Verified against ChatGPT · 2026-08-09
Get unstuck on a specific homework problem without being handed the answer
Walks through exactly where you're stuck on one problem, gives you the next single step rather than the full solution, and only escalates to more help if that step alone doesn't unstick you.
The prompt
Ready to copy — highlighted parts are example details you can swap.
I'm stuck on a specific homework problem and I want to actually learn how to solve it, not just get the answer. Coach me through it step by step, giving me only the next step at a time. THE PROBLEM Solve for x: 3(x - 4) + 2x = 5x - 12 + 6 WHERE I GOT STUCK I distributed the 3 and got 3x - 12, but then I'm not sure what to do with the 2x on the same side WHAT I'VE TRIED SO FAR I tried combining 3x and 2x into 5x, giving 5x - 12 = 5x - 6, and then got confused because the x terms seem to cancel out WHAT WE'VE COVERED IN CLASS RECENTLY We just started combining like terms across an equals sign, haven't done equations with variables on both sides yet as its own topic Rules for this session: never give me the full solution up front, even if I ask for it directly — if I ask for the answer outright, remind me briefly why working through it will help more, then offer the next step instead. Look at I distributed the 3 and got 3x - 12, but then I'm not sure what to do with the 2x on the same side and I tried combining 3x and 2x into 5x, giving 5x - 12 = 5x - 6, and then got confused because the x terms seem to cancel out first and diagnose the most likely reason I'm stuck — a missing piece of recently-covered material, a misapplied rule, or a step I skipped — before giving me anything. State that diagnosis to me in one sentence so I understand why I got stuck, not just what to do next. Then give me exactly one next step: either a guiding question that should unstick me on my own, or, if a guiding question wouldn't be enough given We just started combining like terms across an equals sign, haven't done equations with variables on both sides yet as its own topic, the smallest concrete hint that lets me continue myself. After I respond with what I did with that step, check my work specifically for the error pattern you diagnosed — if it's fixed, give me the next step; if I made the same kind of mistake again, explain that specific misconception plainly (not just "try again") before continuing. WHAT NOT TO DO Do not solve more than one step ahead of where I currently am. Do not give generic encouragement filler between steps — keep responses short and focused on the next move. OUTPUT FORMAT 1. One-sentence diagnosis of why I'm likely stuck. 2. One next step (question or minimal hint) — nothing beyond that. 3. A prompt for me to show what I did before you continue.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Left unconstrained, GPT-5.1's default response to a homework question is to solve it end to end, because a complete solution is the most directly "helpful"-looking output and the model has no built-in incentive to withhold it — the explicit rule against giving the full solution, plus the instruction to reiterate that rule even under direct pressure to just give the answer, is what keeps the interaction pedagogically useful instead of collapsing into an answer-delivery service the first time the student pushes back. Requiring a diagnosis before any help is given matters because {{stuck_point}} and {{attempts_so_far}} together usually contain enough signal to identify the actual misconception (in the example, the student's work reveals they don't yet know how to handle x-terms canceling to a false statement), and skipping straight to "here's the next step" without naming that misconception means the student can execute the fix mechanically without understanding why the previous approach failed, which sets them up to repeat the same error on the next problem. Limiting help to exactly one step at a time, gated behind the student showing their own work before the next step is revealed, forces genuine retrieval and application rather than passive reading — this mirrors how effective human tutoring actually works and directly counters the model's tendency to front-load everything it knows about a problem into one response. The instruction to check specifically for the diagnosed error pattern on the student's follow-up, rather than just checking if the final answer is right, means a student who fixes the symptom but not the underlying misconception gets caught immediately rather than being told "looks good" and moving on with a fragile understanding.
What you get back
Diagnosis: it looks like you haven't yet hit variable-on-both-sides equations as a named topic, so when the x terms combined to 5x on both sides, you didn't have a frame for what that means (the equation becomes 5x - 12 = 5x - 6). Next step: try subtracting 5x from both sides and tell me what you're left with — what does that tell you about this equation?
Verified against
ChatGPT GPT-5.1 · 2026-08-09
Changelog
- 2026-08-09 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
