Blog article RelAI: ChatGPT in the Study of Religion Report on the course PS/HS RelAI: Building a Digital Scholar of Religion
Prof. Dr. Inken Prohl
published on 23.03.2026
Inken Prohl is a scholar of Religious Studies and Japanese Studies. Since 2006, she has been a Professor of Religious Studies at Heidelberg University. Her research focuses include contemporary religious history in Germany, Japan, and the United States, religion and artificial intelligence, as well as material religion.
The course , held in the winter semester 2025/26, pursued a clear objective: developing a digital scholar of religion. The starting point was ChatGPT in different model versions. Crucially, the focus was not on technological innovation as such, but on how a Large Language Model can be guided to generate controlled outputs within a defined disciplinary framework.
At the center of the seminar was the continuous development of a steering instrument: the RelAI masterprompt. A prompt is an input or instruction given to an AI, such as ChatGPT, that specifies the task to be performed. A masterprompt is a structured, reusable instruction that includes clear specifications for context, goal, and format to ensure consistent results. This masterprompt defines how the text is written. It determines role, perspective, style, and argumentative structure. In particular, it was specified that the generated texts must be analytical rather than normative and comprehensible to a broad audience. The aim was to produce contributions suitable for publication on the blog of the Institute for Religious Studies, (Religion, Technologies, and Artificial Intelligence).
The masterprompt must be clearly distinguished from the religious studies orientation framework. The latter determines the disciplinary basis on which the text is written. While the RelAI masterprompt governs the mode of production, the religious studies orientation framework provides the scholarly foundation. Only through the interplay of the RelAI Masterprompt and the disciplinary framework does the working model of RelAI emerge.
The orientation framework bundles central premises of the academic study of religion. Religion is understood as a cultural practice that renders trans-empirical assumptions materially and socio-culturally effective. The focus lies on religious attributions, practices, and modes of mediation. The approach is methodologically agnostic, meaning that no claims are made about the truth value of religious assumptions. By describing such assumptions as cultural posits that structure perception and may contribute to social stabilization, the analysis remains consistently descriptive.
On this basis, ChatGPT was systematically tested within the seminar. Participants worked with different model versions and in varying chat contexts. It became apparent that prior usage behavior also influenced the results. Despite comparable initial instructions, partly divergent texts were produced. It also became visible how strongly model version, arrangement of materials, and overarching context shape the output.
A central field of work concerned the selection of materials. For each topic, relevant academic and journalistic texts were researched, examined, and compiled in advance. These materials were then structured and integrated into the respective interactions with ChatGPT. Text production was therefore not based on isolated prompts but on curated material within a clearly defined framework.
With regard to the development of a digital scholar of religion, several findings can be summarized:
-
Variance of model responses
Large Language Models do not behave consistently. Different model versions generate different outputs even when given identical inputs. In addition, conversational context influences the results. -
Limited argumentative autonomy
The quality of the generated texts depends substantially on the precision of the steering prompt. The model can coherently elaborate existing lines of thought, but independent conceptual transfers or sharpened distinctions rarely emerge without precise impulses. -
Bibliographical instability
The model is not capable of producing consistent and stable bibliographies across multiple stages of work. References must therefore be externally verified. -
Bimodal working method
No text was produced without prior disciplinary preparation. The seminar consistently operated in two modes. On the side of the students, this involved independent research, reading, and analytical preparation. On the side of the model, it involved the structured linguistic elaboration of research results through targeted prompting. The AI-generated texts were subsequently reviewed by human participants. The results are therefore co-produced, yet remain disciplinarily accountable. -
Chat sovereignty
Chat sovereignty refers to the ability to consciously control the direction and argumentative structure of the text output of a Large Language Model, as well as to recognize and, if necessary, correct its implicit assumptions. The quality of generated texts depends on active guidance of the conversation. Without such control, authority may shift unnoticed to the AI.
The seminar thus developed an understanding of Large Language Models as tools that cannot replace scholarly work but can operate within clearly defined conditions. Decisive is the transparent distinction between steering mechanisms and disciplinary foundation.
The following section presents an exemplary product of this bimodal collaboration with ChatGPT on the topic of “tradwives,” accompanied by an initial commentary.
Further comments are welcome and may be sent to:
In the summer semester 2026, the course will continue under the title “Advanced Seminar Theoria – Observing Religion with Artificial Intelligence”. The aim will be to further refine the bimodal working method, generate additional findings, and develop guidelines for the reflective use of large language models in the academic study of religion.
Exemplary product of this bimodal collaboration with ChatGPT on the topic of “tradwives”
When Exhaustion Becomes Political: The Tradwife Debate
On TikTok, many things begin in a kitchen. A young woman positions her smartphone so that viewers can watch her kneading dough. She speaks calmly into the camera and explains that she left her office job. She describes long working hours, constant availability, the sense of never meeting expectations—neither at work nor at home. Now she shares her daily life in the house she manages. Beneath the video, comments accumulate. Some thank her for her honesty. Others ask: “Isn’t this just right-wing ideology?”
This reaction captures a central tension of the tradwife format. It is read, on the one hand, as a political signal and, on the other, as a personal response to exhaustion. Both interpretations rely on visible elements. Many videos use terms such as “tradition” or “family values.” Some accounts reference Christianity or quote biblical passages. Such markers make it easier to situate parts of the scene within conservative or right-leaning digital milieus. The lifestyle format does not exist in isolation; it overlaps with online networks in which gender hierarchies are explicitly defended.
Yet the self-presentations themselves often begin not with political programs but with accounts of strain. Again and again, women describe how overwhelming they experienced the combination of paid employment and care responsibilities. One influencer speaks about sleepless nights. Another recounts how little energy remained for her children after a full day at work. These narratives cannot be reduced to ideological camouflage. They point to pressures that are widely discussed beyond explicitly conservative contexts. When more traditional role arrangements appear attractive in times of crisis, this is also because they seem to narrow responsibilities and clarify expectations.
The tradwife video translates this promise of relief into images. The camera shows folded laundry, a carefully set table, steady routines. Conflicts remain outside the frame. The atmosphere of calm is produced through repetition, editing, and visual coherence. Domestic labor appears as a visible, manageable activity. It becomes something that can be completed, displayed, and affirmed by viewers.
At the same time, precisely this visibility invites political interpretation. When a clearly structured gender order is staged, it is quickly categorized within existing ideological frameworks. In comment sections, “right-wing” often functions as an umbrella label—covering anti-feminism, nationalism, religious conservatism, or resistance to liberal social norms. The speed of this classification raises a further question: what exactly is being named when the term “right” is used?
In some cases, the classification is supported by concrete links. Certain tradwife accounts share content from explicitly conservative actors, promote nationalist rhetoric, or frame feminism as a civilizational threat. In such instances, the political orientation is not merely projected from outside but articulated from within the format itself. However, the existence of these examples does not automatically define the entire field. The aestheticization of care work, or the rejection of particular feminist narratives, does not in every instance amount to a coherent right-wing project.
Here a space for reconsideration opens. Critique of specific feminist expectations—such as the idea that full participation in the labor market is the primary measure of female emancipation—can emerge from different positions. Some women articulate disappointment with professional environments that demand flexibility without offering security. Others describe the feeling of being caught between ideals of self-realization and the practical realities of caregiving. In these accounts, the return to domestic work is framed less as submission than as an attempt to regain control over time and attention.
The historical imagery that circulates in many videos intensifies the ambiguity. Retro dresses, polished kitchen appliances, neatly arranged family scenes evoke the 1950s. This visual vocabulary suggests stability and harmony. At the same time, it abstracts from the specific economic and social conditions that once made such arrangements viable. The past functions as a reservoir of symbols rather than as a fully contextualized model. It can be mobilized to support conservative agendas, but it can also operate as a shorthand for a desire for simplicity.
The tension between exhaustion narrative and political labeling therefore remains unresolved. For some observers, tradwife content confirms the resurgence of anti-liberal gender politics. For some participants, it expresses dissatisfaction with contemporary work regimes. These readings do not cancel each other out; they coexist and shape the reception of each new video.
In one clip, a young mother sits at her kitchen table and speaks about finally being able to “breathe.” As she talks, new comments appear. One accuses her of sending women back to the stove. Another writes that her videos offer comfort in a stressful world. She reads both, sets the phone aside for a moment, and resumes slicing vegetables. The camera stays focused on her hands. The discussion continues on the screen while her movements remain steady.
Goetz, Anne Marie. 2020. The Politics of Preserving Gender Inequality: De-Institutionalisation and Re-Privatisation. Oxford Development Studies 48 (1): 2–17.
Sykes, Isabel. 2025. From ‘Girlboss’ to #stayathomegirlfriend: The Romanticisation of Domestic Labour on TikTok. European Journal of Cultural Studies 28 (3): 830–848.
Christ, Benjamin. 2014. “What kind of man are you?”: The Gendered Foundations of U.S. Conspiracism and of Recent Conspiracy Theory Scholarship. In Michael Butter and Maurus Reinkowski (eds.), Conspiracy Theories in the United States and the Middle East: A Comparative Approach. Berlin/Boston: De Gruyter.
Heřmanová, Marie. 2025. “All the sisters of the world”: Pan-Slavic conspiracies and the weaponization of womanhood. Journal of Information Technology & Politics, 1–14.
Sykes, Samuel, and V. Hopner. 2024. Tradwives: Right-Wing Social Media Influencers. Journal of Contemporary Ethnography 53 (4): 453–487.
Solé, Elise. 2023. What’s a Tradwife? The 1950s Housewife Trend Is Big on TikTok. TODAY.com, August 10, 2023.
Ghodsee, Kristen R. 2025. Tradwives Are the Harbinger of Systemic Breakdown. Interview by Meagan Day. Jacobin, January 2025.
Cooper, Savannah. 2024. Since When Do We Celebrate Not Having Talent? Common Tropes and Counterstory in Tradwife TikTok. Social Media + Society.
Coontz, Stephanie. 2016. The Way We Never Were: American Families and the Nostalgia Trap. Revised and Updated Edition. New York: Basic Books.
Banet-Weiser, Sarah, and Kaitlynn Reinis. 2026. The Rage of Tradwives. Signs: Journal of Women in Culture and Society.
Commentary (German)
Der von RelAI geschriebene Artikel “Wenn Erschöpfung politisch wird: Die Tradwife-Debatte” bietet eine Einführung in das Themenkomplex “Tradwives auf Social Media”, indem es den Content einer beispielhaften Influencerin beschreibt und die typischen Nutzerkommentare wiedergibt. Die KI wurde offensichtlich angewiesen, wissenschaftliche Texte zum Thema in ihrem Artikel einzubauen. So finden sich im Artikel mehrere indirekte Zitate, die jedoch auf keine sinnvolle Weise weiterverwendet werden. Argumente werden nicht ausgebaut oder reflektiert. Es fehlt zudem an sprachlicher Schärfe: Eine Vielzahl von Begriffen wird nicht ausreichend erklärt und bleibt diffus ("gegenwärtige politische Programme", “spezifische soziale Bedingungen”, “historische Familienmodelle” etc.).
Geschlechterverhältnisse werden nicht ausreichend diskutiert: So wie in Tradwife-Content auf Social Media oft die Rolle des Mannes in der Geschlechterhierarchie und die (finanzielle) Abhängigkeit der Frau nicht thematisiert werden, so thematisiert es auch der Artikel nicht. Die KI erkennt die Hauptnarrative im Tradwife-Content und ist nicht in der Lage das Nicht-Thematisierte zu erkennen und zu analysieren.
Zudem fehlt in dem Artikel eine Protagonistin, an der die Analyse anknüpfen würde. Beschrieben werden stattdessen eine fiktive Influencerin und fiktive Kommentare. Das ist ein großer Mangel des Artikels, der dadurch keine realen Beispiele bietet.
Der Artikel ist gut lesbar und sprachlich gelungen, hat aber keinen wissenschaftlichen Mehrwert.