A Google patent filed in 2023 and published in April 2026, analysed by Search Engine Land, establishes that AI answer engines evaluate images as a primary citation signal — in some cases before text evaluation. AI systems prefer sources where brand packaging text is legible, visual identity is distinctive, and machine parsing of the image is unambiguous. Generic stock photography creates a structural disadvantage: images that could belong to any brand in a category reduce the AI's confidence in source identity attribution. Two GEO metrics are emerging from practitioners working from the patent's implications: "ownership rate" — whether an AI system reliably associates a given image with the correct brand — and "legibility rate" — whether the brand's visual elements are parseable by machine vision at the resolution at which they are indexed. Both are now components of a complete AI visibility programme, alongside the content and structural signals that GEO has addressed in its first iteration.