditions. Write in blue pen; use red pen to mark gaps you cannot remember.
   • Minutes 15-25 (AI Critique & Gap Analysis): Paste your attempt into the AI. Prompt: "Act as a strict examiner. Critique my logic and point out gaps, but do not rewrite it for me."
   • Minutes 25-35 (Active Re-drafting): Close the AI window. Rewrite the response on paper, incorporating the feedback entirely from memory.
   • Minutes 35-40 (Socratic Flashcard Prep): Ask the AI to generate 3 highly specific retrieval questions based on your weak areas for tomorrow's review.
4. Standard 2 Solution: The active plan: (1) The student attempts five stoichiometry problems from a past paper completely closed-book. (2) If stuck, they paste their incomplete working into the AI and ask: "Identify the specific step where my mole ratio calculation went wrong." (3) After correcting the step, they complete the calculation. (4) They add the mistake to their review list and re-test themselves on a similar problem 48 hours later.
5. Challenge 1 Solution: A strong Socratic prompt template: "Act as an expert physics tutor. I am trying to solve a problem involving F = ma but I am stuck. I will paste my problem and my current working. Do not give me the answer, the formula to use, or the correct numbers. Instead, ask me one guiding question that helps me identify my own error or determine the next logical step. Wait for my response before giving another hint."

Common mistakes

• Mistake: Pasting the entire prompt into ChatGPT to get a quick answer.
  Fix: Write your own outline first, then ask AI to critique your structure.
• Mistake: Confusing the AI's clear explanation with your own understanding.
  Fix: Close the tab and write down the explanation from memory.
• Mistake: Using AI to write essays because it "saves time."
  Fix: Use AI to brainstorm counter-arguments, but write the actual sentences yourself.

Quick recap

• Unrestricted AI use leads to "borrowed competence" and a drop in unassisted exam performance (up to 17-20%).
• Active retrieval and "desirable difficulties" are essential for building long-term memory.
• Always produce your first draft or attempt a problem solo before opening an AI tool.
• Prompt AI to act as a Socratic critic, pointing out logical gaps or identifying specific steps where your math logic breaks.
• Close the AI window and rewrite or solve the problem from memory to secure the neural pathways.

To build the mental stamina required for high-stakes assessments, you must practice under real-world constraints. You can find more resources on active learning and critical thinking at the Australian Curriculum website.

Related workflow

Start timed practice mocks and section drills