On this page
"Wrong. The answer is 20.4 m. Good effort!" That is feedback, technically. It is also nearly useless. When you ask an AI to check your physics solution, the kind of feedback you ask for decides whether you learn from it.
Does feedback always help?
No. A 1996 meta-analysis by Kluger and DeNisi, covering 607 effect sizes, found feedback improved performance on average (d = 0.41), but over a third of feedback interventions actually decreased performance. Their analysis suggested feedback works less well as attention moves away from the task and towards the self.
A 2020 meta-analysis by Wisniewski, Zierer and Hattie of 435 studies found a medium average effect (d = 0.48), with large variation, and found that the impact is substantially influenced by "the information content conveyed". Feedback that tells you something useful tends to work. Feedback that only grades or praises you does much less.
What does good feedback look like?
A clear framework comes from John Hattie and Helen Timperley's 2007 review, "The Power of Feedback". They say effective feedback answers three questions: "Where am I going?", "How am I going?" and "Where to next?"
Each can work at four levels:
Task: is this step right or wrong, and what is missing?
Process: what strategy or method is needed to do this kind of task?
Self-regulation: how can you monitor and check your own work?
Self: praise about you as a person, such as "you're so smart".
Hattie and Timperley argue that feedback about the self is the least effective, and that process and self-regulation feedback are powerful for deep processing and mastery. Two more points from the paper matter for students. Task feedback is "more powerful when it is about faulty interpretations, not lack of information." And "if students lack necessary knowledge, further instruction is more powerful than feedback information."
That last point is practical. If you have no idea how to start, asking for feedback is the wrong request. Ask to be taught first.
A worked example: one solution, two kinds of feedback
Problem: A ball is thrown straight up at 20 m/s. Find the maximum height. Take g = 9.8 m/s².
Student's solution:
At the top, v = 0.
v² = u² + 2as, so 0 = 400 + 2(9.8)s.
s = −400/19.6 = −20.4 m. Height is 20.4 m.
The final number happens to be right, but step 2 is wrong. With upward taken as positive, the acceleration is −9.8 m/s², not +9.8. The student got a negative height and dropped the sign to match the expected answer.
Weak feedback: "Correct, 20.4 m. Well done!" This checks only the final answer, adds empty praise, and leaves a sign error in place that will cost marks on the next problem.
Useful feedback, at three levels:
Task: "In step 2, a should be −9.8 m/s² if up is positive. Then 0 = 400 − 19.6s, so s = +20.4 m."
Process: "Before any kinematics equation, write your positive direction and give every vector a sign that matches it."
Self-regulation: "A negative height for a ball thrown up is a signal to stop. Next time, check whether the sign of your answer makes sense before you remove it."
Prompts that get feedback at the right level
Use these with any AI tool after you have written out your full solution:
Feed up: "Before checking my work, tell me what a complete solution to this problem needs to show."
Task: "Find the first step in my solution that is wrong, if any. Quote the step and explain the error. Don't solve the rest."
Process: "What general strategy should I use for problems like this one?"
Self-regulation: "What check could I have done myself to catch this mistake?"
Feed forward: "Give me one similar problem to try next that targets this mistake."
Skip "Is my answer right?" on its own. As the example shows, a right answer can hide a wrong method.
A checklist for judging the feedback you get
Does it point to a specific step, not just the final answer?
Does it explain why, not just what?
Does it give you a method or check you can reuse?
Does it tell you what to do next?
Is it free of empty praise?
If the feedback fails most of these, rephrase your request using the prompts above. And remember that AI can be wrong about your work too, so check its corrections with units and a quick re-derivation; our guide on when AI gets physics wrong shows how.
The Fermi Problem Library has worked problems with a step-by-step solver that explains the reasoning rather than printing an answer, which makes it easier to compare your steps with a worked route. For why errors like this one are worth making, read why mistakes help you learn.
