Abstract

Companies deploy rudimentary Virtual Agents (VAs) to handle customer inquiries. Technical limitations can lead to comprehension issues, resulting in
frustrating interactions. The credibility of the VA, particularly its perceived social presence, influences how users assess interactions for pleasantness,
satisfaction, and their overall attitude towards the VA. The use of Deep Learning language models like GPT-3 holds promising
potential for improving credibility. Dialogue can be tailored to the user's intent, potentially enhancing satisfaction with the interaction. Current
research suggests that incorporating Text-to-Speech systems could positively impact the credibility of human-VA interactions. VAs utilizing speech output
are perceived as more likable and personal. This study aims to examine to what extent the use of GPT-3 generated
dialogue and Text-to-Speech Systems can enhance the interactions and the evaluation of VAs in terms of credibility and user satisfaction.