On this page
Halfway through a hard problem, every student wonders: am I giving up too early, or wasting time? The answer is not a number of minutes. It depends on what you have tried, and whether your attempts are still teaching you anything.
Why should you struggle with a problem before asking AI?
Because the attempt itself prepares you to learn. In a 2021 meta-analysis in Review of Educational Research, Tanmay Sinha and Manu Kapur compared 53 studies (166 comparisons) of two orders: solve first and get taught after, or get taught first and solve after. On average, trying first won, with a moderate effect (Hedges' g of 0.36). There were exceptions: young children in grades 2 to 5, and lessons on general skills rather than subject content, did better when taught first. We cover the details in our post on what the productive failure meta-analysis found.
Asking too early has a cost too. In a 2025 PNAS field experiment with nearly a thousand high school maths students, those given a standard ChatGPT-style tutor got 48% higher grades on practice. Once the AI was taken away, they scored 17% lower than students who never had it. A second tutor, whose prompts were designed to safeguard learning, largely avoided that drop. Our post on whether leaning on AI weakens learning looks at more of this research.
How long should you try a problem before asking AI?
Our rule of thumb is two genuine attempts, then the smallest hint. A genuine attempt means something new on the page: a diagram, a list of knowns and unknowns, an equation, or a method followed until it clearly fails. Rereading the question five times does not count. Neither does the same method with different arithmetic.
Before your second attempt, run this quick check:
Underline exactly what the question asks for.
Write down every given quantity with its unit.
Draw the situation, even roughly.
Name the chapter or law it probably comes from.
Try a simpler case: drop friction, use round numbers, or set one variable to zero.
If both attempts fail and the check gives you no new idea, stop. More time alone probably won't help.
What are the signs you are stuck, not struggling usefully?
Researchers call unproductive effort wheel-spinning: spending too much time struggling with a topic without reaching mastery. Describing Beck and Gong's 2013 study in a 2018 Journal of Educational Data Mining paper, Shimin Kai and colleagues report that about 38% of students had not mastered a maths skill after 10 practice opportunities.
You are probably stuck if:
your last attempts repeat one method with small changes
you cannot say in one sentence what the question wants
you are missing a fact or formula, which no amount of thinking will produce
you keep getting different answers from the same set-up and cannot find the error.
You are struggling usefully if each failed attempt tells you something, and you can say why it failed.
How do you ask AI for the smallest hint?
Climb a hint ladder one rung at a time, pasting your working each time:
"Don't solve this. Which concept or law applies here, and why?"
"Here is my working. Is my first step right? Don't give me the next one."
"Give me only the next step, then stop."
"Show the full solution." Then close it and redo a similar problem the next day without help.
Kai's team found two wheel-spinning patterns. Some skipped hints on at least one problem, yet asked for more than one bottom-out hint (the kind that reveals the answer) within 8 problems. Others skipped hints on some problems and moved quickly between problems on the same skill. The authors suggest spaced practice and sparing use of bottom-out hints. Our post on hints versus answers in tutoring research covers more of this research.
How does the rule work on a real problem?
A ball is thrown horizontally at 15 m/s from a cliff 20 m high. How far from the base does it land? Attempt one uses the range formula u² sin 2θ / g. With θ = 0 that gives zero, which is clearly wrong. Attempt two tries v² = u² + 2as and gets a speed, not a distance. Two genuine attempts, and you are stuck.
Rung 1 is enough: "horizontal and vertical motion are independent; the drop sets the time." Now you can finish alone. From 20 = ½ × 9.8 × t², t ≈ 2.02 s, so the distance is 15 × 2.02 ≈ 30.3 m.
This is the kind of help Fermi's AI tutor is built for. It notices where a student is stuck and decides whether to explain, ask or show. To check a method afterwards, the Fermi Problem Library has a step-by-step solver that explains the reasoning rather than printing an answer.
What can parents do?
Before helping, ask "What have you tried?" Two real attempts on the page earn a hint, from you or from AI. A blank page earns the five-point check above. Tonight, try the rule on one hard problem and notice which rung of the ladder you needed.
