Should my 13-year-old be using ChatGPT for homework?

Probably yes — but not in the way most thirteen-year-olds are currently using it. A ban at home does not hold, because the tool is on a friend's phone, in a school Chromebook, and one browser tab behind the homework document. What actually changes the outcome is whether your child can tell when the machine is wrong, and whether they can say out loud which part of the work was theirs.

The useful question is not whether, but at which step

Homework has stages: understanding the question, gathering material, forming a view, drafting, and polishing. A language model is genuinely useful at the first and last of those, and quietly corrosive in the middle.

Asking it to re-explain a confusing paragraph three different ways is a good use — it is a patient tutor that never sighs. Asking it to check spelling and flag a wandering sentence at the end is a good use. Asking it to form the argument, or to decide what the essay thinks, is where the learning leaves the building. The student ends up as a courier between two machines: one that generates and one that grades.

Three uses that help

Explain it to me again, simpler. A model will rephrase an idea at any reading level, indefinitely, without making a child feel slow. This is the single best thing it does for a struggling learner.

Quiz me. Ask it to generate ten questions on the chapter and mark the answers. Retrieval practice is one of the most reliably effective study techniques there is, and it is tedious to set up by hand.

Find the hole in my argument. Paste the finished draft and ask what a skeptical reader would object to. The child still owns the argument; the machine plays the opponent.

Three uses that hollow the work out

Write it for me. Obvious, and more common than parents think.

Summarize the book so I do not have to read it. A summary of a novel is not the novel; the thing being assessed is usually the encounter, not the plot.

Give me sources. This is the dangerous one, because it looks responsible. Models routinely invent citations that are formatted perfectly and do not exist. A child who has been taught to trust that output will hand in a bibliography of ghosts.

The ten-second test

Ask them to explain the work back to you with the screen closed. Not the whole thing — one paragraph, in their own words, and then one follow-up question you make up on the spot. If the explanation is fluent, the learning happened whatever tools were involved. If it collapses on the second question, the tool did the work and your child was the delivery driver.

This test is better than any detector, because it measures the thing you actually care about. AI-detection software has a well-documented false-positive problem and is particularly unreliable on writing by students with English as an additional language. Your own two questions are more accurate than the software.

Find out what the school's rule actually is

Most schools now have a written position, and most families have not read it. The rules vary enormously — some subjects allow AI for planning but not drafting, some require a declaration line at the end of the assignment, some forbid it outright. A child who follows a reasonable household rule that contradicts the school's rule is still in trouble. Ask for the policy in writing, read it once with your child, and make your household rule a subset of it.

A short script for the conversation

It does not need to be a summit. Something like: "I am not going to stop you using it, and I do not want you hiding it. Here is the deal — you can use it to understand things and to check things. You write the actual work. And if you use it, you tell me which bit." Teenagers respond better to a rule with a reason than to a prohibition they can route around in thirty seconds.

The longer game

The children who will do well over the next decade are not the ones who avoided these tools, and not the ones who outsourced everything to them. They are the ones who can look at a confident paragraph of generated text and say, precisely, what is wrong with it. That is a teachable skill, and it is teachable now, at thirteen, in about an hour a week.

That is the whole reason ZEROTH exists. World 1 is free, takes an evening, and needs no email address — it is the fastest way to see whether this kind of teaching suits your household.