The most effective homework companion does something counterintuitive: it refuses to answer the student's question. Instead, it asks a better one. When a student types "What's the answer to question 3?" a well-built companion responds with "What have you tried so far?" or "Which part of the framework from this week's lesson applies here?" This is Socratic tutoring — guiding students toward understanding through questions rather than handing them conclusions. You can build one as a Custom GPT in about an hour, and the result is an always-available study partner that helps your students do the thinking instead of skipping it.
What you’ll walk away with:
- An AI homework companion helping without giving away answers
- Scaffolding rules guiding students to discover solutions
- A monitoring system for where students get stuck most
Why AI homework companions
The gap between "I taught it" and "they learned it" is where assignments live. But most students do their assignments alone, at odd hours, without access to you or their classmates. When they get stuck, they either give up or look up the answer — neither of which produces learning. A homework companion fills that gap with guided support that is available at midnight on a Tuesday, patient enough to ask the same clarifying question five different ways, and constrained enough to never just hand over the answer.
This is the idea behind Ruzuku's Course Ladder approach to homework companions: interactive prompts that walk students through the hard parts of an assignment without doing the work for them. The AI handles the process — asking what they've tried, pointing them toward relevant concepts, suggesting they look at the problem differently — while you design the destination. You decide what students should produce and what understanding they should demonstrate. The companion handles the coaching conversation that gets them there.
Step by step: Building a homework companion
Define the assignment clearly
The companion can only guide students through work that is well-defined. Before you touch ChatGPT, write out the assignment in full: what students need to produce, what concepts they should apply, what a strong submission looks like, and what a weak one looks like. If you cannot articulate the difference between a thoughtful response and a surface-level one, the companion will not be able to either. The system prompt needs this level of specificity to know which questions to ask and which directions to nudge students toward.
Write the companion’s role — guide, not answer-giver
The system prompt is where everything happens. Start by defining the role explicitly: "You are a homework companion for [course name]. Your job is to help students complete [specific assignment] by asking guiding questions, never by providing the answer directly. When a student asks for help, respond with a question that helps them think through the problem themselves." This framing matters. Without it, the GPT defaults to being helpful in the way ChatGPT is always helpful — by answering the question as fully as possible.
Create Socratic prompting rules
Tell the companion exactly how to guide. Include rules like: "When a student says they are stuck, ask what they have tried so far. When a student asks for the answer, ask which concept from the lesson they think applies. When a student gives a partial answer, acknowledge what they got right and ask a follow-up that leads them to the missing piece." The more specific your rules, the more natural the conversation feels. Generic instructions like "be Socratic" produce stilted, repetitive questioning. Specific instructions like "when the student mentions X, ask them to consider Y" produce conversations that feel like working with a real tutor.
Set boundaries on what it can and cannot reveal
Decide in advance what the companion is allowed to share freely and what it must withhold. It should freely share definitions, restate concepts from the lesson, and point students to relevant sections of the course material. It should not provide complete answers, write any part of the student's submission, or confirm that a student's answer is "correct" without asking them to explain their reasoning. Write these boundaries into the system prompt as explicit rules: "Never provide a complete answer to any assignment question. Never write text that the student could copy directly into their submission."
Build it as a Custom GPT
In ChatGPT, go to "Explore GPTs" and click "Create." Paste your system prompt — the role definition, Socratic rules, and boundaries — into the Instructions field. Upload your assignment document and any relevant course materials (lesson notes, key frameworks, rubrics) as knowledge files so the companion can reference them. Give it a clear name that students will recognize, something like "[Course Name] Assignment Companion" rather than a clever title they will not find later. Set the sharing to "Anyone with the link."
Test with sample student responses
Before you share the companion, test it yourself. Pretend to be three different students: one who asks directly for the answer, one who shares a partial attempt and asks for feedback, and one who is confused about the concepts. Does the companion guide each of them appropriately? Does it hold its boundaries when pushed? Does it avoid giving away the answer even when asked in different ways? Adjust the system prompt based on where it breaks. Most companions need two or three rounds of testing before the Socratic behavior feels natural rather than mechanical.
Share alongside the assignment
Introduce the companion as part of the assignment, not as an afterthought. Tell students what it does, what it will not do, and how to use it. Something like: "This companion is here to help you work through the assignment. It will ask you questions to guide your thinking, but it will not give you the answer. If you get stuck, start by telling it what you have tried so far." Framing matters — students who understand the companion's purpose use it more productively than students who discover it on their own and expect it to do their homework.
Prompts to try
Copy and paste these into the Instructions field when building your Custom GPT, replacing the bracketed text with your details.
- System prompt for Socratic companion: "You are a homework companion for [course name], helping students complete [assignment description]. Never provide the answer directly. When a student asks a question, respond with a guiding question that helps them reach the answer through their own reasoning. Reference concepts from the uploaded course materials. If a student says they are stuck, ask what they have tried and which lesson concepts they think are relevant. Always acknowledge partial understanding before guiding further."
- Hint-not-answer rules: "When a student asks for the answer to a specific question, say: 'I can't give you the answer, but I can help you find it. Let me ask you this instead —' and follow with a question that isolates the concept they need. If a student submits text and asks 'Is this right?', respond by asking them to explain their reasoning rather than confirming or denying. You may provide examples that illustrate a concept, but never examples that directly mirror the assignment question."
- Progress tracking prompt: "After each exchange, briefly summarize what the student has figured out so far and what remains. Frame it as encouragement: 'You've identified that [concept] applies here, and your thinking about [aspect] is on track. The piece you haven't addressed yet is [remaining element] — what do you think might be relevant there?' This helps students see their own progress and stay motivated through difficult problems."
The human layer
The companion guides the process, but you design what students are working toward. The assignment quality determines the companion quality. A vague assignment — "Reflect on this week's lesson" — gives the companion nothing specific to guide toward. A well-designed assignment — "Apply the three-part framework from Module 4 to a real situation in your practice and explain which element was hardest to implement and why" — gives the companion a clear destination and specific concepts to reference when students need direction.
This means the real work is upstream. Before you build the companion, build the assignment. Define what understanding looks like. Identify the common sticking points. Anticipate where students will try to take shortcuts. Write all of that into the system prompt, and the companion becomes a reliable extension of your teaching — not a replacement for it, but a way to be present in the moments when students are doing the hardest work alone.
Course creator tips
Start with your hardest assignment
Build your first companion for the assignment that generates the most questions and the most frustration. That is where students need guidance most, and it is where you will see the clearest benefit. Once you have a companion that works well for the hard assignment, adapting it for simpler ones is straightforward — you already know the prompting patterns that work.
Include your rubric in the knowledge files
If you have a rubric or evaluation criteria for the assignment, upload it alongside the course materials. The companion can then reference the criteria when students ask "Am I on the right track?" — not by grading their work, but by pointing them to the specific criterion they have not yet addressed. This is more useful than a generic "keep going" and more appropriate than a premature "looks good."
Review the companion's conversations periodically
You cannot see individual conversations in a shared Custom GPT, but you can ask students to share screenshots of exchanges that helped them or frustrated them. This feedback shows you where the Socratic approach is working and where it is not. Some students will report that the companion's questions helped them see the problem differently. Others will say it felt like talking to a brick wall. Both responses tell you something about the system prompt that needs adjusting.
What it gets wrong
The most common failure: the companion gives answers instead of guiding. No matter how carefully you write the system prompt, some questions will slip through the boundary. A student who asks "Can you explain the difference between formative and summative assessment?" will likely get a clear explanation — which is fine for a concept question but defeats the purpose if the assignment asks them to articulate that difference in their own words. Test your boundaries against the specific questions your assignment generates, not just generic "give me the answer" requests.
The second failure is rigidity. Some companions apply the Socratic method so mechanically that every response is another question, even when the student has clearly demonstrated understanding and just needs confirmation. A good tutor knows when to say "Yes, you've got it" and move on. Build flexibility into the system prompt: "If the student demonstrates clear understanding of the concept, acknowledge it directly and ask if they want to move to the next part of the assignment."
The third failure is frustrating students who need direct help. Not every student is stuck because they have not thought hard enough. Some are stuck because they missed a foundational concept or misunderstood the assignment. A companion that responds to real confusion with more questions creates frustration, not learning. Include an escape valve: "If a student has been stuck on the same concept for more than three exchanges, offer a simplified explanation of the relevant concept and then return to guided questioning."
Frequently asked questions
Will students just find ways to trick the AI into giving them the answer?
Some will try. A well-written system prompt resists most attempts, but determined students can sometimes get around the guardrails — just as they can find answers by searching online or asking a classmate. The deeper question is whether the assignment is designed so that having the answer without understanding it is useless. If the assignment asks students to apply a concept to their own situation, the "answer" is different for every person. Design assignments where the thinking is the point, and the temptation to shortcut disappears.
How is this different from just telling students to use ChatGPT for help?
When students use ChatGPT directly, it defaults to answering their question as completely as possible. A homework companion has a system prompt that constrains the behavior — it asks follow-up questions, gives hints instead of solutions, and guides students through a reasoning process rather than handing them the endpoint. The difference is between a tutor who does the work for you and a tutor who helps you do the work yourself. The system prompt is what creates that distinction. And when students are ready to submit their work, Ruzuku's exercise submissions let you review and respond to each student's work directly within the lesson.
Does this work for courses that are not academic or skills-based?
Yes, but you need to redefine what "homework" means. In a coaching course, the assignment might be a self-reflection exercise where the companion asks deepening questions. In a creative writing course, it might be a brainstorming partner that pushes students past their first ideas without writing for them. The Socratic approach works wherever you want students to arrive at their own insight rather than receive yours. The companion simply adapts its guiding questions to whatever the assignment asks students to produce.
Assignments need a home
A homework companion guides the thinking, but the assignment itself needs to live somewhere students can submit their work and hear back from you. On Ruzuku, exercises are built into each lesson — students complete the work in context, submit it, and get your personal feedback in the same place they learned the material.
That structure matters. When the companion points a student back to "the framework from this week's lesson," they should be able to find it in one click — not buried in an email attachment. Ruzuku's step-by-step course builder keeps assignments, lessons, and discussions together so the whole learning experience feels connected.
Related guides
- How to Build a Custom GPT for Your Course — the foundation for building any course-specific GPT, including knowledge uploads and system prompts
- How to Create an AI Study Companion for Your Course Students — a broader study assistant for answering questions, complementary to assignment-specific companions
- How to Create Practice Exercises Using AI — generate the exercises that your homework companion helps students work through
- How to Track Student Progress Using Google Sheets — monitor assignment completion alongside your companion's guided support
- Ruzuku for Coaches — build a course with feedback-driven exercises and personal coaching support