Get help with homework without cheating
Where the line really falls between homework help and cheating, two reasons to stay on the right side of it, and why AI detectors settle nothing.
The difficult question is where helpful explanation ends and doing the homework for you begins.
The hard part is the middle. A model explains a concept you missed and you learn something. Or it rewrites your paragraph and you hand it in. The line sits somewhere between those two, and nobody draws it for you.
So here is the line, drawn as clearly as we can manage. Two reasons follow, and neither is a lecture.
The side that is plainly fine
These are all learning, so use them without guilt.
Ask for an explanation of a concept that went past you in class. Then ask again, differently, until it lands.
Ask for a worked example of a different problem of the same type, and do yours afterwards.
Ask to be quizzed. One question at a time, and answer before you scroll.
Ask why your answer is wrong, or better still, where the first line of it goes wrong.
Ask for feedback on structure: where the argument sags, which paragraph is doing nothing.
Notice the shape. In every one of those you still do the thing being assessed, and what you hand in stays yours.
The other end, said once
Paste the assignment, submit what comes back, and that is the whole of it. You do not need us to explain why.
Three cases in the middle
A model rewrites your paragraph. It depends on what was being assessed. If the task was to write, the words were the point and a rewrite takes the point away. If the task was to solve something and then write it up, tightening the write-up sits closer to a spellcheck. Ask what is unclear instead of asking for a better version, then fix it yourself. You keep your sentence and you learn the fault.
A model does one step of a multi-step problem. The danger is that the step you skipped is the step you cannot do, which is also the step an exam will ask for. Ask for the method on a different set of numbers, then come back and do yours.
A model finds your sources. Fine as a starting list and nothing more. You still open each one, read enough to know it says what it was claimed to say, and cite it yourself. Models invent references that look perfect down to the page number. A source you have not opened is a source you cannot defend.
The answer you cannot check
Here is the first reason, and it has nothing to do with morality.
A handed-over answer is often wrong, and rarely in an obvious way. It goes off three lines into a derivation, fluently, in a sentence better written than yours.
So ask who catches that, because it will not be you. You skipped the work, so you never built the understanding that would let you spot it. Your name is still the one on the page. That is the worst seat in the house: you carry the risk of an error you cannot detect.
That failure mode is common, built in, and no setting turns it off. We wrote up the mechanism separately in AI hallucinations explained.
Later, there is a room with no chatbot in it
The second reason is arithmetic. Assessment eventually happens where the model is unavailable: a closed exam, a viva, a whiteboard, someone asking you to talk through what you wrote.
Every piece of homework you handed over is a gap that surfaces there. It keeps quiet in October and turns up in a room in May with a clock running.
Homework is cheap practice for that room. Handing it over converts that into an expensive surprise.
About AI detectors
Two things are true at once, and students usually hear only one of them.
The tools are unreliable. OpenAI released a classifier for AI-written text in January 2023. The announcement itself said: “Our classifier is not fully reliable.” On its own challenge set it “correctly identifies 26% of AI-written text (true positives)” while “incorrectly labeling human-written text as AI-written 9% of the time (false positives).” OpenAI withdrew the tool on 20 July 2023, citing its low rate of accuracy. Its help centre still answers “Do AI detectors work?” with “In short, not in our experience.” It adds that its own attempt “labeled human-written text like Shakespeare and the Declaration of Independence as AI-generated.”
The errors land unevenly too. Stanford HAI reported on a study of seven detectors. They classified 61.22% of TOEFL essays by non-native English students as AI-generated, while scoring “near-perfect” on essays by US-born eighth-graders. If English is your second language, the tool is weighted against you before you write a word.
We are not printing a 2026 accuracy figure. We could not source one worth printing.
None of that is reassurance
A detector’s verdict is unreliable evidence in both directions. It can flag an essay you wrote alone at midnight, and a clean score proves nothing about you either. Detectors are also not the only way this surfaces. A tutor who has read your last four essays notices the fifth. An oral follow-up settles the question in two minutes. The score is a poor thing to optimise, and the two reasons above are why you do the work.
Your school wrote the rule that actually binds you
Everything on this page is our reasoning, and your institution’s policy outranks it.
Policies differ per institution, and often per assignment inside one. The same university can invite AI for brainstorming in one module and ban it in the next. Some ask you to declare what you used.
So find yours, reading the wording on the assignment brief rather than the general line on a website. If it is silent or ambiguous, ask whoever is marking it. Asking is safer than guessing. It is a question you can ask without looking like you are planning something.
Working this way in MultiChats
Send the question as a photo rather than retyping it. Many of our models read images, so your own working goes in with it.
Ask it to check your working rather than replace it. “Find the first line that goes wrong, then stop” beats “solve this” every time.
Some models show their working as they go, which helps when the method is what you are after.
Keep the thread. If anyone asks how you got there, a conversation full of your own attempts is the answer.
The free plan runs 50 messages per 24 hours across 22 of our 59 models, from 17 providers. That covers a normal week of homework. Free accounts also get 2 web searches per 30 days on a fixed window, spent automatically when the app judges a question needs the web. On the web app you can turn Intent Detection off in Settings and decide yourself.
We do not train on your conversations. What the provider behind a given model does with them varies by model. Where we know of a difference, that model carries a notice on its info card in the picker.
Start a chat and ask it to explain the thing you did not follow in class. Then do the problem yourself.