The short version
- Most students who take Principles of Macroeconomics never take another economics course, so that one class should leave them able to find real data, explain it to someone who has to act on it, and judge what an AI tool gets wrong.
- FREDX builds that into eight short FRED assignments. Each one ends with a 100-word note to a future employer, and AI only comes in after the student commits to an answer.
- Among 93 students surveyed in Spring 2026, confidence rose most in the skills they started out least sure of: finding data, building charts, and explaining what a chart shows.
One shot at economics
If a student takes only one economics class, it should leave them able to find real data, explain what it shows to someone who needs to act on it, and judge what an AI tool gets wrong.
Most students who sit in Principles of Macroeconomics will never take another economics course (Stock, 2024). Many are business majors meeting a requirement. Yet they'll spend their careers reading jobs reports, arguing about inflation at the dinner table, and deciding whether this is the year to lock in a mortgage rate.
So I keep asking what that one class should actually equip them to do. My colleague Usamah Alfarhan and I wrote a paper on our answer, which we call FREDX. It's forthcoming in the Journal of Economics Teaching, and this week I'm presenting it at the St. Louis Fed Professors Conference. Below is what we built and what our students told us.
What's wrong with how we teach it now?
Two things pushed us to redesign the course.
The first is old. Thirty years after economists started complaining about "chalk and talk," lecture still dominates the intro course (Asarta, Ahlstrom, & Harter, 2026). Data shows up as a figure in the textbook. Students learn what GDP measures without ever downloading a GDP series and plotting it themselves. I call that data literacy by proxy, and I don't think it sticks.
The second is new. More than a third of college students already use AI tools regularly in their coursework (Stöhr et al., 2024). If we ignore that, students use it anyway, just quietly. If we bolt it on badly, it does the thinking we wanted them to do. I believe the answer is to design for AI on purpose, and to decide exactly where in the work it shows up.
What is FREDX?
FREDX is eight short data assignments and a final report, built into the regular order of a macro principles course. The week we cover inflation, students calculate inflation themselves from the Consumer Price Index and the Producer Price Index. The week we cover the Fed, they chart what the Fed actually did in 2008 and 2009.
Every assignment runs the same five steps, and AI comes last
- RetrievePull the series from FRED
- VisualizeBuild the chart in Excel
- ReasonTie the pattern to the model
- Communicate100 words to a future employer
- EvaluateAsk AI to critique it, then judge the critique
↺ Next assignment: same five steps, less support from the instructor
The steps stay the same all semester. What changes is how much I spell out. Early on, every data series is named and every chart is specified. By the end, students pick their own data to compare the 2008 financial crisis with the COVID-19 recession, and that report replaces the final exam.
Who are students writing for?
Every FREDX assignment ends the same way: explain what you found, in about 100 words, to a future employer.
That one framing changes what students write. An essay for a professor invites them to recite the definition so I know they read the chapter. A note to a manager makes them decide what a non-economist needs to know and what can be left out. In the first assignment, that means comparing long-run U.S. growth with the last five years and telling a boss whether the recent pace is normal.
The word limit matters too. A hundred words forces a claim, the evidence behind it, and a reason anyone should care. By the end of the semester, students have eight short reports and a capstone they can actually show an employer.
Robert Shiller made the case in his presidential address to the American Economic Association that the stories people tell about the economy move the economy (Shiller, 2017). If narratives matter that much, I think our students should practice building them from data, and checking them against it.
So where does AI fit?
AI comes last, after the student has already committed to an answer.
Students finish the analysis and their 100-word report on their own. Then they paste that report into any chatbot they like (the free versions of ChatGPT, Claude, Gemini and Copilot all work) and ask it, as an economic expert, to critically evaluate their reasoning. After that, they answer three questions about what came back:
- What did I learn that I didn't already know?
- Which of the AI's claims go beyond what my own data show?
- Did the exchange actually improve my understanding?
The order is the whole design. We borrowed it from Chin and colleagues, who argue that AI works best as a complement to student reasoning rather than a substitute for it (Chin et al., 2026). A student who starts with the chatbot tends to anchor on whatever it says. A student who commits first can see where the AI disagrees with them and ask why.
To be clear, I'm not saying students should try to catch the AI being wrong. Chatbots tend to agree with whoever is typing, so "looks good to me" is a common reply. The second question is where the learning happens. Even an agreeable reply can reach past the data, into causes or forecasts the student's chart can't support, and spotting that is the skill.
What we want from students is AI judgment, not just AI use.
Did students feel any different by the end?
They did, and the biggest jumps came where they started out least sure of themselves.
We surveyed students at the start and end of the Spring 2026 semester, and 93 of them answered both times. The questions asked how confident they felt finding economic data, building charts in Excel, explaining what a chart shows, and using AI, plus how much they valued those skills.
Skill confidence rose most where students started lowest
Average agreement before and after FREDX, 93 students who took both surveys, Spring 2026
1 = strongly disagree, 7 = strongly agree
What I find interesting is the gap between the top and bottom of that chart. Students walked in already confident about AI and already convinced data skills matter. What they didn't believe was that they could do the work: find the series, build the chart, explain it. Those are the scores that moved most. One item stood out to me: "I feel confident evaluating whether a graph or chart is misleading" went from 3.76 to 5.24.
I want to be honest about what this is. These are students rating their own confidence, there's no comparison group yet, and confidence isn't the same thing as skill. Every change was statistically significant except the last row. A proper comparison with sections that didn't use FREDX is the next study.
Can you use this in your class?
Yes, and you don't have to rebuild your course to do it. Everything runs on FRED, which is free, and Excel, which most campuses already license. The full set of eight assignments, the capstone, and the grading rubric are in the paper's appendix.
If you only take one piece, I'd take the bottom of every assignment: the 100-word note to a future employer, followed by the three AI questions. You can add both to something you already assign this week.
What it means for you
If you're a parent or a student reading this, try it on the next economic headline you see. Find the actual number on FRED, write down in two sentences what you think it means, then ask a chatbot to poke holes in it.
Pay attention to what it tells you that the data never said. That gap is where most bad economic arguments live, and you'll start spotting it everywhere.
For the classroom and your conversations
- Pull U.S. real GDP from FRED. Has growth over the last five years been faster or slower than the long-run average? What would you tell a boss about why?
- Ask a chatbot to critique your answer. Which of its claims could you check with the data you pulled, and which couldn't you?
- Why might someone feel very confident using AI but much less confident reading a chart? What would change that?
- If you could take only one economics class, what would you want to be able to do when it ended?
Read and listen more
- FRED, Federal Reserve Bank of St. Louis: the free data source behind every FREDX assignment
- Journal of Economics Teaching, where the FREDX paper is forthcoming
- Teach economics at any level? The Decode Econ Educator Network is a free, selected cohort that meets monthly. Applications for 2027 are due December 1, 2026, and the form is on the For Educators page.
References
Asarta, C. J., Ahlstrom, L. J., & Harter, C. L. (2026). "Chalk and talk" 30 years later: Insights from a seventh national survey on teaching undergraduate introductory economics. The American Economist, 71(1), 5–14. Find it on Google Scholar
Chin, S., Io, K., Kırşanlı, F., & Perry, T. (2026). From cheating to learning: AI integration in introductory economics. Journal of Economics Teaching, 11(2), 208–228. https://journalofeconomicsteaching.org/
Shiller, R. J. (2017). Narrative economics. American Economic Review, 107(4), 967–1004. https://doi.org/10.1257/aer.107.4.967
Stock, W. A. (2024). Who does (and does not) take introductory economics? The Journal of Economic Education, 55(2), 110–127. https://doi.org/10.1080/00220485.2023.2277768
Stöhr, C., Ou, A. W., & Malmström, H. (2024). Perceptions and usage of AI chatbots among students in higher education across genders, academic levels and fields of study. Computers and Education: Artificial Intelligence, 7, 100259. https://doi.org/10.1016/j.caeai.2024.100259
If you add the 100-word note or the three AI questions to one of your assignments, what did your students write, and what surprised you?