Abstract:
The adoption of conversational artificial intelligence does not depend solely on its technical performance but also integrates complex relational dimen¬sions that shape the overall user experience. This study examines the evolving perception of ChatGPT across its four successive versions (GPT-3.5, 4, 4.5, and 5), analyzing how improvements in comprehension and contextualization inter¬act with functional, emotional, trust-related, autonomy, and perceived control dimensions of experience.
The longitudinal analysis draws on more than 20,000 YouTube user comments collected between November 2022 and August 2025, enriched by media and aca¬demic sources. Results obtained through an NLP methodology reveal a growing dissociation between functional quality and user satisfaction. While contextual understanding objectively improves (reduced hallucinations, ability to process long texts up to 25,000 words, enhanced performance in mathematics and programming), the proportion of satisfied users paradoxically declines. This downward trajectory can be explained by the differentiated evolution of experiential dimensions.
GPT-3.5 generated initial enthusiasm despite its limitations, with strong user control and cautious trust. GPT-4.0 consolidated technical performance while initiating emotional distance, as users expressed concerns about AI’s growing autonomy. GPT-4.5 emphasized relational empathy, temporarily reinforcing trust through better adaptation to user tone and intent, but generating frustration regarding fine-grained interaction control. GPT-5.0, despite its technical power and dynamic routing system, crystallizes a relational crisis, being perceived as colder and less personal, and amplifying the sense of lost control.
The study highlights that increasing technical sophistication can paradoxically degrade relational experience. It underscores a major challenge for AI development: users expect systems that are not only reliable and contextually relevant but also warm, controllable, and trustworthy. This research suggests that the future of language models will depend on their ability to strike these balances.
International Scientific Multidisciplinary Conference: AI for a Smarter Tomorrow - AI-SMART , September 25-26, 2025
Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons Attribution-Non-Commercial 4.0 License (https://creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial use, reproduction and distribution of the work without further permission.
