The reading pile is long and you need the key points of pages, including ones that are mostly charts or slides.
The summarize action accepts a capture as input. A full-page screenshot becomes a summary of the whole page; a region capture of a chart becomes a description of the trend and, where readable, the underlying numbers. Because the model sees the image itself, it can summarize visual content with no text equivalent in the DOM: dashboards, charts, slide decks, and image-heavy pages. For long documents, full-page capture plus summarize is often faster than reading.
Capture the page, run summarize, and read the result in the panel. For extremely long pages, remember the 60-frame capture limit and capture sections instead.