Prompt
Do I need bounding boxes from OCR or just extracted text?
Latest observation
Oct 24, 2026
OpenAI APIWeb search: off
It depends on what you’re trying to do:
-
Just extracted text is enough if you only need:
- search/indexing
- summarization
- translation
- NLP tasks
- plain text analytics
-
Bounding boxes from OCR are useful if you need:
- layout-aware understanding
- preserving reading order
- highlighting text in the document
- tables/forms/key-value extraction
- associating text with positions, images, or fields
- reconstructing document structure
- deduping/merging OCR with existing layouts
Rule of thumb
- If your downstream task treats the document like plain text, you probably only need extracted text.
- If your downstream task cares about where text appears on the page, you need bounding boxes.
If you tell me your exact use case, I can recommend the right output format.