Alpna Bhatia is an Associate Teaching Professor of Economics at the University of Colorado Boulder.

When people ask how I use AI in my Principles of Macroeconomics course, I usually start with what I don’t use it for. I don’t use it to replace reading, skip over data, or outsource thinking. I use it, ironically, to slow students down—and to make their economic reasoning visible, both to me and to themselves.

Teaching students to navigate the messiness of real-world economics doesn’t happen overnight. It requires a scaffolded approach that shifts the challenge of thinking critically back onto the student so they can learn reasoning and relate theory back to their lives. The real pedagogical magic happens when we stop viewing AI as a sophisticated plagiarism machine and start framing it as a “reasoning partner.”

If you want to transition students away from using AI as a crutch, you have to engineer the interaction for them. In The Best of the Ultimate Guide to Teaching Economics, which accompanies Mateer and Coppock’s Principles of Economics series textbooks, I outline several exercises designed to do exactly this. Depending on your learning objectives, you can deploy AI to support everything from foundational concept mastery to advanced critical analysis. Here are two distinct ways I structure that progression, from foundational concepts to complex data:

Early Phase: Using AI to Build Foundational Logic

When students are just grasping the basics, AI makes a fantastic Socratic tutor—but only if you give them the right prompt. A great example of this is the “AI vs. You—Sorting GDP Components” exercise.

I give students a highly specific prompt instructing the AI tool to act as a guide: “Ask me one question at a time… Don’t classify the activity right away—instead, help me arrive at the correct answer through reasoning.” Students then compare their own classifications against the AI’s. This not only deepens their understanding of macroeconomic indicators, but it introduces the crucial digital literacy skill of evaluating AI-generated content for nuances and accuracy.

Critical Thinking Phase: Tackling Messy Data and Trade-offs

Once students have learned the foundations of macroeconomics and how we use AI in our course, I try to setup AI to help simulate complex decision-making using messy economic data. To see what this looks like in practice, I highly recommend the “Fed Oracle” exercise. This exercise guides students through a monetary policy simulation using an AI tool.

First, students prepare by reviewing the Fed’s dual mandate and selecting an economic scenario. They can either pull live, messy data from FRED or choose one of the simulated challenges—like the dreaded combination of 5% inflation and 6.5% unemployment. Next, they feed a highly structured prompt to the LLM (Large Language Model), casting it as a “Monetary Policy Oracle.” When they provide their scenario, AI proposes a policy action, explains its reasoning, and offers an underlying economic principle. The best pedagogical value of course happens after they step away from AI. Using the Oracle’s output as a starting point, students must conduct independent research to find a historical instance where the Fed faced similar conditions. They compare real-world history against AI’s advice. This forces them to navigate the inherent trade-offs of monetary policy, building upon the core information from the textbook and lectures.

Engineering Metacognition

AI requires us to be exceptionally intentional about student engagement. In a classroom, I can see when a student is confused; online, I rely on the artifacts they produce. I require students to share their AI conversation logs alongside a short reflection so I can see exactly how they reasoned through a problem. In the reflection, I ask students to articulate exactly why using the LLM in this highly structured way helped or hindered them. I ask them to document whether they caught AI making errors or exhibiting sycophantic behavior—did the LLM eagerly agree with a flawed economic premise just because the student suggested it? Hunting for these AI flaws introduces a crucial digital literacy skill: evaluating generated content for nuances and accuracy.

Of course, a cornerstone of inclusive pedagogy is student autonomy. I always provide an opt-out option. Students can choose to complete the assignment entirely without AI, or they can do the work first and use AI solely as a post-assignment check. This isn’t just an administrative workaround; it is an intentional choice that gives students ownership over the cognitive tools they use.

Teaching the Limits, Not Just the Possibilities

Just as important as how we use AI is talking explicitly about when it helps—and when it doesn’t. AI can clarify terminology and help students test their assumptions. But AI cannot tell us causation from correlation. It cannot evaluate trade-offs or weigh competing values. Those judgments—the ones that matter most in economics—still belong to the student.

Framing AI this way isn’t just a pedagogical choice—it mirrors how economics actually works. The economy is noisy, evidence is imperfect, and good analysis lives in the gray areas between data and interpretation. AI becomes one more lens for examining that messiness, not a shortcut around it.

Find out more about the Principles of Economics series and the Ultimate Guide here.

MEET THE AUTHOR