What Grok knows about health? Analysis of posts on X
ANÁLISE DE POSTAGENS NO X
Keywords:
generative artificial intelligence, scientific information, Grok, health, digital platformsAbstract
This article analyzes the mediation of generative artificial intelligence tools in the circulation of scientific information, focusing on Grok’s responses about health and vaccination on X. The study seeks to understand how the AI responds to users in online health debates and what types of evidence are mobilized in its answers. To this end, posts published between June 2025 and May 2026 were analyzed using data from the Online Panel for the Detection of Anti-Vaccine Narratives project. The research adopts a quantitative-qualitative approach, combining automated textual analysis, source identification through regular expressions, semantic grouping using SBERT embeddings and HDBSCAN clustering, as well as a qualitative analysis of a sample of posts about dengue. The results indicate that Grok tends to provide answers that convey neutrality and certainty, frequently mobilizing official, scientific, journalistic, and generic sources. However, the analysis reveals low transparency in the indication of consulted sources, a lack of direct links, and limited hierarchization among different types of evidence. The study concludes that, although Grok contributes to summarizing information available online, its performance may produce only apparently scientific information, as it equates sources of different natures and does not adequately explain the methodological paths that support its responses.