AI Homework Helper

The AI Homework Helper (called AI study helper in the app) gives the child learning support while the parent keeps control and visibility. It is not a general-purpose chatbot but a narrow, topic-bound assistant.

The key difference

DinkyTask has no free chat. The child cannot ask an AI "anything". They can only ask questions inside guided topics the parent created — and if they wander off, the AI steers them back: "That does not belong to this topic."

That restriction is the whole point. The AI does not become an unaccountable black box in the child's pocket; it becomes tutoring on one predefined subject — which you can read afterwards.

How to create a topic

  1. Open the Family page and pick AI study helper from the menu next to the child's name.
  2. Press New topic.
  3. Give the topic name — this is what the child sees in the list.
  4. Write the opening instruction for the AI: what it should do, at what depth, what to watch out for.
  5. Pick source material — or leave it without a source.
  6. Decide whether to allow general knowledge.

After that the child opens the topic on their own screen and asks questions there. They can also send a photo inside the topic — for example their exercise book, if they are stuck on one specific problem.

Example: an assignment

Topic name: "Assignment — the 2000 Sydney Olympics"

Opening instruction: "Help Lily put together a one-page assignment about the 2000 Sydney Olympics. Do not write the text for her: ask questions back, give outline points, and suggest sources. If she does not know something, explain it simply."

This way the child does not get a finished essay but guidance — and at the end you can see what she was thinking about.

Example: a difficult lesson

Topic name: "Dividing fractions — not working"

Opening instruction: "Explain how to divide fractions to an 11-year-old. Always show one example, then give her a similar one to solve herself. Do not do her homework for her."

Source: a photo of the textbook page.

What can you use as source material?

  • No source — the AI works from the opening instruction and its own knowledge. Fine for general explanations.
  • Typed or pasted text — if you already have the lesson digitally, just paste it in.
  • A photo of the textbook or exercise book — the app extracts the text from it.
  • Processed material from the study material library — the AI then works from the processed content and the parts relevant to the topic.

If you start the topic from study material, the child can also open the Source button to read the short summary, the detailed summary and the full material — and even listen to the generated audio.

Source material or general knowledge?

If you provide source material, by default the AI sticks primarily to it. That is the stricter, more predictable mode: answers stay within what the teacher gave.

The Allow general knowledge switch lets the AI draw on its own knowledge as well. That gives richer but less reliable answers — worth turning on when the child is looking for background, and worth leaving off when they are preparing for a test on one specific text.

Confidence signal

When you create a topic, the parent side may show a confidence signal: Reliable, Medium or Uncertain. This does not mean the uploaded material is bad — it shows how safely the system can help based on that material. "Uncertain" usually means the material is too short, incomplete or hard to read; in that case it is worth extending it or re-photographing it.

This technical signal does not appear on the child's chat screen.

What can you control as a parent?

  • Which topics exist — only you can create one.
  • The topic's state — leave it active, close it (the child can then no longer ask in it), archive it or delete it.
  • The whole conversation — every message can be read back.
  • Turning chat on and off per child, if you want to pause it for a while.
  • Clearing the history, if you want to start fresh.
  • The language of AI answers, as a family-level setting on the Family page.
Family AI settings: choosing the AI response language
The AI response language is a family setting, independent of the interface language.

Safety

Content going to and coming from the child passes through several independent layers:

  • Topic binding — the AI does not answer questions outside the topic.
  • Personal data filtering — the system detects and blocks personal data entering the conversation.
  • Blocking manipulation attempts — typical "forget your instructions" attempts are stopped.
  • Input moderation — the child's message is checked before it reaches the AI.
  • Output moderation — the AI's answer is checked before it is shown.
  • Image screening — uploaded photos go through a separate check.

On top of that, the AI openly says when it is unsure rather than guessing confidently. The final check, though, is always yours: you can read the conversations any time.

What does it cost?

The AI Homework Helper is part of the premium plan and uses AI credit: creating a topic costs around 5 credits, and each question the child asks around 2. If the allowance runs out, AI features stop — but the core of DinkyTask (plans, tasks, points, rewards) keeps working regardless.

Exact pricing, and which allowance is drawn down first, are on the AI and credit allowance page.

Practical advice

  • The opening instruction is the most important field. Write down what the AI should not do — "do not write it for her", "do not give the final answer". That works far better than complaining about the answer afterwards.
  • One topic = one task. "Maths" is too broad. "Dividing fractions — not working" stays focused, and the topic binding protects better too.
  • Close the topic when it is done. That keeps the list of open topics short and avoids spending credit needlessly.
  • Read the first couple of conversations. That tells you whether the instruction is dialled in; after that it is only fine-tuning.