Supplementary data for the paper De Winter, J. C. F., Dodou, D., & Stienen, A. (2023). ChatGPT in education: Empowering educators through methods for recognition and assessment. Informatics.



workshop.m				MATLAB script used during the workshops. It loops through the last 30 emails
					and displays them on the screen. The code also contains parts where
					selected participants’ prompts are automatically forwarded to the API, 
					and a part where the workshop host can enter custom prompts manually.
repositories_and_publishers.xlsx	An Excel file containing the number of records with target and control words from
					publishers and thesis repositories is provided. These data correspond to Table 1 
					and Figure 1. An update of these data as of 25 November 2025, 
					including information from more publishers and repositories, is also included.



Further remarks about the use of the workshop.m script:

A challenge in organizing the workshops was to enable interaction between the workshop participants and GPT-4, without requiring participants to subscribe to OpenAI, possess a laptop, or install software. Our solution was to allow participants to email their prompts to one of the organizers. Through our custom script, these prompts were displayed on a projection screen and were then forwarded to GPT-4 using the API. The use of email proved to be an easy solution compared to more technical polling tools and software that are known to exist.

Note that participants were instructed to use the word ‘chatgpt’ in the subject line, and include only the prompt in the message body. By automatically filtering the incoming emails based on their subject line, it was ensured that private emails were skipped. More specifically, emails were selected for display when the term ‘chat’ (case insensitive) appeared in the subject line. The selection was solely based on the word ‘chat’ because, as anticipated, some participants wrote ‘chat gpt’ (with a space) or ‘chat gtp’ (a typographical error).

An oversight encountered during Workshop 1 was the legibility of the prompts displayed on the projected screen; the font size was too small for the size and resolution of the screen, particularly for those situated at the back of the room. Therefore, it is advisable for workshops similar to ours to have a collaborator who can read out the ChatGPT input and output. Readability was not a problem in Workshop 2, where two large digital screens were available.
