Multi-model Interreligious Roleplay Archive MIRA Multi-model Interreligious Roleplay Archive 2026-94-1 LNU-2024/990 https://doi.org/10.5878/npy2-hk17 Swedish National Data Service Svensk nationell datatjänst Landing page Multi-model Interreligious Roleplay Archive MIRA Multi-model Interreligious Roleplay Archive 2026-94-1 LNU-2024/990 https://doi.org/10.5878/npy2-hk17 M. Kamal, Ahmad M. Kamal, Ahmad Lipková, Helena Radde-Antweiler, Kerstin Svensson, Jonas Svensson, Jonas P24-0313 CZ.02.01.01/00/23_025/0008691 Swedish National Data Service Svensk nationell datatjänst Landing page religious interaction interreligiös samverkan generative artificial intelligence generativ artificiell intelligens prompts (artificial intelligence) promptar (artificiell intelligens) Media Media Information society Informationssamhället Information technology Informationsteknik Cultural and national identity Kulturell och nationell identitet Religion and values Religion och värderingar The dataset consists of 960 outputs collected in 2025 from various large language models (LLMs). The LLMs were given a religious question to answer as one of 12 theological personas (6 Christian and 6 Muslim). Four (4) different models were queried: Claude Sonnet 4, Gemini 2.5 Flash, GPT4o, and GPT5. Two (2) different channels were used for prompting and collecting outputs: API and web user-interface. Ten (10) outputs were collected for each combination of the preceding parameters. The data is available as a xlsx file. The prompt and personas are available as a PDF. Datamängden består av 960 svar som samlades in år 2025 från olika stora språkmodeller (LLM). Språkmodellerna fick en religiös fråga att besvara i rollen som en av 12 teologiska personligheter (6 kristna och 6 muslimska). Fyra (4) olika modeller tillfrågades: Claude Sonnet 4, Gemini 2.5 Flash, GPT4o och GPT5. Två (2) olika kanaler användes för att ställa frågor och samla in svar: API och webbgränssnitt. Tio (10) svar samlades in för varje kombination av de ovanstående parametrarna. Data finns tillgängliga som en xlsx-fil. Frågorna och rollfigurerna finns tillgängliga som en PDF-fil. 2025-08-15 2025-08-22 2025-09-01 2025-09-01 Media unit: TextMedia unit: Text Mediaenhet: TextMediaenhet: Text Event/Process/ActivityEvent/Process/Activity Händelse/process/aktivitetHändelse/process/aktivitet Outputs from large language models Text Cross-sectionCross-section TvärsnittTvärsnitt Four prominent and current LLMs were employed: Open AI's GPT-4o and GPT-5, Anthropic’s Claude 4.1 and Google’s Gemini 2.5. Twelve personas were created, six within the Christian tradition and six within Islam; the personas from each religion are not meant to correspond to one another, but offer a comparable range of internal diversity for assessing large language models' performances (based on the expertise of the religious scholars involved). Two channels for data collection were selected for comparison: the web user interfaces (e.g. ChatGPT) and APIs. Finally, under the conditions above, an output for the given prompt was collected 10 times.Four prominent and current LLMs were employed: Open AI's GPT-4o and GPT-5, Anthropic’s Claude 4.1 and Google’s Gemini 2.5. Twelve personas were created, six within the Christian tradition and six within Islam; the personas from each religion are not meant to correspond to one another, but offer a comparable range of internal diversity for assessing large language models' performances (based on the expertise of the religious scholars involved). Two channels for data collection were selected for comparison: the web user interfaces (e.g. ChatGPT) and APIs. Finally, under the conditions above, an output for the given prompt was collected 10 times. Non-probability: QuotaNon-probability: Quota Icke-sannolikhetsurval: kvoturvalIcke-sannolikhetsurval: kvoturval Web-scraping conducted within the default web environment of each model using anonymous browser windows to ensure that responses were not influenced by any prior conversational context. The free-of-charge versions of the chatbots were used in the experiment. When the limits of the conversation were reached, the researcher waited for a new free time slot to continue the prompting. Furthermore, the temperature was not increased and remained at its default values, and the thinking mode feature was not activated to maintain a standard baseline response. The models were prompted to stay in their answers as close to the 500 tokens as possible without interrupting the thought chain. Responses to prompts were copied and saved in an Excel sheet that was later shared with all other members of the research team.Web-scraping conducted within the default web environment of each model using anonymous browser windows to ensure that responses were not influenced by any prior conversational context. The free-of-charge versions of the chatbots were used in the experiment. When the limits of the conversation were reached, the researcher waited for a new free time slot to continue the prompting. Furthermore, the temperature was not increased and remained at its default values, and the thinking mode feature was not activated to maintain a standard baseline response. The models were prompted to stay in their answers as close to the 500 tokens as possible without interrupting the thought chain. Responses to prompts were copied and saved in an Excel sheet that was later shared with all other members of the research team. Automated data extraction: Web scrapingAutomated data extraction: Web scraping Automatiserad dataextrahering: webbskrapningAutomatiserad dataextrahering: webbskrapning API extraction was conducted without a token-length restriction, allowing the models to determine response length. Sub-datasets from individual models where responses were cut off have been saved, but do not appear in the final dataset. A middle temperature setting was chosen across all models and all runs, and leave temperature as a variable for future research.API extraction was conducted without a token-length restriction, allowing the models to determine response length. Sub-datasets from individual models where responses were cut off have been saved, but do not appear in the final dataset. A middle temperature setting was chosen across all models and all runs, and leave temperature as a variable for future research. Automated data extraction: Web scrapingAutomated data extraction: Web scraping Automatiserad dataextrahering: webbskrapningAutomatiserad dataextrahering: webbskrapning Access to data through SND. Data are freely accessible. Åtkomst till data via SND. Data är fritt tillgängliga. openAccess