Variation Summary

A concise presentation format for AI-generated image descriptions that explicitly organizes information into three categories: agreements (claims supported by all or most models), disagreements (claims where models conflict), and unique mentions (information provided by only one model). This format was the most preferred by BLV users (ranked first by 11 of 15 participants) because it provides a quick overview of reliability—areas of agreement signal higher confidence, while disagreements and unique mentions flag potentially unreliable information. The variation summary helps users efficiently prioritize which parts of a description to scrutinize further.