Search for an AI homework helper and most of what comes back is a list of apps. A list cannot tell you the thing that matters, because two apps can look identical and do opposite jobs. One writes the answer on the page. The other asks the learner what they have tried. This guide is about telling them apart, and about choosing by the age of the learner.
Four jobs hiding under one name
When people say "AI homework helper" they usually mean a tool that responds to a question about schoolwork. What the tool does with the question is the difference between help and a shortcut.
Four kinds of homework help
1Answer-giver
Returns the finished answer or the full worked solution.
Ask: Who did the thinking?
2Explainer
Explains an idea in plain words, then stops.
Ask: Could the learner say it back?
3Coach
Holds the answer back and asks a question or gives the next hint.
Ask: Who wrote the last step?
4Checker
Looks at work the learner has already finished and points at errors.
Ask: Was the attempt made first?
The first kind, the answer-giver, is the one most people picture. The other three leave the learner with something to do. Explaining helps as long as the learner then closes the chat and tries the problem. Coaching is the strongest, because the learner keeps doing the work and gets support at the moment they are stuck. Checking is useful after a real attempt and wasteful before one.
Many tools do more than one of these. A study tool built around your own notes, such as Gemini Notebook, answers questions from the sources you give it. Quizlet generates flashcards, summaries and practice from a student's own notes. Gizmo turns notes and PDFs into quizzes and flashcards inside a spaced-repetition system. SchoolAI describes its assistant as one that supports thinking without doing the work, and gives teachers a dashboard with real-time insight. Those are descriptions of what the tools say they do, from the tools' own pages as summarised in our Learning Tools Index. Whether a given tool coaches or answers also depends on how the learner uses it, which is why the test further down matters more than any label.
Why the kind of help matters: one trial
The strongest evidence comes from a school trial. In a randomised field study of nearly a thousand high-school maths students, published in PNAS, students practised with either a plain GPT-4 interface or a version built to give hints. During practice, grades were 48% higher with the plain version and 127% higher with the hint-giving one. When access was removed for the exam, students who had practised with the plain version scored 17% below students who never had AI. The abstract says the safeguards in the hint-giving version largely mitigated the harm (Bastani et al., PNAS, 2025). We go through that study in detail in Does AI homework help hurt learning?.
The lesson for homework is narrow and useful. A tool that finishes the work makes the work look good, and takes the learning with it. A tool that holds the answer back largely avoided leaving learners behind in that trial.
A second line of evidence points the same direction. In a randomised trial called Tutor CoPilot, an AI tool wrote suggestions for human tutors rather than talking to students directly. Students were 4 percentage points more likely to master topics, and students of the lower-rated tutors gained 9 points (Wang et al., arXiv, 2024). That is a preprint, and it is about tutors rather than homework, so treat it as a pointer, not a promise. It fits the same pattern: the gain came when the tool improved how a person helped, not when it replaced the person.
Help that supports the tutor: more students mastered the topic
Percentage-point gains in topic mastery against tutors without the tool. The AI wrote suggestions for the tutor; the tutor still taught.
Source: Wang et al., “Tutor CoPilot” (arXiv:2410.03017), randomised trial, 900 tutors and 1,800 K-12 students
Does it fit the age?
Learners are not one group. What is sensible for a ten-year-old is not sensible for a seventeen-year-old, and what a tool's own terms allow may be different again.
In the UK, Ofcom research reported in the trade press found that 56% of children aged 8 to 17 use AI, rising to 66% of 16 to 17 year olds (Advanced Television, May 2026). The figures are about AI use in general, not schoolwork. Even so, most older teenagers already use AI in some form, so expect it to come up in homework.
Many children in the UK already use AI
Each bar is a share of its own age group.
Source: Ofcom children’s online experiences research (May 2026), as reported by Advanced Television
Age limits are set by the tools, not by us, and they change. Google's help page for Gemini says a parent can give a child under 13, or the applicable age in their country, access to Gemini Apps, and that the parent stays in control of that access (Google, Gemini Apps help). The page also says the service is not available for supervised accounts in the European Economic Area, Switzerland and the United Kingdom, so UK families cannot use that route. For any tool, read the terms yourself and use the school's approved tools if it has a list. England's Department for Education says settings should comply with the age restrictions set by AI tools, and that pupils "should only be using generative AI in education settings with appropriate safeguards in place, such as close supervision and the use of tools with safety and filtering and monitoring features" (DfE, Generative AI in education).
That points to a plain rule by age.
House rules by age: a starting point
1Primary years
About 5 to 10
An adult sits with the child. Use AI help only through an account the tool's own terms allow for that age, never an adult's account.
Coach or explainer only; no open chat.
2Lower secondary
About 11 to 14
Attempt first, on paper. Then one hint at a time. The learner writes the last step.
Coach, explainer, checker. Check each tool's own terms: some allow under-13s only through a school-managed account or with consent, and some not at all. Never use an adult's account.
3Upper secondary
About 15 to 18
Own the method. Use AI to test understanding and to check finished work, never to produce the final marked work unless the task allows it.
Coach, explainer, checker; drafting only where the task allows it.
Those are our suggested starting rules. They are not a regulation. Your child's school, and the exam board for older students, may set stricter ones, and those win.
One stuck question tells you what a tool is
You do not need a trial of your own. Take a problem that really is hard for the learner, sit beside them, and watch what the tool does when they say they are stuck.
- If it states the finished answer, it is an answer-giver. It will finish worksheets and leave the learner no better.
- If it asks a question back, or offers the next hint and waits, it is acting as a coach.
- If it explains and then asks the learner to try, it is an explainer working as intended.
- If the learner can push it into giving the answer anyway, the guardrail is weak, and you should know that before you hand it over.
Run the same test on a second tool and compare. Where there is a difference, you have your answer.
Four house rules for homework with AI
- Try first, on paper. The learner makes a real attempt before any tool is opened. A blank page followed by a question is a request for the answer.
- One hint at a time. Ask for the next step, not the whole solution. If a hint is not enough, ask for another.
- The learner writes the last step. The final line, in their own words, is theirs.
- Say what you used. Where a teacher allows AI, a short note on what the tool did keeps everyone honest. Where a teacher does not, do not use it.
For older students, the last rule is the one that matters for marks. Whether a use is acceptable is the school's call, and we cover that in Is using AI for homework cheating?.
What to watch for at home
Rising homework marks are not proof that a tool works. In the trial above, the harmful version produced the larger short-term gain. A better check is a few unaided questions on last week's topic, a few days later, with nothing open. If the learner can do them, the tool is helping. If they cannot, it has been doing their thinking.
Also notice what the tool does on a wrong turn. A good coach points at where the work went wrong and waits. A weak one silently fixes it and moves on, which teaches nothing.
Frequently asked questions
Which AI homework helper is best? There is no single best, because the right one depends on the learner's age, the exam system and what they need help with. Choose by what the tool does on a stuck question, then check its age terms and whether the school has approved it. Our Fit-Finder narrows tools by learner and system.
Is an AI homework helper the same as ChatGPT? Not necessarily. A general chatbot is built to answer. Some homework-specific tools are built to hold answers back or to work only from the learner's own notes. Both kinds exist, and the same test tells them apart.
Can AI do my child's homework for them? Yes, which is exactly the risk. A tool that returns finished answers can produce correct homework with no learning behind it. The house rules above are there so the learner stays the one doing the work.
How young is too young? Each tool sets its own minimum age and says who may open an account, and some give under-13s no route at all. Use only an account the tool's own terms allow for that age, never an adult's account. Keep an adult beside younger children, and prefer tools a school has set up.
Should I ban it? In our view, a ban is hard to keep and pushes use out of sight. A clear rule about how it is used, and a tool that coaches instead of answers, is easier to stick to.
Sources
- Bastani et al., "Generative AI Without Guardrails Can Harm Learning: Evidence from High School Mathematics," PNAS 122(26), 2025 (abstract via RePEc): https://ideas.repec.org/a/nas/journl/v122y2025pe2422633122.html
- Wang et al., "Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise," arXiv:2410.03017: https://arxiv.org/abs/2410.03017
- Advanced Television, "Ofcom: harmful social media content still an issue" (22 May 2026), reporting Ofcom's children's online experiences research: https://www.advanced-television.com/2026/05/22/ofcom-harmful-social-media-content-still-an-issue/
- Google, "Gemini Apps for children under 13," Gemini Apps Help: https://support.google.com/gemini/answer/16109150?hl=en-GB
- UK Department for Education, "Generative artificial intelligence (AI) in education": https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education