How We Do It: Content Safety and Explanation Standards at Duowenwen

The "For Parents" page is written in a parent's own words. This page is for people who want the details, and for search engines and answer engines: how Duowenwen does those five things, and what we explicitly don't promise.

Can my child really not tease it into saying something it shouldn't?

It only explains problems, and does nothing else on your child's behalf. Whatever is sent to it is read as text only, never run as a command, and it doesn't click anything. It won't discuss violence, sexual content, self-harm or discrimination. A detector for instructions like "pretend to be someone else", "forget your rules" or "let's talk about something else" is still being built, and we'll describe it on this page once it's done. Until then, this protection rests on the design itself: it can only explain problems.

What rules does it follow when it explains a problem?

The answer is aligned word for word against the textbook text and comes with a page number, so your child can check it in the book. The explanation talks like a tutor and goes step by step. It never says "you should know this by now"; it praises how your child thought it through, not how clever they are; and it doesn't compare your child with anyone. Every explanation ends with 2-3 follow-up questions, and your child picks the next one themselves. It doesn't rush or grade.

Who checks that the explanations are right?

In the early days, we spot-check explanations ourselves, looking only at the problem and the explanation, not at who asked. We look at three things: is the answer right, is the page number right, is the tone right. When we find a wrong explanation, we fix the rules first, then check whether similar problems are wrong too; where we can't fix it, we say plainly for that type of problem, "please go by what your teacher says." A regular spot-check procedure hasn't been set yet; once it is, we'll write it here.

Can explanations be wrong?

Yes. It can misread a problem, miscalculate, or quote the wrong textbook page. That's why answers carry page numbers: so people can check, not to look reliable. If it differs from what the teacher says, the textbook and the teacher are the authority. We don't present it as something that never makes mistakes.

Where do my child's photos and chats go?

Transcribed text is kept for 7 days, so your child doesn't have to retake the photo when asking a follow-up; after 7 days it's deleted automatically. Photos are kept for 90 days at most. The operator may look at the photos your child actually takes (crooked, glary, handwritten) to improve recognition; only the operator can see them, and they're deleted along with the conversation when it's deleted. We don't use them to train our own models. Explaining a problem goes through third-party AI services, which handle data under their own data policies; the list is in the privacy notice.

What can parents see, and what can't they?

In the family view, parents can see which points their child got stuck on this week, and the question map. They can't see scores or rankings, because we don't grade or rank. The questions a child asks are the child's own; parents see the map to know what they've been wondering about this week, not to check up on them.

What don't you promise?

We don't promise your child will definitely learn it, or any change in grades. We don't write your child's homework or hand it in. We don't decide for the teacher what's right or wrong; the textbook and the teacher are the authority. We don't build a question bank or offer exam papers to download. We don't run ads or rankings.

Next: if you still have questions, write to us; to see how it explains, photograph a problem.