When AI Lies with Charts: Misleading Infographics in Text-to-Image Generation

While these systems offer substantial efficiency gains for data communication, their outputs risk being adopted uncritically by users who lack the data visualization literacy needed to identify misleading elements — a competency that the systems themselves do not possess.
We present a systematic analysis of 100 AI-generated infographics produced with Nano Banana Pro (Google) across 20 thematic categories, annotated by a human expert and an AI model using the taxonomy by Lo et al. [14]. We find that 99 out of 100 infographics contained at least one misleading element, with recurring patterns spanning the full visualization pipeline: fabricated data, inconsistent encodings, inappropriate chart types, incomplete charts, and misleading framing. Compared to the web-scraped corpus of Lo et al., visual design errors remain equally prevalent, while the Garbage-In and Inconsistency categories stand out prominently — suggesting that AI-generated visualizations do not merely reproduce classical chart crimes, but additionally amplify issues of data provenance and internal inconsistency.