The Impact of AI-Driven Information Technology on Social Media Communication and Consumer Trust
DOI:
https://doi.org/10.63125/2jq2bx46Keywords:
Artificial Intelligence, Social Media, AI-Mediated Communication, Consumer Trust, Personalization, Generative Content, Chatbots, Privacy, Survey Methodology, SimulationAbstract
Artificial intelligence now mediates most brand communication on social media: recommendation models decide what consumers see, generative models write part of it, conversational agents answer them, classifiers moderate them and labels mark what is synthetic. This article asks how these AI-driven information technologies affect consumer trust and what a survey study of the standard design can learn about it. It develops the Algorithmic Communication Trust Framework, in which five exposure constructs (personalised recommendation, AI-generated content, conversational agents, algorithmic moderation and AI labelling) act on three perceptions of AI-mediated communication (communication quality, authenticity, and manipulation and surveillance), which act on trust, which carries their effects to engagement and purchase intention, with AI literacy and privacy concern as moderators. Because no survey observes the true effect, the framework is evaluated on a transparent synthetic testbed: a pre-specified data-generating model with parameters anchored to published findings, ordinal measurement through 48 Likert items and a method factor. The pipeline of a recent template study (reliability, factor analysis, Kruskal-Wallis tests, hierarchical regression, bootstrapped mediation and moderation) is applied to synthetic samples of 130, 500 and 2,000 and repeated in a Monte Carlo evaluation. At n = 500, thirteen of eighteen directional hypotheses are recovered with the designed signs: the three perceptions are the strongest predictors of trust, trust mediates ten of eleven indirect paths, synthetic content lowers trust through two routes and labelling raises it through both. Coefficients are attenuated by 15 to 30% and interactions by more; at n = 130 only ten hypotheses survive, significant estimates are exaggerated up to 2.7-fold and the factor structure replicates in 36% of samples; interactions of 0.10 need 1,000 to 2,000 respondents;


