How we use AI
What AI does on Informative Academy, what it will never do, and the reasoning behind both.
AI helps our learners think about their own work. It is not someone they talk to.
That one sentence drives almost every decision on this page. A lot of AI built for young people is designed to be engaging, conversational, and always available. We think that is the wrong shape for a learning platform used by children, so we have deliberately built something narrower: AI that responds to a piece of work a learner has already done, then gets out of the way.
Everything below is a plain-language summary of a policy we hold ourselves to internally. It sets out how every AI feature is classified, what protections it must have before it can launch, and who has to approve it.
Where AI is used
Feedback on work you have already done
After you write a reflection, finish an assessment, or build out a business idea, AI can offer comments and a question to think about. It responds to your work. It does not do the work.
Explaining what happened
After a game or a simulation, AI can explain why a decision played out the way it did, using what actually happened in the activity.
Behind the scenes
Our team uses AI to help draft and review lesson material. A person reviews and approves everything before a learner sees it.
Wherever we offer an AI feature, there is also a way to get a useful result without it. AI is never the only route to something worth having.
What we will not build
An AI friend or companion
No character to chat with, no simulated personality, no AI that remembers you and asks you to come back and talk. Children can form real attachments to conversational AI, and we are not willing to design for that.
An AI counsellor or confidant
AI will never be the thing a young person turns to about their safety, their mental health, or something worrying them. Those conversations belong with a trusted adult, and we route them there.
A general chatbot
There is no "Ask AI" box on Informative Academy. Every AI feature is attached to a specific piece of work, with a beginning and an end.
AI that decides things about a learner
No model decides a grade, an outcome, what a learner is like, or what they can access. AI output is a suggestion a person can ignore.
AI advice about money, health, or the law
Our platform teaches financial thinking with simulated money. AI will not tell a young person what to do with real money, or give medical or legal advice.
We write these down as prohibitions rather than preferences on purpose. A preference bends when a feature looks exciting. A prohibition has to be formally changed, by a named person, in a document with a version number.
What happens to what a learner writes
When a learner uses an AI feature, the work it is responding to is sent to our AI provider so it can generate a reply. That means a reflection, an assessment answer, or a business idea does leave our platform for that request.
We never use learner data to train AI models. Not ours, not anyone else's. Our provider does not train on content sent through their API, and our own policy prohibits it outright.
We want to be precise about the limit of that promise, because not training on something is not the same as not holding it. Our AI provider keeps API requests for up to 30 days so it can monitor for abuse, and longer only where the law requires it, unless stricter retention controls are approved for our account. We are also still completing the formal data-processing agreements with our AI providers, and we would rather say so here than let the wording imply the paperwork is already done.
There is no back-and-forth conversation for us to keep, because our AI features are one-shot by design. Some AI results are saved on purpose: the feedback attached to a ThinkType result, and the commentary in a completed Build report. Those are saved so a learner can return to their own work, and they carry the same rights as everything else — a learner or parent can ask us to export or delete them.
We keep AI content out of our internal logs. We record that a feature was used and how long it took, not what was written in it.
An adult connected to the account does not automatically get to read what a learner wrote to an AI feature. For anything new we build, sharing with a parent has to be something the learner positively turns on.
One current exception, which we are fixing: journal entries flowing into a parent report still follow an older setting where they are included unless a learner marks them private. We have decided to reverse that so nothing is shared unless a learner chooses to share it, and that change is queued.
Age changes what is allowed
Ages 11-12
Short, single-purpose AI only, on a specific task, explained before it happens. No conversation, nothing kept.
Ages 13-15
The same, and anything more open would need its own separate safety review before we would consider building it.
Ages 16-20
Still tied to a task. Being older does not unlock a general-purpose AI, because that is not what we are building.
A Premium subscription can unlock more feedback. It never removes a safety limit, widens what a learner can type in, or turns a younger learner's experience into an older one's. Paying more buys more value, never fewer protections.
Learners under 16 need a parent or guardian to confirm their account before they can use the platform, and that check happens on our servers rather than in the browser, so it cannot be clicked past. The finer-grained age check that applies these bands to each AI feature individually is being built now, and it has to be in place before we add any new AI feature.
A person is always the answer for anything that matters
If something a learner writes suggests they are worried, unsafe, or dealing with something difficult, the answer is never a longer AI reply. Our policy requires every AI feature to recognise that moment, stop, and point towards a trusted adult, and no new AI feature can launch without it. Our forums, messages, and support channels are staffed by people rather than models, and we do not intend to automate them.
This is also why we will not let AI ask a young person to keep something from a parent, teacher, or carer. No AI feature we build will ever suggest secrecy.
How we hold ourselves to this
Every AI feature is classified before it is built, and its classification decides which protections are mandatory: who it can serve, what it may receive, what it may say, how often it can be used, and how quickly it can be switched off.
A feature cannot ship until each of those is in place and evidenced. An idea being written down is never the same as an idea being approved.
We review the policy every quarter and after any material change, and we record where we are not yet meeting our own standard rather than quietly rounding up.
Being specific about that, as of August 2026: the per-feature age check described above is being built, the parent-report sharing default is being reversed, and our formal data-processing agreements with our AI providers are not yet signed. Our existing AI features are older than this policy and are being brought up to it rather than being treated as already compliant.
Questions about this
If you are a parent, educator, or partner with a question about how AI works here — or you think something on this page does not match your experience — please get in touch at privacy@informative.sg.