The Teacher Portal: Who Checks the Machine, and What They Earn for It

Any serious claim about AI-generated educational content rests on a single question: who reads it before a child does? It is easy to answer with the word teachers and leave it there. The harder and more useful version is how the queue is built, how many people must agree, what stops a bad actor, and what the reviewer gets for the twenty minutes.

The queue

A teacher who turns on reviewing in their settings gets a queue of quizzes awaiting approval. Their own submissions are excluded. Anything they have already decided on is excluded, so the list is always work that is actually theirs to do.

Each item shows the whole quiz, not a summary: every question, every option, every marked answer, every fill-in answer, every ordering and every grouping. For picture questions the picture is drawn, because approving a question about base-ten blocks without seeing the blocks is not approving anything.

How many approvals

There are two paths, and they differ because the situations differ.

A quiz written by a teacher publishes when two different teachers approve it. Two is a considered number. One is a single point of failure for content going to thousands of children. Three slows the queue to the point where nothing moves and the review becomes a formality people click through.

A quiz an administrator assigns to a named teacher publishes on that teacher's single approval. The second opinion already happened: an administrator looked at the batch and chose who should check this part of it. Asking for another signature after that is ceremony. Assigned quizzes sort to the top of that teacher's queue and are marked as theirs.

In both paths, one rejection is enough to stop a quiz. Agreement should be hard and objection should be easy, because the costs are not symmetric.

What it pays

• Approving a quiz: one credit, paid at the moment you decide, regardless of whether your approval is the one that publishes it. The work is reading the quiz, and you did that either way.
• Having your own quiz published: three credits, paid when it reaches publication, not when you submit it. Submitting fifty weak quizzes earns nothing at all.
• Student engagement with your quiz: a further credit for each ten questions answered by students on their first attempt at it. Replays by the same student never count twice.

The timing of the writing reward is the part worth dwelling on. Pay at submission and you have built a machine for generating volume. Pay at publication and the reward is attached to the only thing anyone wants, which is content good enough to pass two colleagues.

Turning credits into money

From fifty credits a teacher can request a redemption. The conversion rate is set per country with a fallback for countries not listed, because fifty credits should mean something comparable in Lahore and in Dubai. Requesting holds the credits immediately, so the same balance cannot be claimed twice, and an administrator processes the payout. A rejected request returns the credits in full.

This is a manual payout model and we would rather describe it plainly than dress it up as something automated. The amounts are modest. The thing being bought is attention from a qualified person, and it is better to pay something real for that than to ask for professional judgement as a favour.

The queue that does not fit this model

Some content never reaches peer review. A keyword filter runs over every submitted question, and anything it flags goes to an administrator-only queue instead. Not rejected: routed. A Class 9 biology question can look alarming in a word list and be entirely correct in its chapter. That decision belongs to an administrator rather than to whichever volunteer is next in the queue.

Why this is the part we talk about most

The generation is the impressive-looking half and the easy half. Anyone can call an API. What is hard, and what decides whether a platform is trustworthy, is building the layer where qualified people look at the output, having enough of them, making it worth their time, and making sure nothing routes around them.

So when we say experts check the content, this is what that sentence is standing on: a queue, two approvals or one named one, a rejection that outranks an approval, a filter that escalates rather than blocks, and credits that convert to money. The mechanisms that keep unreviewed content out of a student's hands are described separately.