Problem

The rapid proliferation of generative AI (GenAI) chatbots is fundamentally transforming digital interactions, offering societal benefits while simultaneously posing risks of widening digital divides. These divides stem from uneven adoption rates and a low awareness among users regarding the limitations of these powerful tools. Despite a global surge in GenAI use, there has been a critical lack of comprehensive empirical research mapping its adoption, usage patterns, and literacy specifically within the Italian context. This gap is particularly significant given Italy's documented challenges in general digital competence, which could exacerbate the risks associated with uninformed GenAI adoption and use.

Method

This study employed a comprehensive empirical approach based on newly collected survey data from 1,906 Italian-speaking adults, encompassing both GenAI adopters and non-adopters. The survey, distributed between May and August 2025, gathered rich demographic information and aimed to address four key objectives: examining GenAI adoption in comparison to other longstanding language technologies (LTs), documenting GenAI usage patterns (activities, intents, contexts), assessing awareness of benefits and risks (literacy gaps), and exploring sociodemographic divides in adoption and usage. Quantitative statistical analyses included logistic regression to predict GenAI chatbot adoption and ordinary least squares (OLS) models for intent-specific usage frequency, controlling for age, gender, geography, socioeconomic status, education, prior LT experience, and literacy. Qualitative insights were derived from open-ended responses, analyzed using the BERTopic modeling tool to identify prevailing themes and user attitudes.

Results

The findings revealed widespread GenAI chatbot adoption in Italy (80.5%), surpassing voice assistants and approaching the near-universal use of machine translation, with ChatGPT dominating the platform landscape (74.5%). GenAI is significantly displacing other technologies, notably assisted writing tools (24.5% partial, 19.8% complete replacement) and web search (34% partial, 5% complete). This shift is primarily driven by perceived convenience (38%) and flexibility/personalization (30%), rather than higher accuracy or reliability. Learning and Information Retrieval emerged as the most frequent usage intents, with activities split almost equally between work/study (52%) and personal contexts (48%). Concerning high-risk applications, 24.3% of users sought medical advice and 10.2% used GenAI for emotional support, with older users (65+) disproportionately seeking medical advice and fact-checking. Users showed a strong preference for written interactions (85%) due to perceived control and precision, while Italian was the primary language of interaction (95%), followed by English (52%) for work or perceived reliability. Overall GenAI literacy was found to be low, even among users, with 39.5% reporting encountering errors such as factual inaccuracies (hallucinations), gender/racial bias in image generation, and linguistic bias. While awareness of bias was relatively high, the ability to spot errors was low. Significant digital divides were identified: adoption decreased sharply with age, women were half as likely to adopt as men, and higher education and socioeconomic status correlated with increased adoption. Prior LT experience and literacy were strong positive predictors of adoption. Crucially, the gender divide persisted in usage frequency across all intents, even after controlling for literacy, and was more pronounced in older generations.

Implications

The study's findings underscore the urgent need for targeted educational initiatives to enhance GenAI literacy among Italian users. Given the widespread and convenience-driven adoption of GenAI, often for sensitive tasks like medical advice and fact-checking, and in light of low literacy levels and user struggles in identifying errors, structured learning opportunities are essential to protect vulnerable users from misinformation. The persistent gender divide in both adoption and usage, which is not fully explained by technical competence or socioeconomic status, highlights the necessity for further investigation into underlying social and cultural barriers to equitable participation. Policymakers, educators, and developers must collaborate to ensure equitable access to GenAI while guiding users towards informed and responsible adoption, thereby preventing the exacerbation of existing digital inequalities and potential biases in future model training data.