The AI assistant a course creator actually needs

Somewhere in your first week of running an online academy you will hit a question shaped like this one: the course is published and visitors still cannot see it, or checkout will not complete, or a setup is one toggle short and you have no idea which toggle. So you open the documentation and find a correct explanation of something that is not quite your situation, or you write to support and spend the rest of the day waiting. The irony is that the problem itself takes a minute to fix once you know where to look.

This article is about the kind of help that ends that minute. Not a chatbot that writes you a competent paragraph about pricing, but an assistant that lives inside your dashboard, reads the state of your academy, and answers the question you actually asked. The difference between the two is not the quality of the prose. It is what each of them is able to know about you.

The blocker is not what you do not know, it is what you cannot find

Ask any academy owner about the last time they were stuck, and the answer is almost never "a missing feature". Far more often the step exists and they cannot locate it. The platform does exactly what they want; the distance between them and it is the name of a screen.

That distance costs more than it looks. Searching documentation is slow because you search in your words and the docs were written in someone else's, so you read three accurate pages and none of them is about your case. Waiting on support turns a small stall into a postponed day. And both routes charge a third cost that shows up in no ledger: you lose the working state you were in, and come back an hour later to a task you have gone cold on.

So the first thing worth having in an assistant that lives in the dashboard is not eloquence. It is the ability to say: open this section, then this screen, then set this. An answer that ends in the name of a screen ends the problem. An answer that ends in a general explanation returns you to the starting line in better English.

Four rungs an assistant can climb

Not all help is the same kind of help. It pays to sort what an AI assistant can offer into four rungs. Each rung is worth more than the one below it because it moves you a step closer to done, and each one requires something the previous one did not.

Rung	The question it answers	What it takes to reach it
Explaining how something works	How do I create a quiz?	Knowledge of the platform
Pointing at the path	Where exactly do I set this?	Platform knowledge plus the context of the page you are on
Diagnosing your own state	Why is my course not visible to visitors?	Reading your academy's state at the moment you ask
Doing the work	Build me a draft lesson on this topic	The ability to create inside your academy

A general chatbot stops at the first rung, and that is not a flaw in it so much as an honest limit: it holds nothing about your academy, so the best it can offer is the general case. The second rung requires being inside the platform and knowing where you currently are, so the answer is about your page rather than about the manual. The third and fourth are out of reach for anything without a real connection to your data.

Keep this ladder in mind the next time a product offers you an "AI assistant". The useful question is not whether it has AI in it. It is which rung it reaches, because the gap between the first and the third is the gap between pleasant conversation and a finished task.

Diagnosis: the rung a general chatbot cannot reach

This is where the whole practical difference sits. Ask an in-platform assistant why your course is not showing to visitors and it does not recite the possible reasons. It checks what is holding your course back and names the specific cause and the step that fixes it. The distance between a list of possibilities and one confirmed cause is the distance between half an hour of trial and error and a minute of doing.

The second diagnosis is about money directly. When a purchase does not go through, the assistant inspects your payment setup and points at what is missing. That case deserves particular attention because it fails silently: nobody tells you that a visitor tried to buy and could not. Buyers do not file complaints, they close the tab. A visible outage you find yourself, but an incomplete payment setup can run for a week while you quietly conclude that demand is soft.

Note the structural point underneath both examples. The answer is built on your academy's state at the moment you ask, not on what was true when the documentation was written. Documentation ages. Your live state cannot, because it is read fresh every time.

When an answer is not enough: agent mode, and how it is counted

The fourth rung is called agent mode, and it sits on the Premium plan for the academy owner alone. Here the assistant does not only explain; it creates: a lesson, a quiz, an assignment, an image, or a complete page with its blocks.

The route from request to output is four steps and no more:

1. You describe what you want in ordinary language, the way you would explain it to a colleague, with no special syntax to learn.
2. The agent builds the output as a draft inside your academy, in its proper place rather than in a side window.
3. You open the draft, read it, and change whatever you want.
4. You decide whether it publishes. The final call stays yours on every output without exception.

The accounting is deliberately simple, simple enough that you can estimate a project before you start it:

• One credit for a lesson, a quiz, an assignment or an image.
• Three credits for a complete page.
• You are charged only when the creation actually succeeds, and small edits to what was created cost nothing.
• Credits renew monthly and do not roll over, and you can top up with 50 credits for 49 SAR whenever you need.

That "charged only on success" rule matters more than it first appears, because it moves the risk of a failed attempt off you and onto the platform. You experiment without keeping a meter running in your head, ask for a second draft when the first one misses, and pay nothing for an attempt that produced no result.

And notice where this genuinely earns its keep. The first draft is the heaviest part of building a course: the blank page, the ordering of ideas, the first ten lines. When a draft already exists, your job changes from composing to editing, which is faster, more pleasant, and much closer to where your expertise actually lives. The same is true of writing assessments, where drafting ten questions from scratch is the task everyone postpones until after launch and then never does.

The limits that make an assistant trustworthy

An AI assistant is useful in proportion to how clearly its limits are drawn, because the worst thing to build a decision on is a tool that does not know where it stops. Three limits here are explicit.

First: no access to Google Calendar data at all. Your personal appointments are entirely outside its reach, and its work is confined to your academy's data inside the platform. This is an architectural boundary rather than a setting you switch on and off, and it is the difference between an assistant that knows what it needs to know and a tool that collects everything it can reach simply because it can.

Second: agent mode is limited to the academy owner. Your team keeps working within their usual permissions, nobody creates content in your name, and nobody spends your allowance without your knowledge. The right to create follows ownership, not membership.

Third: everything it produces is a draft. Nothing reaches your learners before it has passed you. That also means the agent cannot damage a published page or alter a working lesson, because its output begins life outside anything the public can see.

Where AI will let you down

Asking a general chatbot about a specific academy. Ask an outside tool why your course is not visible and you will get a confident, well organised answer built on nothing at all, because it has never seen your academy. Then you will work through steps that were never your problem. The rule is simple: a question about your platform gets asked inside your platform.

Treating agent mode as a publish button. The output is a draft, and that is a design decision rather than a shortcoming. Approving a draft you have not read means publishing under your own name something you have never read, which is the fastest way to lose the trust of a learner who paid you.

Upgrading in order to get chat. Chat is on every plan at no extra cost, so there is no sense in paying for something you already have. What the Premium plan buys is execution, not answers. If your question is "where do I set this", your current plan already serves you. If your question is "build me this", the upgrade is aimed at the right thing.

Hoarding credits. They renew monthly and do not carry over, so the allowance you did not spend this month does not enlarge next month's. Make agent work a monthly habit instead of saving it for the big project that deserves it.

The point of all of this is that the assistant worth a place in your day is the one that knows your academy, not the one that knows the language. The first saves you an hour of searching with a single sentence ending in the name of a screen. The second writes you a graceful paragraph about a problem that was never yours. Try the difference during the free trial before committing to anything.