Touch vs. Chat: Study shows tactile charts beat LLMs for BLV chart learning

A new study from Tingying He and colleagues at the University of Utah, together with collaborators, investigates how blind and low-vision (BLV) individuals can best learn to understand complex chart types. The researchers compared two learning formats: one providing a tactile chart, a textual explanation, and an LLM-powered chatbot; the other offering only the textual explanation and the same chatbot. In an interview study with 12 BLV participants, the team measured how well each approach helped users form mental models of chart structures (like scatterplots, heatmaps, network diagrams) and then apply that knowledge while exploring an unfamiliar dataset with alt text and an LLM.

The results from thematic analysis are clear: tactile templates significantly supported BLV participants in building accurate mental models of chart types, which subsequently improved their ability to use the LLM for data exploration. Participants who had access to tactile charts could ask more targeted questions and interpret spatial layouts better during the exploration phase. In contrast, the text+LLM-only approach showed clear weaknesses for spatial reasoning tasks — users often misdirected questions or struggled to form coherent mental images. The study has practical implications for inclusive data visualization tools, suggesting that pairing tactile aids with conversational AI offers a more effective learning pipeline than relying on LLMs alone for chart education among BLV individuals.