A few months ago, we spent part of a Saturday morning with a group of middle school students at Virginia Tech’s “Big Ideas in the Humanities” camp, a weekend program sponsored by the Center for Rural Education that invites gifted students from rural areas to explore how humanities questions help us make sense of life. Our lesson focused on local history during World War II, and we wanted students to look specifically at a photograph connected to wartime fat collection efforts in Roanoke, Virginia. We projected the image onto the screen and asked students to ask and work with questions using the Question Formulation Technique (QFT).
Figure 1
Manpower, junior size, by Valentino Sarra, 1942 Oct.
While looking at the image, students offered a few factual questions:
- How old are they?
- Where are they?
- What time period is this?
- Is this a school or food charity?
- Where are their parents?
Figure 2
Middle School Students Engage in the QFT
Then we added a twist.
After students generated their own questions, we introduced Chatty Geeps (our nickname for ChatGPT) and asked it to generate questions about the same image. We framed the activity as a chance to examine what this tool could and could not do. Before we compared the lists, we had also spent a little time demystifying the technology: Chatty Geeps is not a historian, a magical oracle, or a better student. It is a pattern-making system that can produce plausible-sounding text, sometimes helpfully and sometimes very badly. Naming those limits gave students permission to be skeptical rather than intimidated.
Some of the AI-generated questions were useful and pushed students to think of more complex ideas and move beyond factual questions. Students added some of the AI questions to their own lists and also used them as a springboard to develop more of their own. But they also noticed that some of Chatty Geep’s responses misread or entirely overlooked important details in the image.
Perhaps more importantly, however, when students prioritized the questions they wanted to investigate further, they overwhelmingly chose their own.
Right now, much of the public conversation about generative AI (GenAI) in schools tends to swing between extremes. GenAI is either framed as an inevitable revolution that will transform education for the better, or as an existential threat to thinking itself. As history educators, we find ourselves somewhere in the middle, taking a stance we are calling technocuriosity.
In our role as educators, we do not think the answer is pretending these tools do not exist. Students are already encountering them. Teachers are already trying to make sense of them. The question we need to answer is not necessarily whether GenAI belongs in classrooms—we may have already lost that battle—but what kinds of classroom practices might help students engage with it thoughtfully and critically. That is part of why we became interested in pairing GenAI with the QFT. The QFT creates structure around curiosity. Because it begins with noticing, wondering, and struggling to frame meaningful questions, it slows students down long enough to notice their own thinking.
What surprised us during our work with students and teachers was that introducing AI into the QFT process did not necessarily diminish student thinking. In some cases, it actually made student thinking more visible. When students compared their own questions with AI-generated ones, they began evaluating the quality of inquiry itself:
- Which questions felt more meaningful?
- Which questions led toward deeper investigation?
- Which questions assumed too much?
- Which questions sounded impressive but did not help us think?
In other words, this unusual pairing helped students develop stronger historical questions while also developing judgment about questioning itself.
At the same time, we also do not want AI to flatten inquiry into answer-seeking alone. One of our concerns is that GenAI can encourage students to move too quickly toward conclusions before they have had time to sit with ambiguity, wrestle with evidence, or formulate questions worth pursuing. Pairing AI with the QFT offers one possible way to protect space for student thinking while still allowing room for experimentation and exploration.
Note: We are still figuring this out! Some things work well. Some things go very poorly. Sometimes the same thing that goes well on Monday is a nightmare on Thursday. Sometimes students become more reflective about their own thinking because of the comparison. Sometimes AI becomes a distraction. But we think those tensions are worth exploring rather than avoiding altogether.
In the webinar below, we walk participants through the same process we used with students and teachers. We experiment with pairing the QFT and GenAI in real time, compare human and AI-generated questions, discuss what happens when the technology gets things wrong, and reflect on what inquiry might look like in this strange, rapidly changing moment. We are not approaching this webinar as experts with definitive answers; we are history educators, thinking carefully about how disciplinary practices can remain at the center of learning even as new technologies reshape the landscape around us. Click here to access the recording that includes the chatbox transcript.
This webinar builds on ideas from our forthcoming practitioner book on teaching history with GenAI, where we explore ways history teachers can experiment with AI while keeping inquiry, evidence, and student thinking at the center of the classroom. Sign up to be notified when the book is released here.
Mini Q&A with David and Amy:
Below, we respond to a few of the questions that came up during the webinar.
You mentioned that you used a custom version of ChatGPT for the lesson described above. What is a custom GPT, and how did you create one for the social studies classroom?
A custom GPT is a version of ChatGPT that has been given a more specific purpose, set of instructions, and, in some cases, a small collection of supporting materials. We created ours as a classroom-facing thought partner for historical inquiry. That meant giving it instructions about the QFT, historical thinking, age-appropriate language, and the kind of pedagogical stance that would allow it to generate questions, make its reasoning visible enough to critique, and still occasionally fail.
A custom GPT is not perfect or “trained” in the way people sometimes imagine. It is still limited, and it still needs teacher judgment. The benefit is that teachers can narrow the task, shape the tone, and build a tool that better reflects their classroom goals. For teachers who want to try this, we recommend beginning by identifying one tightly bounded purpose, such as helping students generate historical questions from a primary source, and then revising the instructions after using it with actual student work. In a recent article in Social Education, we outline step by step instructions on how to build a custom GPT.
Students already have a tendency to judge or feel insecure about their own questions. How do you avoid students concluding that the bot knows more than them or asks better questions?
This was one of our concerns as well, so we introduced Chatty Geeps as a tool, not an authority. We explained, in student-friendly language, how it predicts and produces text based on patterns. It does not see the image the way a person sees it. It does not know what students notice, what they are curious about, or what matters to them as learners. It can sound confident even when it is wrong.
That framing changed the way we interacted with Chatty Geeps, asking questions like: Did the question fit the evidence? Did it assume something not visible in the image? Did it help us investigate the past, or did it just sound fancy? Once students saw that Chatty Geeps could miss obvious details, the comparison became less intimidating and more empowering. Additionally, we were careful not to treat their questions as rough drafts that AI would improve but as serious attempts at historical thinking that were valuable on their own.
What are your top three tips for using ChatGPT in the classroom in a way that preserves student thinking?
As AI becomes increasingly embedded in the tools students use every day, we think less about “using ChatGPT” and more about helping students learn how to think in an AI-saturated world. Whether students encounter AI through ChatGPT, Google Search, or tools we have not yet imagined, they will need opportunities to develop judgment about how and when to engage with these systems.
First, have students do the intellectual work before the tool enters the process. In our example, students generated their own questions before seeing anything from ChatGPT. That sequence protects student curiosity. They need opportunities to notice, wonder, puzzle, and struggle before a polished response appears on the screen. If AI becomes the starting point, students can easily shift from inquiry to reaction.
Second, design activities that require students to monitor and evaluate thinking, not simply consume outputs. One of the reasons we paired ChatGPT with the Question Formulation Technique is that it creates opportunities for comparison and reflection. Students are deciding which questions are stronger, which are weaker, which assumptions are embedded within them, and which are worth pursuing. Students learn through processing; what gets processed gets learned. The more opportunities students have to interrogate, revise, and justify their thinking, the more likely meaningful learning becomes.
Third, use the tool’s mistakes as part of the lesson. We are often tempted to hide the messy parts of technology, but the errors are where some of the best teaching happens (for more on this idea, see our recent post in The Times Higher Education). When ChatGPT misreads an image or invents context, students get a concrete example of why human judgment still matters. Learning requires sustained and increasingly sophisticated work. Students will always begin with approximations, and comparing stronger and weaker questions, whether generated by humans or AI, can help nurture their capacity to ask better ones over time.
Underlying all three suggestions is the importance of making the learning visible. Students should understand why they are generating their own questions and why they are evaluating the tool’s responses. As part of this process, students will always begin with approximations of sophisticated thinking. In our minds, the goal is to help them become increasingly aware of how they think, how they learn, and how they can monitor and improve their own inquiry over time. When students understand that purpose, the technology becomes one resource within the learning process rather than the focus of the lesson. Beyond this, we feel strongly that if AI is becoming part of the learning environment, then students need repeated opportunities to decide when to use it, when to question it, and when to ignore it, alongside guidance that they should always trust their own judgment.
Does this help students learn? Having students critique the questions can have value, but does the value added justify the time taken? Is the benefit more about historical literacy or artificial intelligence literacy?
The question of whether the activity is worth the time is an important one because every instructional choice carries an opportunity cost. If students spend twenty minutes analyzing and refining questions, they are not spending those same twenty minutes doing something else. Teachers make these decisions every day.
For us, the more important question is not whether an activity takes time, but what students are doing with that time. Learning does not happen simply because students encounter information. Learning happens when students actively process ideas, wrestle with them, revise them, and make judgments about them. Again, what gets processed gets learned. That is why we see value in comparing student-generated and AI-generated questions. Students are producing questions, but also evaluating them, identifying strengths and weaknesses, considering different forms of inquiry, revising their thinking, and deciding which questions are worth pursuing. The intellectual work lies in the comparison, reflection, and discussion that follow.
We are also not entirely convinced that historical literacy and AI literacy are as separate as this question suggests. While they emerge from different traditions, they share many of the same habits of inquiry. Historians ask where information comes from, what evidence supports a claim, what perspectives are missing, how context shapes interpretation, and whose interests are served by particular representations of knowledge. Increasingly, those are the same questions students need to ask when working with generative AI.
From this perspective, students are not choosing between learning historical literacy or AI literacy. They are practicing forms of judgment that matter in both domains. They are learning to question claims, examine evidence, identify assumptions, recognize omissions, and make informed decisions about what deserves their attention and trust.
Of course, not every use of AI justifies the time required. If students simply generate questions and move on, the activity offers little value. But if the comparison becomes an avenue for reflection, critique, and revision, then the time is being spent cultivating habits of inquiry that extend far beyond a single lesson.
About the authors:
Amy Allen, Ph.D., is an Assistant Professor of History and Social Science Education in the Elementary Education program at Virginia Tech. Her research examines how teachers and students navigate history, technology, and inquiry in the age of GenAI. She explores how digital tools can support (not replace) the messy, human work of meaning-making. Her scholarship appears in Theory & Research in Social Education, The Social Studies, and Social Studies and the Young Learner, and she frequently presents at NCSS, NCHE, and AERA. Outside of work, Amy maintains a deeply analog relationship with karaoke, trivia, and chaotic good energy. When she’s not researching technocuriosity, she’s testing the limits of “place-based learning” at Disney parks, listening to Taylor Swift, and pretending she’s in a musical number nobody asked for but everyone secretly enjoys. She is deeply committed to serious scholarship but never taking herself too seriously.
David Hicks, Ph.D., is a Professor of History and Social Science Education at Virginia Tech, where his research explores the integration of digital technologies and media to deepen historical inquiry and disciplinary thinking. With more than ninety publications, he has contributed significantly to understanding how technology can scaffold the analysis of historical sources, foster critical literacy, and humanize digital learning environments. Dr. Hicks’s scholarship combines historical rigor, technological innovation, and a commitment to helping teachers navigate the promises and perils of AI in history education. Outside of academia, David remains tragically unrecognized by the Academy, the Recording Industry, and, most cruelly, Mardi Gras king cakes. A self-proclaimed “Gen AI curious” historian, he divides his time between mentoring pre-service teachers, perfecting small animal puppetry, contemplating why Twitter’s rebrand destroyed civilization, and surfing TikTok. When not philosophizing about cognitive engineering, David dreams of a One Direction reunion and still insists Blur is better than Oasis.