There are three methods for browser information retrieval ranked from least to most token-expensive: reading the accessibility tree uses the fewest tokens, followed by HTML, and finally screenshots or direct pixel clicking