Do We Still Need Courses When We Have AI?

By Vadym Leonov, web developer for course creators and membership businesses · 15-minute read

You spent months on your course. You recorded lessons after work, rewrote the tricky explanations three times and answered the same questions in the community until you could do it half asleep.

Then a student mentions, almost in passing: “I asked ChatGPT and it explained it more simply.”

It stings. Underneath the sting is a scarier question. If anyone can ask a chatbot anything, why would they keep paying you?

I build and look after learning platforms for English teachers and other course businesses, and some version of this question comes up every few weeks. My honest answer: AI has killed one kind of course. It isn’t the kind worth building.

Why information alone is a weak product now

The reference library has a new competitor

If your course mainly answers standard questions, AI is now your direct competitor. Explaining the present perfect, a VLOOKUP or the AIDA framework takes a chatbot a few seconds. Unlike a recorded lecture, it adapts: simpler words, a different example, a scenario from the student’s own job.

The evidence that this produces real learning is getting harder to wave away. In a 2025 randomised study with 194 Harvard physics students, a carefully designed AI tutor produced larger learning gains, in less time, than the active-learning classes it was compared with, and students reported feeling more engaged (Kestin et al., Scientific Reports). Read the fine print, though. This was a purpose-built tutor on specific lessons, not “ask ChatGPT about physics”. Still, the direction is clear.

Defending a premium price on access to explanations alone gets harder every year.

Routine correction is losing its premium

Students already paste essays into AI for grammar fixes, ask it why their code throws an error and get it to write extra exercises. For routine work, that feedback arrives faster than any teacher can manage.

It isn’t always right. A fluent correction can be wrong for the context, and code that runs can hide reasoning that won’t survive the next problem. The valuable part of your feedback was never the red pen anyway. It was spotting the misunderstanding behind the mistake and deciding what the student should practise next. If your offer is “I’ll correct your mistakes”, show students that deeper layer, because the surface layer is now free.

A box of videos is hard to justify

Recorded lessons are still useful. Students can learn on the train, and you stop explaining the same idea forty times. A focused course can still save someone hours of searching.

The product in trouble is the unsupported archive with a big promise on the sales page: 87 lessons, lifetime access and no help deciding what matters this week. AI makes that model look worse, because at least the chatbot talks back.

My view is that selling information on its own is finished as a premium business. People will keep buying recorded content, the way they still buy books. They just won’t pay coaching prices for it.

Five fixes that don’t fix anything

1. Recording more lessons

When sales slow down, adding a module feels productive. You get something to announce and the package looks bigger.

More volume usually means more homework, not better results. Someone who freezes in English-language meetings needs repeated conversation practice and specific feedback. Another grammar library is one more thing for them to put off.

Before you record anything, find what is actually stopping progress and build the smallest useful fix for that.

2. Competing on price

A discount can help a launch. As your main answer to AI, it’s a trap. You can’t undercut a free chatbot by making human support cheaper, and you’ll end up with students who need lots of help and no margin to pay for it.

A better pricing conversation spells out what the money buys: how much practice, how much review and how often students see you.

3. Pretending AI can’t teach

Your students have probably already found AI useful. Telling them it’s worthless makes you look out of touch.

Explain where it helps, where it goes wrong and how to use it well. That makes you the expert who helps students choose their tools, and a chatbot can’t take that role from you.

4. Bolting on a chatbot and calling it a transformation

A chatbot is a feature. Whether it helps depends on the job you give it. Which skill does it help students practise? Does it follow your method or a generic one? Can students push back on its feedback? What happens when it is confidently wrong?

That last question isn’t hypothetical. The US National Institute of Standards and Technology lists “confabulation”, meaning false information and made-up reasoning or citations, as a known risk of generative AI (NIST Generative AI Profile). And a bot that hands over finished homework undermines the skill your course is meant to build.

5. Buying more ads before fixing the offer

Ads can bring people to the course page. They can’t fix a weak offer. If visitors land on a long list of videos, broad promises and no visible support, more traffic just means more people leaving confused. Fix the product first, then make the website explain it. (If the site itself is the leak, our 30-day plan for a website that isn’t generating leads is a good place to start.)

Start with a skill students can show

Start with a situation your learner wants to handle.

For an English teacher, that might be joining a work meeting without rehearsing every sentence. For a coach, running a difficult conversation with a team member. For a technical instructor, building a working project and explaining the decisions behind it.

Write down four things: where the student starts, what good performance looks like, what practice gets them there and what evidence will show they have improved. Keep the promise inside what your programme can actually support.

This fixes your sales page too. Instead of “unlock your potential”, you can show the task, the practice and how progress gets checked, and prospective students can picture the before and after.

	Information course	Practice-based course
What the student buys	Access to lessons	A skill they can demonstrate
After each lesson	The next video	A task: record, solve, draft or rehearse
Feedback	Comments section, if anyone replies	Scheduled review against clear criteria
When a student goes quiet	Nobody notices	Someone checks in
Sales page promise	“120 lessons, lifetime access”	“Run a 10-minute meeting in English by week 8”
Competes with AI on	Explanations, and loses	Practice, judgement and accountability

Put practice at the centre

Go through your course module by module and ask one question: what does the student do after watching this?

If the answer is “watch the next video”, that module needs work. Better options are recording a spoken answer, solving a problem, writing a draft, rehearsing a conversation or making a decision with incomplete information.

Research backs this up. Freeman and colleagues pooled 225 studies of undergraduate science, engineering and maths teaching. Students in active-learning classes scored about 6% higher on exams, and students in traditional lectures were 1.5 times more likely to fail (Freeman et al., PNAS). That is university STEM, not your Tuesday-night business course, but it is a good reason to distrust passive delivery anywhere.

Live sessions should serve the practice. A workshop or case discussion earns its place when students bring something to try, examine or improve. A live lecture is still a lecture.

Keep short recorded explanations where they help students prepare, and spend your own time where observation, discussion and judgement matter most.

Build a weekly rhythm students can keep

A workable week might be one short explanation, a few brief practice sessions, one submission and a live session about whatever went wrong.

Bring old material back, and ask students to recall and apply it rather than re-read it. In their review of learning techniques, Dunlosky and colleagues rated two methods as broadly useful across subjects and ages: practice testing and spreading practice out over time (Dunlosky et al., Psychological Science in the Public Interest).

AI is good at generating extra exercises and revision schedules. You decide whether those exercises match the skill and level the student actually needs.

Accountability and community that do something

Accountability with a structure

AI can send reminders, suggest schedules and say encouraging things, and that helps a little. A commitment to a real teacher or group works differently. The student knows someone expects their work, will notice if they disappear and will help them get back on track when life gets messy.

Build that into the programme with manageable weekly tasks, clear submission dates, a predictable feedback schedule and a simple, guilt-free way back after a missed week.

Don’t let it turn into surveillance. The point is to get people back to useful work. Your support should answer one question: “When I get stuck, who helps me take the next step?”

Give the community a job

A group chat is not a community anyone will pay for. People need a reason to talk to each other.

Language learners can practise conversations in pairs. Coaching students can discuss anonymised cases. Technical students can demo projects and defend their choices to peers. Set up recurring activities with a purpose, such as matched practice partners, small feedback groups or a weekly “here’s what I finished and what I learned” thread.

AI can help organise all of this, but it can’t be the other people. For a lot of students, those relationships end up being a big part of why they stay.

Make your judgement visible

Your expertise is worth more when students can see how you assess things. Explain why an answer works in one situation and not in another. Show the difference between a beginner’s mistake and a choice that depends on context. Be open about where your advice stops.

Being human doesn’t make you right. Trust comes from showing your reasoning and correcting yourself when you’re wrong. A teacher who says “here’s how I checked this, and here’s when I’d do it differently” gives students something a confident chatbot answer doesn’t.

What AI does well, and what still needs you

Plenty of teaching tasks can be shared. The useful line isn’t “AI bad, human good”. It runs between capability and responsibility: AI can do a lot of the work, but it can’t be accountable for a student’s progress or be in a real relationship with them.

Learning need	What AI can do	What still needs a human teacher
Explanations	Rephrase ideas and generate examples instantly	Draw on real experience and take responsibility for teaching decisions
Routine feedback	Suggest corrections and offer extra practice on the spot	Make an accountable professional judgement about the student’s work
Motivation	Prompts, encouragement and study plans	A real commitment to someone who knows the learner
Conversation practice	Simulate scenarios for rehearsal	Real conversation with real social stakes
Community	Help organise groups and activities	Relationships between people with shared goals
Assessment	Draft questions and help with marking	Stand behind a grade or an endorsement

The best course designs make these roles obvious. Students should know when they are working with AI, when a teacher reviews their work and how to ask for a human.

Using AI without letting it do the learning

Teach students how to use the assistant. They should try the task first, ask for a hint before a full answer, and then explain what they changed and why.

Why bother with rules? A 2025 field experiment with nearly 1,000 high-school maths students in Turkey shows what happens without them. Students given a plain GPT-4 interface did better during practice, then scored 17% lower than the control group on an exam taken without AI. A tutor version with learning safeguards largely removed that drop (Bastani et al.). That’s one setting and one subject, so it doesn’t prove AI always harms learning. It does show that “did well with AI” and “can do it alone” need to be measured separately.

Guardrails worth building in

• Set some tasks that students must complete without the assistant.
• Where using AI is part of the skill, assess how well students use it and whether they check its output.
• Keep the assistant inside your curriculum, approved material and teaching rules.
• Test it on typical mistakes and ambiguous questions before students ever see it.
• Give students an obvious route to a human review.
• Tell students what data the tool processes, and don’t collect sensitive details you don’t need. UNESCO’s guidance puts human agency, age-appropriate use and data protection at the centre (UNESCO guidance on generative AI in education).

This is how the AI tutors I’ve built for English with Sue and Dovydenko School work: they answer from each teacher’s own material rather than the open internet. If you want an assistant on your public site as well, the same thinking from our AI chatbot rollout guide applies: a narrow job, approved content and a clear handoff to a person.

Pricing mentorship without burning out

Mentorship can justify a higher price when it includes real access to your expertise and feedback. “Unlimited support” is how teachers burn out.

Define it before you sell it: how often students meet you, how much work you review, how quickly you reply and which questions need a separate booking.

Tiers help. Independent study, supported group learning and one-to-one mentoring can sit at different prices, as long as each one says plainly what is included.

Then do the maths. Add up teaching and support hours per student, platform fees, acquisition costs and refunds. AI savings only count if they cut work without cutting what students are paying for.

A four-week plan to rebuild one course

You don’t need to rebuild the whole business. Pick one learning journey and work on that.

1. Week 1. Talk to recent students. Find where they struggle or quietly stop.
2. Week 2. Choose one outcome and redesign the practice around it.
3. Week 3. Add one AI or automation feature that removes a real bottleneck.
4. Week 4. Pilot with a small group and review what happened.

Track whether students start their first task, practise regularly, submit work and show improvement. Track your own support hours, conversion and how many students stay. Compare similar groups where you can, and note anything else that changed at the same time. A busier dashboard is only useful if it changes a decision.

Where a developer fits in

I’m Vadym Leonov. I build websites and web apps, set up AI and automation, and connect them to the marketing side of a business. A good share of my clients teach for a living.

Two of them are Aussie English and English with Sue, and they show why one template doesn’t fit every course business. Aussie English combines Australian English with podcasts, cultural context and speaking practice. English with Sue is built around conversational confidence and one-to-one lessons. Students choose them for different reasons, so each website has to make different strengths easy to see and use.

With a course business, I usually start by connecting what the student wants to achieve, how the teacher gets them there and where the current website or systems get in the way. The work that follows might be a clearer course page, smoother registration and payments, booking and email tools that talk to each other, or a student area where the next task is obvious.

I add AI where it has a defined job: a course assistant, guided practice, help finding the right lesson or answers to the admin questions you’re tired of. How far it goes depends on your materials and how much teacher oversight you want. You work with me directly, so the build follows your teaching method instead of a generic template. If you’d rather have ongoing help than a one-off project, my tech partner plans for course creators cover the platform side month to month.

So, do we still need courses?

Yes. We need fewer courses that are just libraries, and more that get people practising, give them honest feedback and notice when they drift.

You have already put years into becoming a good teacher, and AI doesn’t cancel that. It takes over much of the explaining and leaves you the parts students struggle with on their own. Teachers who use it carefully will set a standard that’s hard to match, whether you ignore AI or let it run the course for you.

Frequently asked questions

Sources

• Kestin et al. (2025), AI tutoring outperforms in-class active learning, Scientific Reports
• Bastani et al. (2025), Generative AI can harm learning: evidence from high school mathematics
• Freeman et al. (2014), Active learning increases student performance in science, engineering, and mathematics, PNAS
• Dunlosky et al. (2013), Improving students’ learning with effective learning techniques
• NIST AI 600-1 (2024), Artificial Intelligence Risk Management Framework: Generative AI Profile
• UNESCO (2023), Guidance for generative AI in education and research