On this page
When you are stuck on a problem, the quickest way out is to ask for the answer. An AI chatbot will give it to you in seconds, with every step written out. You read it, it makes sense, and you move on. The trouble is that following a solution and producing one are different skills. This post looks at what tutoring research says about how much help to ask for, and gives you a simple ladder to use every time you are stuck.
What happens when students get answers instead of hints?
The clearest recent evidence comes from a field experiment published in PNAS by Bastani and colleagues (2025). Nearly a thousand high school maths students practised with one of two GPT-4 tutors. "GPT Base" behaved like a standard chatbot. "GPT Tutor" had prompts designed to protect learning, giving hints rather than full answers. During practice, grades went up by 48% with GPT Base and 127% with GPT Tutor. But when access was taken away for the exam, the GPT Base group did 17% worse than students who never had AI. Those negative effects were largely avoided with GPT Tutor. The authors describe students using the plain chatbot as a "crutch".
Is this a new problem?
No. Researchers saw the same pattern with computer tutors long before chatbots. Many intelligent tutoring systems offer hints in levels, ending with the full answer. Baker and colleagues (2004) found that "gaming the system", for example clicking through hints quickly to reach the answer, was associated with substantially lower post-test scores, even after controlling for prior knowledge.
A review by Aleven and colleagues (2003) found that good help-seeking is linked to better learning, but that learners often do not use help well. Some avoid help when they need it; others ask for too much too soon.
What kind of help works best?
Step-level help seems to matter. In a review of tutoring experiments, VanLehn (2011) found that human tutoring had an effect size of d = 0.79 compared with no tutoring, and intelligent tutoring systems were close behind at 0.76. Most of those systems give feedback and hints on each step of a solution, not just on the final answer.
There is also evidence that struggling first can help, as long as instruction follows. In Kapur (2014), students who tried maths problems before being taught the concept ended up with greater conceptual understanding and better transfer to new problems than students taught first, with similar procedural skill. Our post on Vygotsky's zone of proximal development explains why help pitched just above what you can do alone is the most useful kind.
A help ladder for when you are stuck
Use these rungs in order. Only climb when the one below has not worked.
Try for a set time. Give the problem a real attempt, say ten minutes, and write down what you tried.
Ask what principle applies. "Which law or idea should I start from?" not "What's the answer?"
Ask for the next step only. Show your work so far and ask what to do next, then continue on your own.
Ask what is wrong with your step. If you have a wrong answer, ask the tutor to find the first incorrect line, not to redo the whole thing.
Get a full worked solution, then redo it. If you do need the whole solution, close it and solve the problem again from a blank page the next day.
Here is what rung 3 looks like with a JEE-style mechanics problem. A block of mass 2 kg is pulled along a rough floor by a 20 N horizontal force, with coefficient of kinetic friction 0.5. You have written F − f = ma but are stuck. A good next-step hint is: "What is the normal force, and how does it set the friction?" From there you can find N = mg = 2 × 10 = 20 N, f = 0.5 × 20 = 10 N, and a = (20 − 10)/2 = 5 m/s², taking g = 10 m/s². You did the physics; the hint only pointed.
How Fermi's AI tutor handles this
The design question in the PNAS study, whether the tutor gives answers or teaches, is the one that matters when you choose a tool. Fermi's AI tutor holds a line of reasoning across a whole session, notices where you are stuck and decides whether to explain, ask or show, so you are pushed to take the next step yourself. You can still ask for more help when you need it; use the ladder above to decide when. Our comparison of AI tutors and answer chatbots goes into the difference.
The next time you are stuck, ask for the next step only, and write the rest of the solution yourself.
