Introduction: AI in History Education
Each day, new reports on AI tools demonstrate their volume of technical capabilities that allow for instantaneous production of student work, including summarizing long readings, analyzing historical primary sources, adapting antiquated language, and writing largely passable historical essays that can pass reading comprehension exams (Hughes-Warrington 2023). AI is transforming the field of history by providing new tools for research, analysis, and accessibility at lightning speed; however, many educators remain understandably hesitant to integrate AI tools. Educators are at a historic moment for pedagogical change in how we teach. For historians, what does it mean to teach historical inquiry in this AI-saturated moment—especially in ways that still meet the expectations of the American Historical Association’s “Core Learning Outcomes in History” (2025)? While universities are rapidly developing their own policies, training guides, and courses on AI, the American Historical Association has released its own—encouraging history faculty to “responsibly and effectively incorporate generative AI into their teaching practice” (AHA Council 2025). If generative AI is radically shaping education, educators should gain their own “AI literacy” to understand the pros and cons of AI-integrated learning and how those shape teaching in their field.
In particular, AI is radically shaping the practice of digital public history that can serve as a constructive model to history students. AI technologies have shown to be particularly effective in making archival histories accessible, visible, and personally meaningful, including aiding in the transcription of oral history interviews. AI transcriptions have facilitated new partnerships between historical associations and corporations to aid in improving archival access, such as the Imperial War Museum’s recent projects using AI to extract metadata from more than one million primary sources in its archives (Coates 2025; Charr 2025). This new digital frontier can help bring to the forefront the role of history educators, the methods of historical inquiry, and the value of accessible historical data to the broader public in a political climate hostile to history and the broader humanities. Digital tools like AI, when taught and used through an ethical framework of digital literacy, can help history educators accomplish the larger goal of reinforcing the relevance of the past in a future-focused present.
Comparing pedagogical techniques using AI across four history courses (seven sections, with a total of over 100 students, ranging from surveys to upper-level seminars) taught between January 2024 and May 2025, data collected reveals how free to low-cost AI tools such as ChatGPT can be used tactically to assist in teaching basic principles of historical inquiry. While data is taken largely from Asia Campus history surveys, this data includes student work from a targeted independent seminar on the topic, “AI in the Archives,” which sought to measure how minimalist AI software could be used as a method of teaching and learning history, specifically focused on increasing accessibility and usage of nontraditional primary sources. An analysis of student coursework and reflections across campuses and courses revealed that minimalist AI pedagogy can aid instructors in empowering students to more deeply, creatively, and critically engage with primary sources even when faculty operate within technical and institutional constraints. When used effectively—and especially when approached through a framework of change and growth for both faculty and students—AI can assist history faculty in teaching digital historical methods in ways that facilitate critical inquiry about difference, power, and patterns over time.
In assisting students with introductory research and evidence-based analysis, AI tools proved helpful in these courses in guiding students through principles of historical inquiry while also demonstrating opportunities for students to advance their research. By creating an isolated space free from judgment, AI tools supported students in using historical narratives to explore complex topics and build empathy across generations, geographical borders, and identities. At-your-own-pace AI tasks alleviated some pressure off of faculty to work one-on-one with students in converting subjective analyses into skill-specific checklists. However, AI did not surpass faculty guidance in any area of historical inquiry, only in reinforcing the synthesis of course material with AI literacy. AI literacy approaches to teaching history helped clarify what artificial intelligence can and cannot do, reinforcing software as a tool rather than a replacement for historical inquiry while also underpinning student and educator accountability. Taken together, one of the most constructive aspects of using AI in teaching historical methods is the ability to teach students and faculty to hone their communication skills. By using large language model (LLM) guidance techniques, such as prompt engineering (coaching the AI to produce a better output by revising your prompt) and grounding (giving a GPT verifiable sources as models), students can more clearly recognize the software’s limitations and, therefore, what humans can uniquely bring to the work of history.
Context: Fear, Unease, and Digital Illiteracy
The first hurdle for the newly developed course, “AI in the Archives,” was neither cost nor faculty derision but student unease. When compared globally, the United States, despite being the current leader in AI development, boasts the world’s worst score in overall public perception of artificial intelligence (Poushter et al. 2025). Assignments that regularly solicited student feedback as critical reflections on AI illuminated that Salt Lake City Campus students in advanced history seminars were much more reluctant to use AI software, especially when instructed to independently utilize AI tools that best match their research topic. Comments expressed by students at the Salt Lake City Campus reiterated claims concluded in a recent Gallup Survey of Gen Zers who expressed both anxiety and even anger about artificial intelligence, unpreparedness in its usage during post-graduation due to school bans, and fears that AI usage will harm their understanding of digital literacy (Walton Family Foundation 2025). Even when upper-level students enrolled in the AI-specific seminar were provided a substantial foundation in digital history, AI functionality and terminology, and when early assignments were structured with a palette of capabilities, a majority of students still denigrated using AI tools in the Spring of 2024. Little instruction is being provided to students on the definitions and effective use of AI, leading to a broader climate of fear and distrust heightened in an era of “fake news.” However, engaging with new technologies requires time, patience, and curiosity, which can be a challenge to muster in our contemporary political climate. Problems with using generative AI heighten extant concerns educators and students have had in education, including: privacy and security, bias, reduced human interaction, academic misconduct, and information inaccuracies (Office of Communications 2024).
The foundation of this project’s exploration of AI in teaching history rested on testing if/how free to low-cost AI tools could be used to teach key analytical concepts of digital information literacy—what the Society for American Archivists outlines as a combination of using primary sources to teach students about contextualization, evidence versus opinion, ethics of access and privacy, as well as the power dynamics shaping both the source and its collection. Rather than relying on AI as a replacement teacher (in providing instantaneous lectures) or as a replacement student (in producing instantaneous response papers), could AI be used as a tool to guide students through digitized materials to help faculty teach historical inquiry? In particular, the focus of the project rested on ways in which AI could add to the course’s implementation of higher order critical thinking skills in Bloom’s taxonomy using “minimalist” technology—software requiring only a basic browser, at no cost to faculty or the students, and needing little tutorial while still maintaining the course’s learning objectives. Through this lens, the minimalist component of this project meant the highest potential for accessibility with the lowest barriers to entry, in which students could, relatively quickly, find the technology, be able to use the tool freely, and engage with the tool enough to allow for a critical analysis. More importantly, the goal was to utilize easily accessible AI tools to more quickly, comprehensively, and creatively engage with historic primary sources as a tool of information literacy.
Course Design and Tactical Assignments
Teaching the capabilities of AI for historical inquiry first required developing a protocol on AI literacy that best fits your discipline and course. As part of this AI literacy, your assignments should clearly outline to your students the limitations (not merely the opportunities) of using AI for historical inquiry. Just like with free internet use at a cafe, LLM-based chat systems available for free might have distracting ads or prompts for subscriptions so students should be cognizant about what they should click on and never enter in credit card information or link to a social media account for this schoolwork unless of their own volition. Students were encouraged to use their own accessible spam email account separate from their school emails as preparation for the frequent hassle of required accounts. AI requires active engagement and should never be used as the sole source of information. Teaching students how ChatGPT works can help them understand that GPTs do not find and retrieve stored information but instead scour pre-programmed datasets to rely on probabilities to generate text. Prompts—instructions communicated to a GPT—must be clear and precise. Reviewing GPT output and revising prompts requires active engagement. Because a foundation of topic is knowledge necessary for historical inquiry, students were instructed early in the semester on academic pathways for research using the library catalog. While most AI tasks focused on using AIs for source analysis, citation formatting, summarization, and creative generation, assignments that required students to use ChatGPT for basic research equally instructed students on verifying the sources it provides.
Most importantly, demonstrate to your students how the AI can hallucinate or quickly generate mistakes based on predictions used from its pre-training. Intentionally prompting an LLM like ChatGPT to generate an output that is incorrect, misleading, or nonsensical despite being presented as factual can help students understand the software’s limitations. For example, if you ask ChatGPT, “Can you tell me about the great war between the dragons and fairies in 1732?” it will likely respond with what is called semantic proximity by providing information about the subjects and time period that it has been trained on without rejecting the question outright. As evidenced by recent reports, the percentage of hallucinations that ChatGPT produces is not reducing despite updates to the LLM. The more you use a computer, the more opportunities it has to malfunction, and because AI does not “think” like a human, it generates text by predicting what words or phrases are likely to appear together based on patterns in large quantities of data. While GPTs are designed to counterfactually improvise to continue a chat when it does not know the answer, you can give them instructions to say “no” if they do not know the answer, and they will do so.
Students (and faculty alike) are shocked that an AI might lie, however teaching these concepts can help students see their own techniques for teaching and learning. A test taker might similarly use semantic proximity when guessing the answer to an unknown question in an effort to hopefully score higher. By framing the LLM’s drive as one centered on continuing the conversation and comparing its responses to guesses generated by a student on a sudden pop quiz, students can more quickly think of AI as a peer or student rather than an omnipotent being or wolf in sheep’s clothing. Just as test questions must be specific, clear, and relevant to the topic, so must prompts to GPTs to ensure the most accurate output.
Students should also be instructed on ChatGPT’s gaps and how to understand and correct for them through machine learning concepts like prompt engineering (coaching the AI to produce a better output by revising your prompt), grounding (giving a GPT verifiable sources as models) and tokenization (dividing text into smaller units called tokens—a core process in machine learning, that also resonates with how historians examine language in fragments). By using LLM-specific terminology, students can better understand how their prompts can be revised for more accurate output. Instructors can model to students how to ground a GPT by providing an example source, just as they would for a student. One method suggested by Google Gemini that uses Google search engine results as its evidence is retrieval augmented generation. This approach involves feeding an AI tool with real information—such as a document or a website—so that the GPT’s answers are based on facts you’ve provided, rather than just its predictions from unknown datasets. Not only is this exercise helpful in demonstrating the tool’s boundaries, but also teaches students key components of information literacy, including evaluating bias, analyzing evidence, and using information ethically in their work as historians. Through output critique, students became mini teachers to the ChatGPT in analyzing its thought process.
The more an instructor can use AI themselves to understand its own limitations and capabilities, the more effective they will be at designing clear, precise, and discipline-appropriate pathways to AI literacy. Every week, more and more scholars and practitioners release tutorials and reports on their discipline-specific use of AI that can help further train faculty in effective strategies for critical digital pedagogy. Faculty should strive to establish a solid foundation of knowledge in AI tools relevant to their field. While the amount of data on AI can seem overwhelming at first, this foundational knowledge can help alleviate concerns and help faculty anticipate potential issues. With my focus on using AI for historical inquiry into oral history transcripts in my “AI in the Archives” course, I learned through webinars from crowdsourcing consultants From The Page (2023) that ChatGPT’s own ethical safeguards and datasets currently prevent it from conducting a balanced rhetorical analysis of pro and anti-abolition documents. In my own experiments using “runaway slave” ads as a basis for humanizing resistance efforts of freedom seekers, ChatGPT can be quick to flag and even censor derogatory racial language embedded within historic sources. It is helpful to demonstrate to students how, even within well-known LLMs, you can find specific GPTs that could be more helpful to a specific task, such as Benjamin Breen’s GPT The Historian’s Friend trained for historical source analysis and prohibits creative output.
Key strategies for designing an AI task
Rather than leaving students to work with AI on their own, I approached their use of AI similarly to other technological tools, like a university library’s search engine or an interactive digital museum. In particular, attention was directed to developing an AI-guided history curriculum through new “AI Tasks” that used AI tools to explore archival sources. Modeled after the National History Education Clearinghouse’s “inquiry lessons” (National History Education 2025) and UCLA’s Public History Initiative’s Historical Thinking Standards (UCLA 2025), the AI tasks used in this study largely focused on contextualizing sources, prompting students for data collection, aiding in source retrieval, pushing analysis, and providing moments of critical reflection for nonmajors new to the discipline. Each AI Task guided students in strengthening a specific historical research skill, analysis, or communication method while also providing information on a specific era.
Reviewing student AI tasks taught me how to revise assignments to account for technological limitations or advancements and a lack of instructional clarity, as well as where to push critical reflection. Inquiry lessons like my AI Task 3: Adapting Text (Lovell 2025a) are designed to expose students to various perspectives and even evidence that contradicts their own understanding or viewpoint. With a majority of ESL students in my US history surveys at the University of Utah’s Korean campus, GPTs that allow for intralingual translation enable students to update older primary sources to modern English and other languages. AI tasks designed on analysis are paired with directed moments of self-reflection, requiring students to analyze how new information and evidence shaped their understanding. These moments reaffirm to students that historical inquiry is not focused on one correct answer but rather a medium through which to analyze evidence-based arguments.
In particular, AI tasks function as learning guides that are organized around several intertwined phases that seek to streamline aspects of already-designed assignments:
- laying a knowledge foundation—such as learning content-specific key terms like “redlining” when learning about segregation, or historiographical basics like basic components of an annotated bibliography
- self-guided exploration within set parameters—such as perusing subjects in a specified digital archive
- document-based critical analysis—analyzing the causes, effects, and contexts for subjects and sources to push deeper learning, usually coupled with more discussion-based questions
- augmented multimodal learning—steps in which additional information, terms, or sources are introduced to build on knowledge foundations while laying the groundwork for synthesis—very helpful in challenging students’ AI dependence
- personal reflection—moments for pause that ask students to personally react to the material without restraint, usually requiring a sentence or word count rather than editing to encourage extemporaneous thoughts
- grading rubric/grading checklist—a clear list of the assignment expectations that include content and formatting specific requirements, underscore the course’s learning objectives and the discipline’s inquiry methods
To begin the course, I use a simple AI task, like the linked example, that requires students to use an LLM to analyze the syllabus and respond to the prompts via a series of questions. Requiring the students to copy the ChatGPT URL into their submission allows me to review their LLM engagements. Just as we trust math students to “show their work,” AI tools like ChatGPT where URLs of conversations with the app can be shared can help students demonstrate their processes of learning. Through shared URLs of student-led conversations with ChatGPT, I was able to observe exactly when and how students were interrogating a primary source, identified moments when students were missing key components of historical inquiry, and pinpointed instances where students lack confidence in conducting evidence-based, argumentative analysis of historical artifacts and narratives. In the case of this AI task, seeing their work was especially important for me to see what types of questions students have about the course, how those questions were answered, and how students wrote and revised their prompts. Finally, while instructors might be more hesitant to include moments for reflection due to limited course time, reflective interludes can help students process their relationship with the tools that can strengthen their understanding of the content and history more broadly. As an instructor, the most commonly revised aspect of AI tasks involves honing these moments for critical reflection to ensure that students are genuinely engaging with the material.
Initially, AI tasks are as explicit as possible, with clear instructions on where to find a required document, how to use a tool, and how to structure their response. While initial AI tasks in history surveys focus more on lower-order critical thinking skills, such as learning terms and gathering facts, as the courses progressed AI tasks develop into more detailed forty-five to sixty minute assignments. For a history survey, topics of AI tasks follow the course content chronologically and are, therefore, designed around what is available on that particular topic as well as what methods of analysis they need to learn. When students engage with primary sources from early colonial settlers in the 1600s, written in outdated English, AI tasks are a helpful way to teach students about intralingual adaptation, or how to translate text within the same language. As part of a set of mid-semester AI tasks with topics interrogating slavery, assignments explored that topic through the lens of guiding students in how to transcribe 19th century-era newspaper print and turn-of-the-century handwriting—skills particularly hard for non-Native English speakers. Taken together, all AI tasks at their core focus on primary source literacy and digital literacy to build from basic course content to larger papers and the final projects that require application. Choose skill sets and topics that are the best fit for your discipline, course, and student level while also keeping in mind that you might need to test assignments three to four times before feeling confident, especially as technology rapidly shifts.
In the AI Task for Week 11 of the history survey, for example, students are asked to choose an image from the KUED Topaz Resident Collection within the university’s digital archives to learn about World War II-era Japanese American internment in Utah. After working with a peer to conduct a lightning round of photo analysis based on the instructional guide offered by the National Archives, students are instructed to run the image through ChatGPT for a detailed visual analysis. Students are guided on the GPT’s methods of visual reasoning or the ability of artificial intelligence to interpret and understand visual information by analyzing images (artwork, photographs, material culture, etc.) to generate conclusions. Because machine learning doesn’t “think,” it identifies patterns, AI can sometimes reveal unparalleled capabilities in deciphering fine details and comparing them with archived information on weaponry, fashion, architecture, farming tools, and more at an accelerated pace, enabling expert predictions about the visual content. GPTs, therefore, rely on digital archives as datasets. Because World War II-era photography is digitized at a much higher rate than Reconstruction-era photography, for example, and those archives often include sequential images due to advancements in camera technology, using GPT skills in visual reasoning for photographic analysis can prove very effective for the right topic. In this case, students with little knowledge of internment could engage with ChatGPT to learn about aspects of the design of internment camps (including guard towers, barrack organization, and barbed wire fencing), social practices (from community-building activities like youth classes to humiliating lines for the bathroom), and more. The AI task not only offers a model for visual analysis that students can apply on to later assignments, but also presents students with opportunities to ground GPTs with metadata from the digital archive.
First-semester survey students are quick to identify objects within a picture and often describe photos as reflecting reality rather than an artist’s construction through the lens of a photographer, perhaps because of the inundation of immediate digital photography in their daily lives. With this mindset, students often struggle with semiotic analysis, which is particularly useful as a tool of visual rhetorical analysis when examining war time government-censored media. As argued by historian Linda Gordon among others, the US military censored the release of Dorothea Lange’s photographs capturing the process of Japanese American incarceration during World War II because of the semiotics embedded within her photos (Gordon 2017). In guiding students on how to conduct a semiotic analysis using ChatGPT, students can see examples of how discourses of patriotism, order, power, paternalism, and assimilation were key to expressing resistance in photography of Japanese American internment.
Just as when designing your own inquiry lessons, AI tasks require time to generate and should be customized to your content, goals, and student level. Several history courses incorporated more in-depth AI tasks that encouraged students to use ChatGPT for its analytical skills as a model for their own analysis. Because of ChatGPT’s capabilities in quickly analyzing and summarizing long-form documents, such as 100+ page oral history transcripts, the tool became a way for the course to utilize oral history transcripts within the University of Utah’s Special Collections, as well as situate local Utah experiences within larger narratives about power and identity in American history. With a focus on humanizing and personalizing geographically, culturally, and chronologically different experiences than those of our Asia Campus students, US history survey students each selected a different interview within our university’s Carbon County Oral Histories collection on local immigrant coal miners and mining strikers, and used ChatGPT to generate brief biographical descriptions and portraits that they used to create diary entries from the perspective of their person in an effort to understand how locals employed resistance strategies against coal corporations.
In a higher-level oral history course, students use ChatGPT to create detailed summaries of assigned people within the Interviews with Japanese Americans in Utah Collection as the basis for metadata creation and visual concept maps. While the AI tool was used to guide students in what to do with primary source material throughout the process of historical inquiry, the tool became a way for students to learn more about Japanese American histories in Utah beyond internment—their work as sugar beet farmers, their experiences at church, their lives as stay-at-home mothers, etc.—in ways that challenged the myth of Asian Americans as perpetual foreigners in the American heartland. This AI task not only introduced them to local experiences but logistically helped prepare them for their final projects in which they were instructed to create zines summarizing, analyzing, and teaching long-form oral history interviews they conducted through interactive exercises. Because ChatGPT functions quite effectively when working in a “sandbox,” AI tasks that require students to utilize contextual question answering and source-grounded generation focused on a specific uploaded document can often serve as helpful models for introductory analytical lessons for survey students.
Scaffolding AI tasks for content, technology, and emotional understanding
Organizing assignments so that they build on one another technologically, topically, and emotionally is essential to ensuring that students are on track. In the history survey, assignments were scaffolded with ChatGPT allowing students to learn about its capabilities as they progressed through the historical research process for their final project, while also gaining a deeper understanding of the software’s capabilities. Introductory AI tasks guided students on how to summarize, outline, and find specific evidence in specific chapters within a PDF of a US history survey textbook (Foner, Give Me Liberty). Mid-to-late semester AI tasks are often more complex. For example, AI Task 10 introduces students to data mining oral history interviews on the Great Depression (Lovell 2025b). Students are provided with a set of oral history transcripts featuring poor southerners during the Great Depression, in which they use ChatGPT for information retrieval and sentiment analysis. With that collected data, students then use The Historian’s Friend to generate a prompt that can be used to generate a portrait of that person on ChatGPT.
Portrait generation using archival data can be very helpful for helping students personalize text data. Because photographs by Dorothea Lange and other Works Progress Administration photographers are well-documented online, students can easily use creative tools to generate (and prompt engineering) portraits of working-class people for whom there is no known digitized photograph. On the instructor side of viewing ChatGPT URLs to monitor student work, image generators are quite limited in their ability to generate historic photography based on an archival data prompt. They do, however, allow students the ability to “time travel,” enabling largely Korean history survey students at the Asia Campus fun and creative ways to see representations of themselves—thereby metaphorically centering Asians—in white-dominated era American history such as the colonial era.
One of ChatGPT’s strengths is that it can serve as a (seemingly) neutral entity when discussing issues of power and identity—bearing the brunt of student politics while also serving as a counterpoint if instructed to. To draw attention to our own biases, students in my History of American Social Movements course were assigned an AI Task that had them use ChatGPT’s sentiment analysis capabilities to analyze a 1971 issue of Drag Magazine from the Digital Transgender Archive (Brewster 1971). The issue centers on celebrating drag queens as local royalty, documenting differing strategies for gender performance and embodiment, and capturing a range of emotions surrounding queer life related to the drag ball community, as well as the trauma they experience due to transphobia and homophobia. To prepare for the AI task, students were instructed to read brief overviews on the history of Drag Magazine as published in the Smithsonian (Boldt 2023) and the New York Public Library’s websites (Golia 2023). While the goal was for students to examine the gay liberation movement through the lens of a small, trans-femme-centered queer community, the AI tool became a way to highlight emotional tone in images and text. Students were then asked to reflect on how ChatGPT’s findings compared to their own. With the notes students collected from comparing and contrasting two issues of Drag Magazine, the task then required students to use AI to create a poem that synthesized their notes in creative form in a style that might have been included in the original issue. The poem as a method of assessment, suggested by a student, offered students a quick way to communicate their findings of the sentiment analysis while also acknowledging aspects that stood out to them. Taken together, the assignment not only exposes students to primary sources in a non-confrontational setting but also puts students in the position of analyzing the GPT’s findings as a pedagogical tool for learning sentiment analysis.
Reviewing their work, students (especially Asia Campus students) revealed how many had never seen a drag performance due to public scrutiny and shaming. While the narrative of transphobic and homophobic harassment was one that currently saturates their contemporary experience, students from Utah expressed admiration at the magazine’s volume of smiling photo collages, the glitz and glamour, as well as the writers’ pride in drag queens referring to themselves in the text as “local celebrities.” With the goal of continuing conversations, GPTs trained to implement Paul Grice’s cooperative principle of communication through conversational alignment such as extending and elaborating on the user’s framing rather than contradicting it. This alignment is why GPTs rarely reject a premise unless directly prompted to because they assume the user’s topic, tone, and direction are intentional (for better or worse). Faculty frequently incorporate this concept into class discussions through improvisational “yes, and” communication techniques, synthesizing student responses with lecture content. GPTs are an endless source for conversational mirroring, allowing the chattiest students to discuss themes they found on a topic. Research shows that students learn more when they find personal connections to course material that evoke emotions, such as empathy (Immordino-Yang 2015). Yet it is often topics about marginalized groups like trans women and drag performers in our contemporary political climate that can invite student backlash that can be harmful to the class and community. In light of recent awareness of the role ChatGPT is playing in mental self-help, OpenAI updated its model to engage in restorative dialogue in moments of distress; however models are not fail-proof.
Similar to faculty, ChatGPT is also not immune to replicating potentially discriminatory communication. A recent study by scholars Hua et al (2024) illuminated how OpenAI introduced reinforcement learning through human feedback to create ethical safeguards for mitigating toxicity (discriminatory language) after evidence of discriminatory language on GPTs surfaced. Despite efforts, discriminatory language cannot be completely prevented due to systemic bias, rendering conversational alignment a double-edged sword. Part of AI literacy must include discussions of the limitations of GPTs as well as the responsibility of students as historians to not only conversationally collaborate but fact check, interrogate, and redirect using evidence-based arguments. Moments of critical reflection, document analysis, or other methods of critical inquiry essential to learning cannot exist solely through AI. Like a calculator, it is a tool that can aid some teaching and learning processes in making traditional styles of learning more dynamic and digitally interactive. However, AI should never supplant the pedagogical relationship between the instructor and student.
Grading AI tasks
Most students understandably are fearful of new methods of assessment which has led me to assess AI tasks through a method of labor-based grading—a method of grading in which student effort is privileged more than perfection. By focusing on effort and growth more than mastery, labor-based grading encourages risk-taking and exploration and allows students more flexibility with accommodations. Students complete a list of assignments and earn more credit for completing more assignments. Rather than rubrics that give points piece by piece, labor-based grading uses checklists to make sure you have all components required to earn a complete. Complete all the items and they earn full credit. Miss one piece, and they need to revise. Rather than providing rubrics which situate student output along a spectrum of mastery, grading an assignment as Complete or Incomplete (full credit or no credit) but allowing revisions ensures many elements: that students are paying attention to small details just as much as larger objectives to be able to gauge the learning process, the process affords accessibility to second-language learners as well as first-semester and first-generation college students, and finally the process gives the student the power to manage their time and their grade.
Just as with grading student work, AI tasks may require faculty to complete multiple revisions to ensure the highest level of understanding. I encourage students to ask clarifying questions as a method of prompt engineering for me, and I work with an undergraduate peer advisor to not only mentor students on the software but feed recurring issues back to me in case of an assignment update. Having tested multiple approaches to digital archive engagement, including free roaming exploration, hybrid learning, and guided instruction, step-oriented AI tasks that require students to complete benchmarks allow faculty a greater ability to identify learning issues earlier in the course and more easily in an assignment versus a traditional paper assignment or quiz. Because AI software is in constant flux, allowing assignment revisions can help students be more patient with you and with this technology.
A common concern among faculty is that students will use GPTs to complete their coursework. Students will use AI to complete their critical reflections, regardless if a course has integrated AI tasks or not. AI and plagiarism detectors are not always accurate, generate ethical issues due to storing and capitalizing on student work, and ultimately do not prevent further plagiarism. Fears of plagiarism, replacement critical thinking, and AI hallucinations have pushed some universities and colleges, from Ivy League universities like Princeton to smaller public institutions like Western Washington University, to ban the use of AI specifically within history courses. The University of Utah where this study was conducted might rank moderately on a national scale of pedagogical AI friendliness, providing materials that could prove helpful to faculty new to incorporating AI. The university’s policies on AI changed drastically over the study’s eighteen months, from initially relying on faculty to create their own AI policies to releasing an AI-specific website showcasing different facets of learning, research, and teaching related to AI at the university. This initiative became part of the expansion of the university’s mission to include its encouragement of “responsible advancement and use of artificial intelligence in support of its mission spanning research, education, and societal impact” (University of Utah, n.d.). Most recently, the university released a new “AI and You: Student Guide to Generative Tools” as well as a guide and training course specifically for faculty, which can serve as models to faculty interested in designing guides and syllabi statements (Martha Bradley 2025b). The university’s approach to guiding students on originality with AI use remains unclear.
As my implementation of AI-integrated assignments increased, my analysis revealed that students did not plagiarize more. However, my ability as an instructor to identify likely AI-generated material became stronger as my own knowledge of the limitations of these tools strengthened. In my experience working with largely ESL students, many students use artificial intelligence tools with no intention of plagiarizing, due to lack of confidence over their English-speaking ability, fears of professor ridicule, unease with reliance on critical thinking in Western-style education, etc. Native English speakers also use AI for various reasons that are not rooted in malice, including a lack of confidence in their writing ability, difficulties with historical inquiry for nonmajors, and issues with reading comprehension. While not excuses for AI dependency, they do indicate a lack of foundation in digital literacy, AI literacy, and primary source literacy skills critical in this moment of education. Incorporating monitored and scaffolded AI tasks that allow faculty the ability to walk students through appropriate AI usage can prove helpful in modeling to students the software’s limitations as well as the need for original critical reflection.
Directions for Further Research
Digital material is not permanent, and designing digital assignments can inadvertently become limiting when reluctance to use AI remains high. Even when working with major digital archives and digital history sites on AI tasks, such as the New York Public Library, the National Museum of the American Indian, and the Public Broadcasting System’s documentary-based exhibits, this year has brought to the surface how shifting political climates change shape federal funding which can put pressure on public history sites in ways that can dramatically change digital archival research (Kawamura 2025; Lucas 2025; Zwarenstein 2025). Over the past two years some historians and archivists have increasingly showcased their use of AI in historical research, including deciphering ancient texts, using chatbots as oral historians, and utilizing machine learning in analyzing large datasets of historical information (Henry Stewart DAM 2021; Stanford 2021; Oral History Association 2024a; NYU 2023; Thomas and Testini 2024; Dentler et al. 2024; American Historical Association 2021; Smithsonian 2022; American Historical Review 2024; Oral History Association 2024b; Royal Historical Society 2024). Despite leading digital historians like Jo Guldi, Kalani Craig, Lauren Tilton, and David Trowbridge, among others, arguing for the need to demystify AI and learn from the LLM’s own learning processes how to teach history, history surveys, like many other classrooms, have yet to pivot as this paradigm shift in the field begins to inundate students’ lives and learning styles. In a recent New Yorker piece, historian D. Graham Burnett empathized with faculty who are stuck in paralysis: “Staggering transformations are in full swing. And yet, on campus, we’re in a bizarre interlude: everyone seems intent on pretending that the most significant revolution in the world of thought in the past century isn’t happening” (Burnett 2025).
Some scholars, like Shae Omonijo, claim that the looming “AI Bubble Burst” could offer both faculty and students opportunities to rethink the purpose of education altogether. On their YouTube channel focused on “Critical Thinking in the Age of AI,” Omonijo argues that because AI is built on datasets that are quickly becoming exhaustive, new pressure will be put on original data, including undigitized archival material and new original creative expressions. More importantly, Omonijo’s argument and the following reflect a growing pedagogical movement to parse the nuances of critical thinking as essential to the human experience. Omonijo argues that the rapid rise of AI relies on the vast amount of original human-created data that will eventually plateau due to AI use. In response, greater demand will be put on that original human-created data, particularly involving human emotions and experiences. She states, “Your ability to connect ideas across time, across disciplines, and across emotions [is key]….Only you can create your own knowledge map” (Omonijo 2025). Because of the current constraints that limit AI generation to the prompt creator, Omonijo argues that the power of AI rests in crafting that prompt. If “critical thinking is a process, not a reaction,” then guiding students on how to effectively use AI as a medium for historical inquiry while still pushing the need for original critical reflection is key.
While many critical of AI fear that an emphasis on the “digital” will supplant the need for “pedagogy,” the field of “critical digital pedagogy,” Jesse Stommel et al. argue, is about “the challenge… of finding ways to teach through a screen, not to a screen” (2020). Echoing Paolo Freire, Stommel et al. argue that digital methods, when approached critically, can help us better understand the very “humanness” of learning. Expedited by the rapid transition to online learning during the COVID-19 pandemic, many scholars of educational technology began to unite with scholars in digital humanities and critical pedagogy to shift teaching methods for digital native students. Scholar Daniel Gutiérrez-Ujaque has argued that these societal shifts have challenged educators to develop methods of critical digital literacy in order to more effectively prepare students to evaluate and apply technological tools “to mitigate the effects of technological determinism, understood as a theory that technology is the main factor shaping a society’s social, economic, and cultural development” (Gutiérrez-Ujaque 2024). More broadly, approaching AI tools through the lens of critical digital pedagogy can help students apply digital literacy concepts, such as disinformation, social networks, ethics, and privacy, to sources of knowledge and power within different cultural and historical contexts.
Conclusion
While the rapidly shifting terrain of AI tools, limitations of free usage, and general American student unease of using AI did present some challenges, overall iterative use of AI in history education met or surpassed every anticipated goal—most importantly when it came to engaging with primary source materials, including higher usage, comprehension, and application of archival materials by students. Through consistent work with students across a variety of backgrounds through this project, I have found that achieving the benefits of AI means pushing students out of their comfort zones in this new digital terrain—pushing them to take time with AI tools, take risks, and sit with the feelings of uncertainty that propel the process of historical inquiry. As Burnett recounted in piece on the positive transformations AI can have on the humanities, ChatGPT offers the ability for students to “descen[d] more deeply into [their] own mind, into [their] own conceptual power, while in dialogue with an intelligence toward which [they] have no social obligation. No need to accommodate, and no pressure to please” (Burnett 2025). This discovery, critical inquiry in its purest form, requires time, patience, and attention that most faculty do not have to spend one-on-one with students.
For a history major in my “AI in the Archives” class, their explorations of ChatGPT led to them to design their own GPT as a pedagogical tool for their final project on the Romanian Revolution. Because a key component of the assignment required students to showcase a digitized oral history collection through interactive play and critical thinking, the student engineering their own GPT to simulate a role-playing scenario with students learning about the Romanian Revolution through an oral history collection—a truly phenomenal creation of their own volition that demonstrates some students’ genuine interests in playtesting AI. Despite current limitations in GPT creation due to subscription minimums, students creating their own GPTs potentially offers an exemplary new mode of assessment because students have the ability to set their own conversational boundaries called parameter settings. In turn, this involves students in rubric creation and pushes them to clarify their learning objectives which can help them understand how they themselves are being assessed.
Taken together, while working with AI tools in the classroom remains a new experience, despite institutional and cultural limitations, my experiences reveal that the future of AI’s impact on pedagogy appears promising rather than perilous. Although these capabilities are limited, AI can help illuminate students’ thought processes via shared chat URLs in ways that traditional methods that require high student-to-faculty ratios cannot. Faculty hesitation and disciplinary resistance remain high and institutional barriers to using AI in many departments and universities are real and urgent. Limitations on tool access, data privacy policies, and uneven digital infrastructure across platforms, and with departmental restrictions on AI usage must be addressed for AI to be sustainable in pedagogy. Despite AI tool potential and decades of digital humanities field development, faculty skepticism of AI (especially when working with American studies) underscores the need for peer-led modeling, transparency, published case studies, and even student feedback on AI. In short, breaking down how to use AI tools and how machine learning functions helped demystify the process in a way that directed both student and faculty attention to critical inquiry.
Free and low-cost AI tools enabled me to think more critically about how we teach historical inquiry and how students are learning it. AI helped facilitate student agency in historical inquiry while still allowing me the ability to interject and redirect learning. Conversational walkthroughs of course material (especially primary and secondary sources) can help students see themselves as collaborative partners in education. Through machine learning, students were able to play with different creative mediums that helped humanize primary sources and helped students find personal connections to archival material at a much faster rate. While AI tools did not directly increase student exposure to marginalized histories, ChatGPT offered a largely neutral platform with which students could explore evidence-based arguments about taboo topics free from criticism. In particular, PDF analysis AI tools helped add depth to explorations of marginalized groups and sources by directing them to concepts missed in their own analyses. Rather than supplanting critical thinking, AI tools that enabled a greater extent and faster use of archival material helped increase student exposure to a wider variety of perspectives in history without altering the course entirely. Despite these successes, AI is not a replacement for traditional methods of education. These uses complemented non-digital pedagogical methods, including: small group discussions, analog creative projects, game-based learning, and film analysis.
Through guided pathways that offered semi-structured learning, AI tools helped lower entry to participation, especially for ESL learners, and increased opportunities for all students to learn history through creative expression and play. Shared chat logs and iterative prompts offered insight into how students reason, question, and build historical arguments—a powerful formative assessment tool. Most importantly, they helped me see moments where students “didn’t get it,” which were plenty. Student feedback pushed me to break assignments down into multiple AI tasks and be more explicit in detailing assignments’ larger motives in preparing students for the final project requiring synthesis of historical inquiry skills. While course time is limited, reflective pauses can help students process their relationship to the tools that can strengthen their relationship to the content and history more broadly. AI tasks can help you gauge students’ understanding of course content just as quickly as they can help you understand how students are feeling about the course formatting. More than anything, AI tasks are barometers for faculty, not simply methods of assessment for students.
Ultimately, integrating AI literacy into your classroom means training yourself to become adaptable to new technology, to be flexible in digital engagement, but also be patient when students fall short of the critical analysis you expected. This struggle is not unique to artificial intelligence, nor is it limited to our contemporary historical moment. In a political climate increasingly hostile to the liberal arts, AI integration presents opportunities for enriching pedagogy that can help make our work as educators more relevant within students’ increasingly digital lives. For educators looking to integrate minimalist AI into their pedagogy, a great place to start is opening up ChatGPT and initiating a conversation with a primary source you would like your students to explore more deeply. Using your own assignments as starting points, prompt the AI to suggest different strategies for learning objectives you would like to explore. Finally, be open and communicative listeners with your students regarding using AI as a tool rather than the answer. These experiences can serve as effective ways to challenge skeptical faculty as the field progresses while also helping faculty hone their own pedagogical skills.
