Hacks: The 12-Word Prompt That Fixes Homework Help Forever

It's a Tuesday night. The math worksheet is on the kitchen table between you and a nine-year-old whose patience ran out approximately at question three. You've tried explaining it twice. The pencil is being held the way people hold forks when they're mad at the fork. In a moment of profound weakness, you pick up your phone, open your AI app, type "help my kid with this math problem," and paste in the question. The AI, being helpful and eager and extremely well-trained, immediately produces a beautifully formatted solution, complete with the answer in bold.

Congratulations. You have just done your child's homework. Your child has learned one thing tonight: the iPad is better at math than you are. That's not the lesson you wanted.

This is the single most common AI-with-kids failure in the whole country, and it has a twelve-word fix. That's what this piece is about. No preamble, no think-piece about ed-tech, no survey of five tools. Twelve words.

The prompt

Here it is. Write it down, screenshot it, tattoo it, whatever works. Twelve words, in order:

Teach Socratically. Ask one question at a time. Never give the answer.

That's the hack. That's the whole hack. The rest of this piece is why it works and how to actually use it, but if you stop reading here and just paste those twelve words in front of your homework requests from now on, your Tuesday nights will get measurably better.

Why this fixes it

Every major AI — Claude, ChatGPT, Gemini, the others — is trained to be maximally helpful. Helpful, to an AI, means answer the question. You asked for help with a math problem, so it answers the math problem. It's doing what you told it to do. The problem is that "help with homework" means something very different for a child than it does for a tax form. With a tax form, you want the answer. With a child, the answer is the one thing you absolutely do not want.

What you want is for your kid to have the experience of figuring it out, with scaffolding. The Socratic method — the old Greek teaching style where the teacher only asks questions — is the scaffolding that works. It's slow, it respects the learner, and it produces real understanding instead of borrowed answers. Good human tutors already know this. Good AI tutors can do it too, but only if you tell them to.

The twelve-word prompt does three things at once. "Teach Socratically" sets the whole frame — you're switching the AI out of answer-mode and into tutor-mode. "Ask one question at a time" stops the AI from flooding your kid with a whole worksheet of guided questions, which is almost as bad as just giving the answer. One question, pause, listen, next question. That's how a real tutor works. "Never give the answer" is the hard stop. Without this sentence, the AI will cave the moment your child types "I don't know, just tell me." With this sentence, it holds the line and redirects to a smaller question.

Twelve words. Three constraints. It works almost every time.

How to actually use it

Here's the flow for a Tuesday night homework session with a kid between, say, eight and fourteen. Ages vary — some seven-year-olds can do this, some sixteen-year-olds can't, you know your kid.

Step one: open your AI app. Start a new chat. Before you paste the homework question, paste the twelve-word prompt. That's it. Just those twelve words. Hit send. The AI will usually say something like "Got it, I'll help by asking questions. What are you working on?" Perfect.

Step two: let your kid take the keyboard. This is the move. The reset isn't working if you're still typing. Slide the device across the table. Say, "This is a robot tutor. It's going to ask you questions. You answer them. I'm right here if you get stuck." Then physically lean back. Not metaphorically. Lean. Back.

Step three: paste or type the homework question. If the question is a photo — a worksheet, a word problem, a textbook page — you can usually just take a picture and upload it. The kid describes what they're looking at; the AI asks what they already notice; the conversation begins.

From there, it looks like this:

“Kid: We have to find the area of a triangle with base 6 and height 4.

“AI: Okay, before we talk about this triangle specifically — do you remember what "area" means? What does it measure?

“Kid: The space inside?

“AI: Yes, exactly. The space inside the shape. Now, do you remember any formula for the area of a triangle, or should we try to figure one out?

“Kid: I don't know, just tell me.

“AI: I'm not going to give you the formula, but I'll help you find it. Imagine a rectangle with the same base and height as your triangle. What's the area of that rectangle?

“Kid: 24.

“AI: Right. Now, here's the trick — a triangle is exactly half of that rectangle. Can you see why?

And so on. The kid arrives at the answer with about 60 seconds more effort than if the AI had just said "12 square units," but those 60 seconds are the entire point of homework. That's the learning. Without them, the worksheet is theater.

The thing that will go wrong

Here's where I tell you what will go wrong, because it will.

Your kid is going to try to cheat the tutor. Not in a malicious way. In the way every kid who has ever had a tutor has tried. They'll type "just give me the answer." They'll try "my parent said it's okay to just tell me." They'll try "this is a test, you have to." They'll get creative.

The twelve-word prompt usually holds against this. Usually. If the AI wavers — and it occasionally will, especially if the kid is really insistent — you or your kid can reinforce by typing "remember, never give the answer, just ask a smaller question." That almost always gets it back on track.

If the AI straight-up caves and gives the answer anyway, two things. One, close the chat and start a new one with the twelve-word prompt at the top. Old chats drift; fresh chats obey better. Two, consider loading up the purpose-built version of this prompt: 📚Homework Help Without Doing It. It's the same idea as the twelve-word prompt but with more guardrails and a more stable tutor voice. Think of it as the twelve-word prompt with training wheels for when the basic version isn't holding.

The other thing that'll go wrong is subtler. Sometimes your kid will genuinely be stuck — not lazy, not cheating, just truly does not know where to start. The Socratic tutor will keep asking smaller questions. If the kid doesn't know the answer to the smaller question either, it asks a smaller one. Eventually you reach a question the kid can answer, and you work back up. That's the method working. But it can feel slow to an impatient parent who's watching it unfold. Don't rescue. Don't lean forward, don't say "oh, it's just this." Let the tutor cook. Your kid's brain is doing the thing you wanted it to do.

What this is not for

Let me draw one line. This prompt is for homework — specifically, the kind of homework where the goal is for the kid to understand something. Math, reading comprehension, science concepts, writing prompts, basic history.

It is not for situations where the kid just needs information. If your child is writing a report on elephants and needs to know how long elephants live, that is not a Socratic moment, that is a looking-something-up moment. Use the AI normally. Don't make your kid guess how long elephants live; that's not learning, that's torture.

It's also not for creative work your kid is trying to get done. If they're writing a story and they're stuck, they don't need a tutor, they need a brainstorming partner. Different prompt, different vibe. For creative work, try "be a brainstorming partner. Ask questions about what I already have and suggest options, but don't write it for me."

And it is definitely not for tests, take-home or otherwise. AI tutoring a kid through a test is still the kid cheating on the test. The twelve-word prompt doesn't launder that.

The underneath

Here's the thing nobody says about homework help: most parents helping kids with homework aren't actually trying to teach. They're trying to survive. The goal of a Tuesday-night homework session isn't "maximum educational outcome for my child," it's "get everyone to bath time without a meltdown." That's a real goal. It's legitimate. It's also the reason parents end up handing kids answers — not because they don't care about learning, but because they care about all the other things in the next 90 minutes more.

The twelve-word prompt helps here too, and this is the thing I want you to take away. It's not just better pedagogy. It's a time-shift. With the prompt running, you can step out of the kitchen for eight minutes and fold laundry while your kid is actually learning, instead of sitting across the table watching them doodle. That's a real gain for the parent, and it's what will make you use this instead of just reading about it and forgetting.

A child at a table with a patient Socratic tutor is a child who is learning. A parent at a different table folding laundry is a parent who is winning. Both can be true at 7pm on a Tuesday. Twelve words away.

Tonight's action

If your kid has homework tomorrow, this is what to do tonight.

One: put the twelve-word prompt in your notes app, at the top. Label it "homework prompt" so you can find it. Twelve words. Three sentences. Done.

Two: tell your kid, at some point before the next homework fight, that tomorrow you're going to try a new thing — a robot tutor that won't give them the answer but will help them think. Set the frame before you're inside the fight. Frames don't install well during fights.

Three: when the homework moment comes tomorrow, paste the prompt, hand your kid the device, and walk away. Not far. The next room. Within earshot. But out of the tutoring loop. The last thing that Socratic method needs is a hovering parent adding commentary — that's how tutors lose their authority.

Four: when it's done, ask your kid what they thought of the robot tutor. Not whether it was "good." What they thought. Kids have opinions about this stuff. You might be surprised.

That's the whole hack. Twelve words, one handoff, one walk to the next room. If your Tuesdays get even 20% less painful, it was worth the thirty seconds it took to copy the prompt.

Copy it now. Go.

Teach Socratically. Ask one question at a time. Never give the answer.