On this page
ChatGPT can solve a lot of physics problems, which is exactly why it is easy to misuse. Ask it for an answer and you get one, neatly formatted, and you learn very little. Ask it the right way and it becomes a patient practice partner. The difference is in the prompt.
Why does copying AI answers backfire?
In a 2025 study in PNAS, Bastani and colleagues gave nearly a thousand high school maths students access to GPT-4 during practice. With a plain ChatGPT-style interface, practice grades went up 48%, but once access was taken away those students scored 17% lower than classmates who never had it. The authors describe students using it as a "crutch". A version with prompts designed to protect learning largely avoided the drop.
Two older findings help explain why. In a classic 1989 study, Chi and colleagues found that "Good" students studying worked mechanics examples generated many explanations of the solution steps, while "Poor" students generated too few and then relied heavily on the examples. And Roediger and Karpicke (2006) showed that taking a test produced better retention two days or a week later than restudying the same material, even though restudying looked better on a test five minutes later. Reading a finished AI solution is restudying. Explaining and answering is what builds lasting memory.
There is also accuracy. Gerd Kortemeyer found that ChatGPT "would narrowly pass" a calculus-based physics course "while exhibiting many of the preconceptions and errors of a beginning learner". So copying can mean copying mistakes.
Seven prompts that make you think
Copy these, then fill in your problem. Each one keeps you doing the work.
Hint ladder: "Here is a problem I'm working on: [problem]. Don't solve it. Give me one hint at a time, smallest first, and wait for my attempt after each."
Check my step: "Here is my solution so far: [steps]. Don't give the final answer. Tell me the first step that is wrong, if any, and ask me a question that helps me see why."
Predict first: "Before explaining [concept], ask me to predict what happens when [variable] doubles. Then tell me whether my reasoning was right."
Explain back: "I'll explain [concept] in my own words. Point out anything wrong or missing, but don't rewrite my explanation for me."
Quiz me: "Ask me 5 short questions on [chapter], one at a time. After each answer, tell me if I'm right and why. Mix easy and hard."
Sanity check: "I got [answer] for [problem]. Help me check it using units and a limiting case, without solving it from scratch."
Make a twin: "Give me a new problem that tests the same idea as [problem], with different numbers and a slightly different setup. Don't give the answer until I submit mine."
A worked example: a block on an incline
Say the problem is: a block slides down a frictionless incline at angle θ. Find its acceleration.
You use the predict-first prompt. The AI asks: "If the block's mass doubles, what happens to the acceleration?" You guess it halves. The AI asks you to draw the forces. You write gravity mg straight down and the normal force perpendicular to the surface. It asks for the component of gravity along the slope: mg sin θ. Then Newton's second law: mg sin θ = ma, so a = g sin θ. The mass cancels. Your prediction was wrong, and now you know exactly why.
Then the sanity check prompt: at θ = 0 the incline is flat, so a = 0. At θ = 90° the block falls freely, so a = g. Both make sense. That last step is a habit worth keeping; our guide to checking an AI's physics solution shows more of these checks.
Compare that with typing "solve this". You would have read "a = g sin θ", nodded, and kept the wrong belief that mass changes how fast the block speeds up.
Ground rules that keep you honest
Attempt first, then ask for the smallest hint that gets you moving.
Never paste a whole assignment. One problem, one question at a time.
Check every final number yourself, with units.
Write the AI's best hint into your notebook in your own words.
Follow your school's rules on AI. Our post on AI and academic honesty helps you find where help ends and copying begins.
If you want this style built in rather than prompted, the Fermi Problem Library has worked problems with a step-by-step solver that explains the reasoning rather than printing an answer. For more quiz prompts, see how to make an AI quiz you properly.
