To get a useful course from an AI, describe the outcome you want rather than the topic, say what you already know, and say how deep you want to go. Ask for the structure first (units, lessons, practice), check it against your goal, and only then start learning. A general chatbot can draft that plan; a course generator also keeps the structure, your place in it and the practice in one place.
The rest of this page shows how to write the request, how to check what comes back, and the mistakes that make a generated course feel generic.
Why a topic is not enough
"Teach me Python" gives the AI almost nothing to work with. It has to guess your level, your reason and your finish line, so it falls back on the average beginner course: variables, loops, a calculator project. That course exists a thousand times already.
A goal changes the content. Someone who wants Python to clean spreadsheets needs files, tables and scheduling. Someone who wants Python for coding interviews needs data structures and timed problems. The topic is the same; the courses barely overlap.
Step 1: write the goal as an outcome
A good request has four parts:
- What you want to learn, in a few words.
- What you want to be able to do when you finish, as something you could check.
- Why, or the situation you are in (your job, an exam, a project).
- Your limits: time per day, a deadline, tools you must use.
Weak: "Learn Python."
Strong: "I work in marketing. I want to use Python to turn our weekly campaign exports into one report automatically. I know Excel formulas but have never written code. I have about 20 minutes a day."
Step 2: say what you already know
This is the part people skip, and it decides where the course starts. "New to this", "I know the basics" and "I'm already familiar, show me what I'm missing" lead to very different first lessons.
Be specific if you can. "I can write a SQL SELECT with a WHERE clause but joins confuse me" is far more useful than "intermediate".
Step 3: choose the size before the content
Decide how deep you want to go:
- A quick overview: the main ideas and how they connect.
- A solid understanding: the essentials plus practice using them.
- A deep dive: details, tricky cases and harder applications.
Some goals are too big for one course. "Become a data analyst" is several courses: spreadsheets, SQL, statistics, a visualisation tool. Treat a goal like that as a roadmap and build one course at a time, starting with the one you need first.
Step 4: review the outline before you start
Read the list of units and lessons and ask three questions:
- Does every unit move you toward the outcome you wrote? Cut the ones that don't.
- Is the same idea taught twice under different names? Generated outlines often repeat themselves.
- Does the first lesson start at your level, or does it re-teach what you said you know?
Fix the outline now. Changing a plan takes a minute; finding out in lesson 6 that half the course is filler costs you a week.
Step 5: make every lesson end in practice
Reading an explanation feels like learning. Answering a question from memory is what makes it stick. A large review of study techniques rated practice testing and spacing practice out over time as the two most useful methods, and rereading and highlighting among the least useful (Dunlosky et al., 2013, summarised by the Association for Psychological Science).
So ask for a short question or exercise at the end of each lesson, and answer it before moving on, even when you think you already know it.
Step 6: start lesson 1 today
A course that waits for the right weekend tends to stay unopened. Do the first lesson the same day you create the course, while the goal is still fresh, even if it is only ten minutes.
Example: one goal, start to finish
Here is the marketing goal from Step 1 turned into a plan. This is an illustration of the shape to aim for, not output from any particular tool.
- Unit 1. Python just for your task: running a script, reading a CSV file, the three or four data types you will actually meet.
- Unit 2. Working with tables: loading exports, filtering rows, joining two files, handling missing values.
- Unit 3. Building the report: totals and comparisons, writing to a spreadsheet, a simple chart.
- Unit 4. Making it run every week: one script for the whole job, checking for broken input, scheduling.
Notice what is missing: no calculator project, no object-oriented theory, no web development. The goal ruled them out.
Doing it with a general chatbot
You can do all of this in ChatGPT or any other assistant. A prompt that works:
I want to [outcome]. Context: [your situation]. I already know [what you know].
I can spend [time] per day. Build a course outline: units, then lessons,
each lesson ending with one practice question. Don't write the lessons yet.
Review the outline, ask for changes, then ask for lessons one at a time. The catch is that a chat is not a course. You have to keep the plan somewhere, remember where you stopped, and remind the assistant of both whenever you start a new conversation.
Doing it with Quri
Quri is built for this exact workflow. You name a subject and write why it matters to you. Then you pick how familiar you already are (new, the basics, already familiar) and how deep to go (quick overview, solid understanding, deep dive).
Quri then sizes the request as a mini-course, a full course or a roadmap of several courses. If your goal is broader than one course, it shows you the roadmap and lets you choose which course to create first; nothing is created until you confirm. Creating the course takes about one to two minutes, and you need a free account to do it (Quri is free while in beta).
The course is split into units, lessons and steps. Each step ends with a short challenge you pass to complete it, and lessons and units have quizzes. While you study, you can select any text and ask the tutor about it; the tutor sees the course outline and the lesson you are on. If the course turns out not to fit, you can adjust it and say what should change.
When a generated course is the wrong choice
- You need a recognised credential. An accredited degree or a certificate an employer checks has to come from the institution that issues it.
- The skill needs supervised practice. Anything where mistakes are dangerous needs a qualified person watching you.
- An exam board sets the syllabus. You can still generate a course around it, but check the outline line by line against the official syllabus.
Common mistakes
- Asking for a topic instead of an outcome. You get the generic course.
- Leaving out what you know. The course re-teaches the basics and you lose interest by lesson 3.
- Trying to fit a career into one course. Split it into a roadmap.
- Skipping the outline review. Repeated and off-goal units are easy to cut before you start and annoying to sit through later.
- Reading without answering. If a lesson has no question at the end, write one for yourself.
FAQ
Can ChatGPT create a course for me?
Yes. Give it your outcome, your level and your time, and ask for an outline first. You will need to track the plan and your progress yourself, because a conversation does not keep them in order for you.
How accurate is an AI-generated course?
Usually good on well-established topics, less reliable on niche, very recent or regulated ones. Check any fact you plan to act on against a primary source, such as official documentation or the exam board's syllabus.
How long should a generated course be?
As long as the outcome needs and no longer. If the outline is much longer than your available time, reduce the depth or split the goal into smaller courses.
Is there a free AI course generator?
Several. Quri is free while it is in beta; you sign up with an email and a password. General chatbots also have free tiers, with the trade-offs above.
Can I change the course after it's created?
In a chat, ask for a revised outline. In Quri, use "Adjust course" inside the course and describe what should change.