How We Built CARL’s Lesson Structure (and Why It Looks the Way It Does)

When we started building CARL, we knew we needed a lesson structure that could work under AI generation and still actually work in real classrooms.

That’s harder than it sounds.

Most lesson planning frameworks are designed for one teacher, teaching one class, with full control over pacing, content, and context. CARL had to generate lessons for any grade, any subject, any teacher, and still feel pedagogically sound.

We couldn’t just pick one instructional model and call it done.

The Starting Point: BOPPPS and Student Engagement

I first learned about BOPPPS during an Effective Learning Environments course in my undergrad. It resurfaced again during my MEd program when I was studying student engagement more deeply.

BOPPPS stands for:

  • Bridge-in
  • Outcomes
  • Pre-assessment
  • Participatory Learning
  • Post-assessment
  • Summary

It’s clean. It flows well. And most importantly, it starts with a hook.

That mattered because the research was clear. Relevance and agency are two of the strongest drivers of student engagement. Hooks are how you make learning feel relevant from the very beginning.

But here’s the thing: one hook isn’t enough.

Students have different interests, identities, and entry points. A hook that lands for one student might completely miss another. So we built CARL to generate dual hooks by design.

The first hook is usually culturally or age-relevant, something that feels immediate and relatable. The second hook is an opposing entry point, designed to catch students who don’t connect with the first one.

It’s about casting a wider net. If one doesn’t land, the other might.

Adding Structure: GRR (Gradual Release of Responsibility)

BOPPPS gave us the flow, but “Participatory Learning” as a phase was too vague on its own. Teachers know students learn by doing, but the real question is how you scaffold that effectively.

That’s where GRR came in.

GRR follows a clear instructional sequence:

  • I Do
  • We Do
  • You Do

We embedded GRR inside the Participatory Learning phase to give teachers concrete instructional moves. Model it first, guide practice together, then release students to try independently.

This also solved a very practical classroom problem: attention span.

One thing I consistently heard from BC teachers during my MEd program was that students do better when they’re actively learning, but lining up enough meaningful activities to keep engagement high is hard.

By breaking lessons into multiple parts, such as Explore, Explain using GRR, and Apply, CARL builds in natural transitions. This keeps lessons moving and prevents long stretches of passive instruction where students start to drift.

The Naming: Leaning Toward 5E

As we iterated, the phase names started sounding more like the 5E model:

  • Engage
  • Explore
  • Explain
  • Elaborate
  • Evaluate

That wasn’t intentional at first. It happened because the language was simply clearer and more intuitive for teachers.

The underlying structure didn’t change. BOPPPS and GRR were still doing the heavy lifting.

We weren’t trying to be purists about any one framework. We were trying to build something that worked.

Building It to Work Under AI Generation

One thing that isn’t always obvious from the outside is that CARL doesn’t just say “make a lesson plan” and generate everything at once.

Each section of a CARL lesson, hooks, learning outcomes, activities, assessments, supports, is generated using its own detailed and specific instructions. That’s how we maintain pedagogical quality at scale.

It also means the AI isn’t freestyling.

It’s following a structured blueprint that enforces the BOPPPS + GRR + 5E synthesis we designed. Dual hooks are not optional. GRR scaffolding isn’t skippable. Reflection is always present.

Those constraints are what make it work.

You can generate a lesson for Grade 2 literacy or Grade 12 physics, and the underlying instructional logic stays sound.

The Flexibility Requirement

One thing BC teachers made very clear, both in academic literature and in conversations with my MEd colleagues, is that they didn’t want rigid lesson templates.

They wanted structure, but also the freedom to adapt.

So CARL generates a lot of content. Multiple hooks. Layered activities. Scaffolded supports. Reflection prompts. The idea is not that teachers must use everything, but that they can choose what works, cut what doesn’t, and edit freely.

This came directly from critiques teachers shared about other AI lesson planners. Either they’re hard to edit, or they produce barebones outlines that still require hours of work.

CARL gives you the structure. You stay in control.

What We Ended Up With

CARL’s lesson structure isn’t BOPPPS. It’s not GRR. It’s not 5E.

It’s a synthesis.

We pulled the strongest instructional moves from each framework and collapsed them into something teachers already use intuitively:

  • BOPPPS’ flow and clarity, including hooks, outcomes, participatory learning, and reflection
  • GRR’s scaffolding logic, model, guide, release
  • 5E’s clearer phase naming, Engage, Explore, Explain, Apply, Reflect
  • Dual hooks to reach more students
  • Built-in flexibility so teachers can adapt without starting from scratch

Because each section is generated with its own thorough and specific instructions, the pedagogical integrity holds across grades and subjects.

This structure wasn’t designed in one sitting. It emerged gradually through iteration, research, and listening closely to what BC teachers actually needed.

And honestly, that messiness, the synthesis, the adaptation, the refusal to follow one model perfectly, is exactly what makes it work.

Curious what this looks like in practice? You can explore real, classroom-ready CARL lessons at hicarl.ai/sample-lessons.

Want to see CARL’s lesson plan framework in action? Explore what makes our lessons classroom-ready, discover our built-in lesson enhancements, or see what teachers found during beta testing.