What makes a floor plan AI-readable?

AI keeps misreading housing floor plans. My hunch: the text annotations on a plan are more machine-readable than its lines.

The idea

AI models keep misreading housing floor plans, and nobody seems sure which representation fixes it. My hunch: reading the text on the plan will work better than reading its lines. Floor plans carry annotations like room names, dimensions, and unit labels, and OCR on those may be more AI-readable than the geometry itself.

Open questions

Does the hunch survive a simple benchmark: lines-only vs annotations-only vs both?