Every owner has two fears about AI, and neither is silly.
The first is that it replaces the tutors. The second is that it makes things up. Both are reasonable because both have happened somewhere, and because the loudest pitches in this space promise the first and ignore the second.
This is the honest version. What a general chatbot is good for, what it can't do for a centre, what "makes things up" means in plain words, and the two rules that make any of it safe when it's under your logo. Where the platform is mentioned, it's to say what it won't do as much as what it will.
- A general chatbot is a tool for the owner: one PDF, tonight, alone. Use it for that.
- It can't run students, classes, marking by topic or a parent report, and it doesn't keep your material private by default.
- Hallucination is a fluent, confident, wrong answer. Under your logo, it's your answer.
- Two rules make it safe: every item cites its page, and you approve before a student sees anything.
- Ask any vendor four questions. Train on my material? Can another centre see it? What happens to it if I stop paying? Who approves?
The two fears, and why both are reasonable
"It replaces my tutors." Parents pay for the person in the room. That's the product, and it's the reason your centre exists rather than a subscription to a study app. Anything that suggests the person in the room is optional is a threat to the business, and the owner who hears "AI tutor" and reaches for the door is reading it correctly.
"It makes things up." You've probably seen this yourself. You put a worksheet into a chatbot, asked it a question, and it answered fluently with a method you don't teach, or a number that wasn't on the page. It didn't hesitate. It never does. Now imagine that answer on a phone at 9pm with your logo above it, and a parent asking on Tuesday why the centre told their child to do it that way.
Neither fear is about the technology. Both are about what it's allowed to do, and who's responsible when it does it. That's the frame for the rest of this guide.
Nothing appears in your platform that didn't come from a page you uploaded, and nothing reaches a student until you've said yes.
What a general chatbot is good for
For you, tonight, with one PDF, use it. You'll learn something.
Reading a long document and asking questions of it. Drafting an outline for a new worksheet you'll write yourself. Rewording a parent email that's come out harsher than you meant. Summarising a syllabus change you haven't had time to read. A sounding board at 10pm when there's nobody else in the building.
Those are real uses and they cost almost nothing. The owner who uses a chatbot this way for a month usually comes out of it with a sharper sense of what the tool is: a very fast, very confident assistant with no memory of your centre, no accounts for your students, and no sense of what's true.
The comparison with ChatGPT and NotebookLM goes through this line by line. The short version is that a chatbot is a tool for the owner, and a platform is a system for the centre.
What it can't do for a centre
The gap opens the moment there's more than one person involved.
| What a centre needs | A general chatbot | What it takes |
|---|---|---|
| 120 students on their own logins, in their classes | One chat window per person, no accounts for your centre | Every student in the class they're actually in |
| A set assigned to a class with a due date | No | Class, set, due date, three clicks |
| Marking by topic, every answer tracked | No | Every answer marked against your key, fed to a class heatmap |
| A parent report with your logo, from your address | No | Every second Friday, automatically, edited by you if you want |
| Your material kept private | Depends on the plan and the settings you chose | Your centre's space isolated from every other centre's, no reuse, no training, in the contract |
| Hints rather than answers | It will do the homework if asked | A hint, then a worked example, then the solution, with the tutor seeing how many were used |
None of those rows is about intelligence. They're about plumbing: accounts, classes, marking, reports, isolation. A chatbot can't be prompted into having them, and an owner who tries usually loses a term working out how to get the results back from 120 students on 120 separate chats.
Hallucination, in plain words
A hallucination is a fluent, confident answer that isn't in the sources or isn't true.
It happens because a general chatbot is built to answer everything. Ask it about a worksheet it has and it answers. Ask it about a worksheet it doesn't have and it still answers, in the same tone, because producing a plausible answer is what it does. It has no reliable way of telling you which of the two it just did.
In a tutoring context, the risk is specific. A made-up question that looks like yours. A method that isn't the one in your notes. An answer that contradicts your own key. Any of those, delivered to a student, is bad. Delivered with your logo on it, it's yours. When the parent asks why, you can't point to the worksheet it came from, because it didn't come from one.
That's the whole reason hallucination matters more in a centre than it does for an owner at a laptop. At the laptop, you're the check. On a student's phone, there is no check unless one was built in.
Rule one: grounding, or every item cites its page
Grounding is the constraint that the system may only produce something supported by a specific set of documents, and must say which document and which page.
On the platform it works like this. You upload your material. Every practice question, worked solution, explanation and flashcard that comes out of it carries a reference to the source document and page it was built from. Anything that can't be traced to a page is rejected before you ever see it. Not flagged. Rejected.
Maya makes 2/3 of a recipe that needs 3/4 cup of sugar. How much sugar does she use?
Multiply the fraction of the recipe by the amount: 2/3 × 3/4 = 6/12 = 1/2 cup.
That's what makes the sentence "it came from your worksheet" true rather than a slogan. When a question looks wrong, you click through to the page. Usually the worksheet had the error, and now you know. When a student gets an explanation at 9pm, it cites the document and page, and a parent or tutor can check it against the original in one tap.
If your material doesn't cover something, the platform says so and suggests bringing it to the next session. It never fills the gap from outside your pages. The glossary entry on grounding has the plain definition; the upload and convert page shows where the citation sits on each item.
Rule two: you approve before a student sees anything
Grounding stops the system inventing. Approval stops anything reaching a student that you haven't looked at.
Every converted document lands in a queue as a batch: questions, solutions, explanations, flashcards. Each item shows its source page beside it. Where your worksheet had an answer key, the item shows whether the platform's answer matches it, and a mismatch is flagged in red for you to resolve. Usually the worksheet is the one that's wrong.
You approve in bulk when a batch is clean. You edit the wording of anything you'd phrase differently, and your edit is what publishes. You bin what you don't want. Students see only what's been approved, and that's by design, not by a setting someone could flip. If you'd rather delegate, a senior tutor can hold approval rights for their own year level.
Most owners read the first few documents closely and then approve in bulk once they trust the conversion of their own material. The approve before publish page walks through the queue. It's the step in a walkthrough where the platform stops being an AI tool and becomes your material, organised.
Hints before answers, so it won't do the homework
The third fear, usually unspoken, is that the student will use it to cheat.
So the help a stuck student gets is staged. The first response is a hint about the method, drawn from your notes. The second is a worked example from your material with different numbers. The third is the full solution. Most students never need the third, and the tutor can see how many hints each student used on each question.
It won't write an essay. It won't do a whole set. It can't be talked into answering from the open internet, and it can't invent a method you don't teach, because it has nothing to draw from except your pages. The stuck at 9pm page shows the three stages on a phone.
What to ask any vendor
Whoever you're talking to, including us, these are the questions. The answers should be short.
| Question | The answer you want | Why it matters |
|---|---|---|
| Do you train on my material? | No, and it's in the contract, not the sales pitch | Your material is your edge. Training on it means it's now in someone else's product |
| Can another centre ever see it? | No. Your centre's space is isolated from every other centre's, and your material is never shared with or reused for anyone else | Look for that sentence in the contract, not the brochure |
| What happens to my material if I stop paying? | Access ends, their copies are deleted within a stated period and confirmed in writing, and you still hold the originals you uploaded copies of | The alternative is a supplier holding the only good copy |
| Who approves an item before a student sees it? | You, or a tutor you've named | If the answer is "nobody", the logo on the mistake is yours |
| Where do explanations come from? | Only my pages, with the page cited | Anything else is the open internet with your name on it |
| Where is it hosted? | A named country | Which privacy law applies, and how fast a request is answered |
| Do you contact my parents? | Never | The parent relationship is the business |
For this platform, the answers are no, no, any day, you, only your pages, Australia, and never. The security and ownership page has each one in writing, including the commitment we hold our own providers to. This is a description of what exists, not legal advice; anything about your privacy or copyright obligations is a question for your own adviser.
What's deliberately not built
Some of what makes AI safe in a centre is what's left out.
There are no video lessons. Your tutors are the video, and that's what parents pay for. The platform covers the days when the tutor isn't there, not the ninety minutes when they are.
There's no student-to-student messaging and there are no public profiles. Tutor-to-student messages are a setting for your centre, off by default, and you decide whether they're on. Your child-safety answer stays one sentence long.
There's no material of ours. The moment a platform ships a question it wrote, it owns that question's quality forever, and so does every centre that published it. Nothing appears in your platform that didn't come from a page you uploaded. The roadmap lists these under "not planned", with the reasons, because a buyer should be able to see what a supplier has decided against.
This week
Put one worksheet into a general chatbot tonight and ask it three questions about the page. Write down where it answered from outside the page, or answered confidently and wrong. That's your own evidence of what the fear is about.
Then take the vendor questions above to anyone selling you anything with "AI" in it. Seven answers, in writing, before a single page of your material goes anywhere.