Why most AI course tools miss the mark - and what good instructional design actually looks like

Every AI course tool on the market promises speed. Most of them deliver it. Type a topic, wait thirty seconds, receive a ten-module course with quizzes, learning objectives and section headers. It looks like a course. It has all the structural components of a course.

But in most cases, it doesn't actually teach anything.

This is the central problem with the current generation of AI course creation tools - and it's a problem that most buyers don't discover until after they've published. Learners complete the course. They tick the compliance box. And three months later, the behaviour the course was supposed to change hasn't changed at all.

After twenty years designing professional courses for over 1,000 organisations across healthcare, finance, legal and the public sector, I've seen this pattern many times. The course looks right. The learning isn't happening. And the reason, almost always, comes back to the same thing: content without instructional design.

The problem with fast content

When you ask a language model to generate a course, you get content. Organised, readable, coherent content - often genuinely well-written. What you don't automatically get is learning.

The distinction matters enormously. Content is information arranged in sequence. Learning is a deliberate change in what someone knows, can do, or believes - achieved through a specific sequence of exposure, practice, feedback and retrieval. Those are not the same thing, and the gap between them is what instructional designers spend their careers trying to close.

"Content is information arranged in sequence. Learning is a deliberate change in what someone knows, can do, or believes. These are not the same thing."

A course that simply presents information and then asks some questions at the end is not teaching. It's testing. And testing without teaching is not how adults learn complex skills, change habitual behaviour, or retain information beyond the next morning.

What instructional design actually is

Instructional design is the discipline of deliberately structuring learning to produce specific outcomes. It is not the same as writing clearly (though that helps). It is not the same as presenting information logically (though that is necessary). It is the engineering of a learning experience - and like all engineering, it requires both expertise and intent.

The core principles that experienced instructional designers apply, consistently, to every course they build include:

• Learning objective alignment. Every section of a course should serve a declared objective. Every assessment question should test whether that objective has been met. Not adjacent to it. Not vaguely related to it. Directly, measurably tied to it.
• Cognitive load management. Human working memory is limited. A well-designed course introduces concepts in a sequence that builds on prior knowledge, spaces complexity appropriately and avoids overwhelming the learner with more than they can process at once.
• Bloom's Taxonomy-appropriate assessment. Not all learning is knowledge recall. If the objective is for a learner to apply a principle in a real situation, testing whether they can remember what it's called is not assessment - it's false assurance. Bloom's Taxonomy defines six levels of cognitive complexity. Professional instructional designers match assessment to the required level every time.
• Spaced practice and retrieval. Learning sticks when it is retrieved, not just received. Well-designed courses build in retrieval opportunities - not as box-ticking exercises but as deliberate reinforcement of the concepts that matter most.
• Scenario-based application. Abstract principles become usable skills when learners practice applying them in realistic contexts. Scenario-based learning - where the learner must respond to a realistic situation rather than choose from a list of facts - is one of the most robust methods for building transferable capability.

Five signs a course wasn't designed

You don't need to be an instructional designer to spot the difference between a course that was built by one and a course that wasn't. Here are the five signs I look for:

1. The quiz questions only test recall. "What is the name of the regulation that covers...?" is a knowledge question. If every question in the assessment is a knowledge question, and the course objective was for learners to apply the regulation in their work, the assessment is measuring the wrong thing.
2. The sections don't connect. Each section covers a topic, but there's no narrative thread linking them. Learners finish the course with a collection of facts rather than a coherent framework they can actually use.
3. The examples are generic. "A manager named Alex faces a difficult situation at work..." If the examples could apply to any industry, any role and any organisation, they won't resonate with any learner specifically. Good design uses examples that feel real to the audience.
4. The objectives are stated but not served. The course opens with five learning objectives, then covers seventeen loosely related topics, none of which directly addresses any of the five. The objectives are decoration.
5. The learner is never asked to do anything. Read, scroll, click Next, answer a recall question. Repeat. A course where the learner is entirely passive throughout is not building capability - it is manufacturing compliance records.

Quick diagnostic

• Do the quiz questions test application or just recall?
• Could you explain what the learner will be able to do differently after completing it?
• Are the examples specific to this audience - or could they belong to any course?
• Does the structure build - or is it just a list of topics?
• Is the learner asked to practise anything - or just to receive information?

What good looks like in practice

A well-designed course for, say, a compliance objective - "learners will correctly apply the data subject access request process in their role" - would look something like this:

It opens not with a definition of DSAR, but with a scenario: a customer calls and makes a request. The learner's first task is to identify whether this constitutes a DSAR. This activates prior knowledge and establishes why the learning matters before a single piece of information has been presented.

The content that follows builds a framework - not a list of facts, but a structured model that the learner can apply. Each element is introduced with an example drawn from the learner's specific context. Complexity is introduced progressively, not all at once.

Retrieval is built in - not as a quiz at the end, but as short check questions throughout, spaced to reinforce the most critical concepts at the moment they are most likely to be forgotten.

The assessment requires the learner to handle three different DSAR scenarios - each with a slightly different nuance - and make the correct decision in each. The scoring does not just record pass or fail: it feeds back on the specific error pattern so the learner understands what they got wrong and why.

That is not what an AI language model produces by default. It requires deliberate design - and either a skilled human applying that design, or a platform where those principles are built into the generation process itself.

The AI question

So does this mean AI course creation tools are useless? Not at all. The problem is not AI - it's the absence of instructional design principles in how most AI tools generate content.

When AI is given explicit instructional design parameters - clear learning objectives, appropriate cognitive level, scenario requirements, assessment alignment rules - the output improves dramatically. The speed advantage of AI is genuinely transformative. A course that would take 190 hours to build with traditional tools can be produced in under seven hours when AI handles the structural and content work.

But the quality of that output depends entirely on whether the AI has been built with instructional design as a foundation - not as an optional feature, or a prompt you have to add yourself, but as a built-in standard that applies to every output automatically.

How CourseAgent approaches this

CourseAgent was built by the same team that has designed professional courses for 1,000+ organisations over 20 years. Every generation prompt, every quality audit layer and every assessment template is built on the instructional design standards we've applied to over 5,000 professional courses. The AI doesn't just generate content - it applies the same principles an experienced designer would.

Read more about how CourseAgent was built →

What to look for in a tool

If you're evaluating AI course creation tools - whether for your consultancy, your L&D team or your own knowledge-sharing practice - the questions to ask are not primarily about speed or price. They are:

• Does the tool align assessments to learning objectives automatically - or do I have to do that myself?
• Can the tool generate scenario-based content - or does it only produce information and recall questions?
• Does the output structure reflect progressive complexity - or is it just a flat list of topics?
• Is there a quality audit process - or do I publish and hope?
• Can the tool explain what it has designed and why - or is the output a black box?

Speed matters. But speed producing the wrong thing at scale is not progress - it is efficiency in the service of ineffectiveness. The organisations that will benefit most from AI in L&D are the ones that insist on both.