This study analyzes the diffusion of lexical innovations on Twitter to understand how the social network position of adopters impacts their success. Looking at both successful and failed neologisms, we categorize them into "changes" which become established and "buzzes" which decline over time. Using a corpus of 650 million French tweets, we reconstruct user networks and characterize adopters of innovations during different diffusion phases based on prestige, centrality, clustering, and external ties. In the early innovation phase, change and buzz adopters have similar peripheral profiles. During propagation, changes spread to prestigious, central individuals while buzzes do not, which predicts their eventual success or failure. By the establishment phase, changes reach highly central users with closer external ties. The results align with sociolinguistic theories about weak ties for innovation and strong ties for establishment. Additionally, logistic regression models based on early adopter profiles can predict the fate of innovations. This work sheds light on the diffusion dynamics of online lexical innovations and the crucial role of user network factors.