AI in Classroom Learning: From Learning to Use AI to AI for Learning

Introduction: Moving Beyond Tool Adoption to Pedagogical Purpose

You have probably sat through at least one training session on AI this year. Someone opened a chatbot, typed a request for a quiz, and the room nodded. Then Monday arrived, and you faced the same question you had before the session started: what is this actually for?

That question matters more than any tool. Many schools have spent the last two years helping teachers learn to use AI. Logging in. Writing a first prompt. Generating a worksheet. All of that is useful, but it stops short. The harder and more important shift is towards AI for learning, where every decision about a tool begins with a decision about thinking.

Stephanie Holt, Director of Teaching and Learning at DSB International School, made exactly this case in a Suraasa masterclass hosted by Loulou Husky. Her argument is simple to state and demanding to practise. AI belongs in your classroom only where it deepens student thinking. It belongs in your planning wherever it frees you to do the human work that no machine can do.

If you feel uncertain, you are in good company. A Pew Research Center survey conducted in fall 2023 found that a majority of U.S. teachers were uncertain about or saw downsides to the general use of AI tools in K-12 education. That caution is professional judgement, not resistance. This guide turns it into a plan.

You will find four things here. A clear reason why the skills your students need have changed. A practical framework for telling students when AI is allowed. A structure for prompts that produce lesson material worth using. And a way to assess AI-supported work that rewards thinking rather than output. Throughout, the teacher stays at the centre, because that is where the evidence says you belong.

The Skill Shift: Preparing Students for the Intelligence Age

Holt frames the change as a move across three ages. In the industrial age, schools prized memory: the student who could recall the most facts did well. In the information age, the premium moved to retrieval: knowing where to find information and how to pull it out quickly. In what she calls the intelligence age, both of those skills have lost much of their value, because a machine now recalls and retrieves faster than any person.

What remains scarce is the human layer. Synthesis, which means combining ideas from different places into something new. Evaluation, which means judging whether a claim, an argument or a source holds up. Critical thinking. And metacognition, the ability to notice and steer your own thinking. Holt argues that AI-augmented workplaces want precisely these competencies, and that classrooms built mainly around recall now prepare students for jobs that are disappearing.

If you want to hear Stephanie Holt explain this shift and walk through her live demonstrations in her own words, the full session is below.

This is not an argument for dropping knowledge. Students still need facts in long-term memory to think with. The point is that knowledge becomes the starting material, and the lesson has to go further. The OECD Digital Education Outlook 2023 examines how digital tools are reshaping education systems, and it is worth reading alongside Holt's argument if you are planning at department or school level.

There is also a real risk in getting this wrong. The Brookings Institution describes a 2024 field experiment with nearly 1,000 Turkish high-school students. Unrestricted access to ChatGPT during math practice improved practice-problem accuracy, yet reduced subsequent test scores by 17 percent. Read that twice. The students looked better while they practised and learned less. AI used without design can quietly remove the very effort that produces learning.

That finding is the strongest case for everything that follows. The question is never whether students use AI. It is whether the task still makes them think.

Modeling Ethics: Transparent AI Usage in Daily Teaching

Students copy what you do far more reliably than what you say. Holt's point is direct: if you never acknowledge when you use AI to produce questions, examples or lesson materials, your students will never learn to declare their own use. Honesty about AI is a habit, and habits are modelled.

"If you can teach without using AI, don't use it. Use the fun thing, use the hands-on thing. But wherever you use AI, and wherever I use AI in the classroom, I acknowledge my use of AI.", Stephanie Holt

Notice the two halves of that statement. The first is a filter. AI is not the default. If a hands-on activity, a discussion or a physical model teaches the idea better, choose that. The second is a commitment. When AI does earn its place, you say so out loud.

What does that look like in a normal week? A few simple routines work well:

• A line on the slide. "These retrieval questions were drafted with AI and checked by me." Eleven words, and students see disclosure as normal practice.
• A think-aloud when you edit. Show the class an AI draft and talk through what you kept, what you cut and why. That is metacognition made visible.
• A shared vocabulary. Agree terms such as "AI-drafted", "AI-assisted research" and "no AI" so students can label their own work precisely.
• An honest mistake. When an AI tool gets something wrong, show it. Students learn more from watching you catch an error than from a flawless output.

This approach matches the wider policy direction. UNESCO's guidance on generative AI in education and research proposes a human-agent and age-appropriate approach to the ethical validation and pedagogical design processes. In plain terms, a human stays responsible for every judgement, and the use fits the age of the learner. Your disclosure routine is that principle in action, scaled down to one classroom.

It also protects you. When you are open about your own use, conversations about student use stop feeling like accusations. You are not policing a secret. You are teaching a professional norm.

The Traffic Light Framework: Clarifying Classroom Expectations

Most confusion about AI in classroom learning comes from unclear rules. Students genuinely do not know whether a task permits AI, and teachers often decide only after the work comes in. Holt's traffic light model fixes this by labelling every task before students start.

The crucial detail is the last column. In Holt's model, every tier requires transparent attribution. Green does not mean "anything goes". It means AI is welcome, and you still say how you used it.

Here is how you might put the model to work in a single unit. Imagine a Grade 9 unit on climate and food security:

• Red: A short written explanation, done in class, of how rising temperatures affect one crop. This checks what students actually understand.
• Amber: Students use AI to suggest three regions worth researching and a structure for their report. The research and writing remain theirs.
• Green: Students ask AI for a persuasive paragraph on a policy solution, then annotate its weaknesses, missing evidence and bias.

Notice how the Green task is not easier. Critiquing a fluent but flawed paragraph demands evaluation, the very skill the intelligence age rewards. The traffic light does not just restrict AI. It helps you choose where AI creates thinking and where it would replace it.

A simple declaration template helps students meet the attribution rule. Ask them to answer three questions at the end of any Amber or Green task: Which tool did you use? What did you ask it? What did you change, reject or add, and why? That third question quietly turns every assignment into a metacognitive exercise.

Maximizing Teacher Time: AI as an Administrative Personal Secretary

Now for the part most teachers feel immediately. You are measured on outcomes, but most of your week disappears into preparation, paperwork and data. Holt describes AI as a personal secretary, and the description fits. It drafts, sorts and summarises so that you can spend your energy on decisions.

Her most striking example is the scheme of work. Using tools such as ChatGPT, Gemini and NotebookLM, she shows how drafting a full scheme of work can shrink from two to three weeks down to roughly two hours. That is not a small gain. It is the difference between a half-term plan rushed together in the holidays and one you can actually refine.

The secretary model covers far more than schemes of work:

• Drafting. Unit overviews, parent letters, newsletter text and first versions of worksheets.
• Processing assessment data. Summarising mark sheets, spotting patterns and grouping students by need.
• Organising. Turning meeting notes into action lists and helping you plan a busy schedule.
• Working from your own documents. Tools like NotebookLM let you ground responses in curriculum documents or texts you upload.

One condition is non-negotiable. Holt stresses that human oversight and rigorous quality control remain your job. A secretary drafts. You sign. Every AI-produced scheme needs checking against your curriculum, your students and your professional sense of what works. The two hours includes that checking. Skip it, and you are not saving time. You are borrowing it from a future lesson that goes wrong.

For a wider tour of tools that go further than a basic chatbot, our guide to AI tools for educator productivity breaks them down by task.

Mastering the Prompt: Structuring Iterative Prompts for Lesson Planning

Ask a vague question and you get a generic lesson. "Make me a lesson on photosynthesis" produces something that could belong to any class in any country, which means it belongs to none. Holt's point is that strong prompts carry pedagogical constraints. They describe the lesson the way you would describe it to a skilled colleague.

According to Holt, a well-built lesson prompt specifies:

1. Student age and stage.
2. Learning objectives: what students should know or do by the end.
3. Success criteria: how you and they will recognise success.
4. A hook activity to open the lesson.
5. Differentiation needs for support and stretch.
6. Assessment methods, including formative checks during the lesson.
7. Required materials.

Here is what that looks like in practice:

"Act as an experienced science teacher. Plan a 50-minute lesson for 13-year-olds on photosynthesis. Objective: students can explain how light, water and carbon dioxide combine to produce glucose and oxygen. Success criteria: students label the inputs and outputs on a diagram and explain one limiting factor in their own words. Begin with a hook using a plant kept in the dark for a week. Include support for two students who are new to English and an extension task for fast finishers. Add two formative checks mid-lesson and an exit ticket. List all materials."

That prompt is longer, but it saves time, because the first draft lands close to usable. Then you iterate. Ask the tool to simplify the language for the English learners. Ask it to replace the exit ticket with one that tests application rather than recall. Ask it to sequence the lesson using the 5E lesson plan model. Each round is a conversation, not a vending machine.

Build a reusable prompt library from your best versions. Keep a document with your standard parameters already filled in for each class you teach, then change only the topic and objective. Over a term, that library becomes one of your most valuable professional tools. Our guide to prompting frameworks for lesson planning offers more structures you can adapt, and ChatGPT for teachers covers assessment and productivity uses in depth.

One warning belongs beside every prompt you write.

"Remember, AI is a sycophant. It wants to please you, and if it hasn't got those citations to hand, it will make them up.", Stephanie Holt

So ask for real-world citations or source texts, and then check them yourself. Open the link. Find the page. Confirm the quote exists. If a source cannot be verified, it does not go in front of students. Teach your students this same discipline, and you have taught them one of the most important research habits of their lifetime.

Aligning AI with Bloom’s Taxonomy from Retrieval to Synthesis

Bloom's Taxonomy gives you a ready-made map for deciding where AI helps thinking and where it replaces it. Holt shows how AI can scaffold every level, from basic factual recall up to creative evaluation. The trick is designing the task so the student, not the tool, does the cognitive work at the level you are targeting.

Holt's most memorable example sits at the top of the taxonomy. Students can ask AI to simulate Shakespeare explaining Macbeth in period language. On its own, that is a novelty. The learning comes next: students critique the persona's reading of the play, compare it with the text, decide where it rings true and where it flattens the character, and reflect on how they reached their judgement. The AI provides a provocation. The students provide the thinking.

"AI should be a tool to enhance learning, not a crutch to avoid thinking. And I'm a big believer in if it's not helping the learning, don't do it.", Stephanie Holt

That sentence is a useful test for any task you design. Ask yourself: if a student used AI here, would they think more or think less? If the answer is less, change the task or switch the light to Red. For a deeper look at writing objectives and assessments at each level, see our practical guide to Bloom's Taxonomy in the classroom.

Personalized Learning, Instant Feedback, and Predictive Analytics

The most powerful use of AI may be the one students never see. Holt describes using AI to analyse patterns in cohort examination results and pinpoint precise conceptual gaps. In her example, the analysis surfaced specific weaknesses in grammar and sentence structure across a cohort. That let teachers design targeted short-term intervention schemes rather than reteaching everything.

Think about what that changes. Instead of a vague sense that "writing needs work", you know which skill, for which students, and roughly how far off they are. Here is a practical cycle you can run after any mock exam:

1. Anonymise the data. Remove names and identifying details before sharing anything with an AI tool, and follow your school's data protection policy.
2. Ask for patterns. Request a breakdown of errors by skill, question type and student group.
3. Check against your knowledge. Does the pattern match what you see in class? If not, investigate before acting.
4. Plan a short intervention. Two or three weeks focused on the one or two gaps that matter most.
5. Re-assess. Use a brief, targeted check to see whether the gap has closed.

The timing is the point. Using predictive analysis on mock exams lets you spot skill deficits before high-stakes assessments, while there is still time to act. That is assessment for learning in its truest form. If you want to sharpen that distinction, our guide on assessment for learning vs assessment of learning is a useful companion.

Instant feedback works the same way at a smaller scale. AI can give students quick comments on a draft paragraph or a set of practice problems, which means they correct misconceptions while the lesson is fresh. Personalisation follows naturally: different practice sets for different needs, adjusted reading levels, extra worked examples for those who want them.

Keep the human in the loop here too. The U.S. Department of Education Office of Educational Technology report on AI and the future of teaching and learning explores these opportunities and risks for schools in detail, and it is a sensible read before you build any data workflow. Meanwhile, the NCES School Pulse Panel publishes recurring survey results from U.S. public schools, a reminder that school leaders increasingly track how conditions and practices shift over time. AI can show you a pattern. Only you know the student behind the number.

Assessing AI Skills: Guiding Students from Beginners to Masters

If students can generate a polished essay in seconds, a rubric that only grades the polished essay stops measuring learning. This is where the angle of this whole article comes together. Assessment has to move towards synthesis, evaluation and reflection, and it has to look at process as well as product.

Holt's practical advice points in two directions. First, shift homework towards flipped learning or reflective evaluation, so students cannot bypass the cognitive effort with a chatbot. Students meet new content at home, perhaps through a video or reading, and do the demanding thinking in class where you can see it. Second, design rubrics that assess the student's process, their reflection, and their critique of AI outputs, not just the final generated product.

It helps to think of AI capability as a progression that students move through:

A sample rubric for a Green-light essay might weight its criteria like this: argument and synthesis, 30 percent; critique of the AI draft, 25 percent; verification of sources, 20 percent; reflection on process, 15 percent; transparent attribution, 10 percent. The exact numbers matter less than the message they send. A student who submits a beautiful AI paragraph with no critique, no verification and no reflection should not score well, because they have not shown learning.

Reflection prompts do much of the heavy lifting. Try questions such as: Where did the AI get something wrong, and how did you know? What did you decide to keep, and why? What do you understand now that you did not before? If you are designing these tasks regularly, our guide to using an AI assignment generator to build assignments that teach shows how to keep the thinking with the student.

Remember the Brookings finding. Practice accuracy rose while test scores fell. Process-focused assessment is how you stop that pattern from happening in your room.

Why the Human Teacher Remains Irreplaceable in an Automated Era

Every conversation about AI eventually reaches the quiet worry underneath it. Will this replace me? Holt answers clearly, and she grounds the answer in research rather than reassurance.

"The number one in every single study, meta-study that you're ever going to find, is the relationship with the teacher. Everything else is just tools.", Stephanie Holt

Student growth rests on emotional connection and trust built with a teacher. A chatbot can explain photosynthesis. It cannot notice that a usually confident student has gone silent this week. It cannot decide that today the class needs a story more than a worksheet. It cannot make a nervous learner believe they can do hard things. That is the work, and it is yours.

Holt goes further. She argues that AI makes teachers more important, not less, and that students want their teachers: they want to connect with you and to make you proud. Seen that way, every hour AI saves on a scheme of work is an hour you can return to the relationship that drives learning in the first place.

She also offers a better benchmark for success. The goal of AI integration is not simply higher grades, welcome as those are. The goal is producing better, deeper thinkers. If your students leave your classroom able to question a fluent answer, verify a claim and explain their own reasoning, you have prepared them for the intelligence age far better than any tool could.

So where do you start? Pick one thing this week:

• Label your next three tasks Red, Amber or Green.
• Add a disclosure line to the next slide you build with AI.
• Write one fully structured prompt and save it as the first entry in your prompt library.
• Add a reflection question to your next assignment rubric.

None of this asks you to become a technologist. It asks you to do what good teachers have always done: decide what learning looks like, then choose the tools that serve it. For the broader picture of how AI is reshaping the profession, read our guide to AI in education and what teachers should do about it, and if you are planning your professional growth for the year, our guide to CPD for teachers will help you build a path rather than a pile of certificates.

You were asked to shape the future. The intelligence age makes that work harder and more important at the same time. You do not have to figure it out alone. At Suraasa, we build the system that starts with the teacher and stays with them, including the Professional Graduate Certificate in Teaching and Learning (PgCTL) for teachers who want structured, career-long growth.

The tools will keep changing. The relationship at the centre of your classroom will not.

For the Love of Teaching.