Advancing Al-Generated Inoculation Messaging: A Comparative Study of ChatGPT-40, Microsoft Co-Pilot, and Gemini
Category
Business, Education and Humanities
Department
Communication
Student Status
Graduate
Research Advisor
Dr. Alicia M. Mason
Document Type
Event
Location
Student Center Ballroom
Start Date
10-4-2025 2:00 PM
End Date
10-4-2025 4:00 PM
Description
A 2023 exploratory study examined ChatGPT-4's ability to analyze, adapt, and independently generate inoculation messages through a 10-week training initiative. This initiative assessed whether ChatGPT-4 could identify key structural elements (e.g., forewarnings, preemptive refutations), enhance message features (e.g., linguistic style, length), customize messages for specific audiences, and replicate the inoculation message design process on novel issues. Twenty-four previously published inoculation messages were used to train ChatGPT-4 through various prompting techniques, including sequential, active, iterative, and chain-of-thought prompts. While the Al demonstrated originality in its responses, structural weaknesses were evident, such as difficulty producing clear, accessible language and advanced education-level comprehension requirements. Additionally, ChatGPT-4 struggled to develop explicit forewarnings and limited its messages to two refutations. Open Al was unable to craft threat components with consistency leading Mason and colleagues (2024) to observe that "inoculation strategies are more than merely information and transmission, but rather tools for carefully crafted appeals which are psychological arousing in order to generate threat and threat awareness in message recipients," (p. 7). The current investigation attempts to replicate the 2023 study using ChatGPT-4o and expands the focus of inquiry to two additional generative Al platforms -- Microsoft Co-Pilot and Gemini, Findings, discussion, practical significance, and limitations are offered.
Advancing Al-Generated Inoculation Messaging: A Comparative Study of ChatGPT-40, Microsoft Co-Pilot, and Gemini
Student Center Ballroom
A 2023 exploratory study examined ChatGPT-4's ability to analyze, adapt, and independently generate inoculation messages through a 10-week training initiative. This initiative assessed whether ChatGPT-4 could identify key structural elements (e.g., forewarnings, preemptive refutations), enhance message features (e.g., linguistic style, length), customize messages for specific audiences, and replicate the inoculation message design process on novel issues. Twenty-four previously published inoculation messages were used to train ChatGPT-4 through various prompting techniques, including sequential, active, iterative, and chain-of-thought prompts. While the Al demonstrated originality in its responses, structural weaknesses were evident, such as difficulty producing clear, accessible language and advanced education-level comprehension requirements. Additionally, ChatGPT-4 struggled to develop explicit forewarnings and limited its messages to two refutations. Open Al was unable to craft threat components with consistency leading Mason and colleagues (2024) to observe that "inoculation strategies are more than merely information and transmission, but rather tools for carefully crafted appeals which are psychological arousing in order to generate threat and threat awareness in message recipients," (p. 7). The current investigation attempts to replicate the 2023 study using ChatGPT-4o and expands the focus of inquiry to two additional generative Al platforms -- Microsoft Co-Pilot and Gemini, Findings, discussion, practical significance, and limitations are offered.
