Introduction

A student buys your photography course, watches the first three lessons and stops at an exercise. They understand what the controls do. They cannot work out which one to change for the picture they want to take.

The question is small: “The subject looks sharp, but the background is still distracting. What am I missing?”

Finding an answer may mean replaying a long video, searching a community thread or waiting for a reply. A general AI assistant can offer photography advice, but it may introduce techniques you have not taught, assume different equipment or skip the exercise’s purpose.

AI support inside an online course is most useful when it helps the learner work through that particular moment. The student should be able to ask in their own words, understand the explanation and return to something they can try.

For the creator, the challenge is deciding what kind of help the agent should provide, what material it can rely on and when a human answer is needed.

Start with the learning task

Before choosing a tool, write down the situations that regularly bring students to a halt.

Some need a simpler explanation of a term. Others understand an explanation but cannot apply it. A student may need help finding an earlier lesson, choosing the right exercise or understanding what a course includes.

These are different tasks. An answer that solves one can fail at another.

If a student asks where to find the practice file, give them a clear route to the file. If they ask why their result looks wrong, a useful response may begin with a question about what they tried. If they ask for the completed assignment, supplying it immediately might defeat the purpose of the lesson.

The OECD’s Digital Education Outlook 2026 draws an important distinction between better performance on a task and genuine learning. It cautions that outsourcing work to general-purpose AI without educational guidance can improve the immediate output without producing learning gains. For a course creator, this is a reason to judge the help by what the student can subsequently do, rather than how impressive the response sounds.

Make the course material usable as a source

Your teaching method may be obvious to you because you built it over years. An AI agent needs explicit material to work with.

For each lesson, prepare a concise source sheet containing:

  • The lesson’s purpose and the skill the student is practising.
  • The concepts and terminology used in your explanation.
  • Any prerequisites, equipment or earlier lessons it assumes.
  • The exercise and what a reasonable attempt should demonstrate.
  • Common misunderstandings and examples you are comfortable teaching.
  • Topics that belong elsewhere or require your direct input.

Suppose a photography lesson introduces aperture through a simple exercise. The source sheet could explain that the learner should keep the subject in the same place, change the aperture and compare the background. It might also explain that distance from the background affects the result.

That is much more useful than a short instruction to “answer photography questions in a friendly way.” It gives the agent a teaching purpose and a reference point.

For video lessons, provide written teaching notes covering the explanation, examples and common questions. The Coach uses the written material supplied to it; it does not watch the video itself.

Choose the kind of help you want to offer

A useful course agent can be evaluated against three levels of help.

Clarification: Explain an idea in simpler language while preserving its meaning. The learner should not need to recognise the correct technical term before asking.

Guided practice: Help the student examine an attempt or decide what to try next. This may require a clarifying question, a small hint or a reference to an earlier concept.

Course navigation: Help the learner find the relevant material and continue. This is valuable even when the question does not require a sophisticated explanation.

You do not need every level for every lesson. A reference course may favour direct answers. A skills course may favour hints and practice. An assessment may need a clear boundary around giving away solutions.

Tell learners what the support is for. “Ask for help understanding a lesson or finding the next exercise” gives a clearer expectation than “Your personal expert for anything.”

Keep the explanation connected to the experience

Consider this illustrative exchange in the photography course:

Student: “I tried the exercise, but the background is still too clear.”

Agent: “How far is your subject from the background? This exercise changes aperture, but subject-to-background distance also affects the result.”

After the student answers, the next response can stay focused on one thing to try, with the relevant lesson or exercise available alongside it where the system supports that interaction.

The important design question is what the student sees and can do next. A long answer that leaves them searching for the exercise adds work. A short explanation beside the relevant material can keep the learning task in view.

In PAKA Learn, students can ask the course Coach questions in the context of the lesson they are viewing and their progress. The Coach uses the published course material and teaching notes available to that student, can point them to relevant lessons or discussions, and supports referring questions to the creator under the creator’s teaching rules. The customer space keeps purchased content and course progress together, so support continues after the sale.

Give the agent a useful way to say “I don’t know”

Sometimes the material does not contain an answer. Sometimes the student’s description is too vague. Sometimes an answer would require seeing or hearing something the system cannot inspect.

In those cases, the agent should make the limitation clear. “I cannot assess the photograph from your description alone” is more useful than confidently diagnosing a problem it has not observed.

Define what happens next. Depending on your course and available tools, that might mean asking for a missing detail, pointing to the relevant lesson or explaining how to contact the instructor. Do not promise instant personal feedback unless that is part of what you sell.

Set equally clear boundaries around the teacher’s identity. An AI-generated explanation should not be presented as a personal review from the instructor. UNESCO’s guidance on generative AI in education advocates a human-centred approach. Keeping the creator accountable for the learning experience, while being transparent about AI support, is a practical way to apply that principle.

Test the moments that a polished demo misses

Before inviting students to use the agent, test it against your own material. Keep a record of the question, the expected behaviour and the actual response.

Try these six cases:

  1. A beginner’s question without technical vocabulary. Does the explanation meet the learner at their level?
  2. A plausible misunderstanding. Does the answer correct the idea without introducing a new error?
  3. A request for the full exercise solution. Does it follow your teaching approach?
  4. A question outside the course. Does it acknowledge the limit rather than inventing your position?
  5. A request for material the person has not purchased. Does access remain correctly restricted?
  6. A returning student looking for the next step. Can they find and resume the relevant work without unnecessary friction?

Access controls deserve a real test with separate accounts. A friendly instruction to avoid sharing paid content is not a substitute for the system enforcing who can access it.

Also test a learner who does not want to chat. Lessons, resources and basic navigation should remain usable without requiring a conversation.

Measure progress, not just message counts

A busy chat is not necessarily a successful learning experience. A student may send many messages because an explanation is confusing or the next action is hard to find.

Start with observable questions: Did the learner find the material? Could they attempt the exercise after receiving help? Which questions repeatedly required an instructor? Which explanations needed correction?

If you compare course completion before and after introducing AI support, account for other changes, such as a different audience, revised lessons or a new promotion. Do not attribute every improvement to the agent.

Repeated questions can be useful editorial feedback. If several students get stuck at the same point, improve the lesson itself. The support system should help you spot a teaching gap, not become a permanent workaround for it.

Build around one difficult lesson first

Choose a lesson where students regularly need help. Prepare the source material, define the allowed help, test the edge cases and watch whether learners can move forward.

That small exercise will tell you more than a demonstration filled with questions the agent already handles well.

The opportunity for a course creator is straightforward: make your expertise easier to use at the moment a student needs it. Keep the explanation tied to your material, make the next step easy to find and retain a clear route back to you when AI is not enough.

Explore PAKA’s agent-led business experience.

Experience the idea.Try a customer experience