Pre-processing images and charts into structured text is an industry best practice, for reasons including token economy (images cost approximately 1,000–2,000 tokens each, while equivalent text requires only 100–300 tokens) and inference consistency.