Key answer
Upload success means receipt; correct full parsing and inclusion in the current processing scope need separate checks. Some tools retrieve relevant excerpts, some extract text, and some also process page images.
Full explanation
OpenAI Responses API's file-input documentation distinguishes these cases: vision-capable models receive PDF text and page images; non-PDF documents mainly yield text without embedded images/charts; its spreadsheet augmentation parses up to the first 1,000 rows per sheet and produces summaries and header information. These describe that interface, not universal file limits for all chat products. File-processing documentation
Check three kinds of omissions:
· Structure: do actual pages, contents, appendices and sheets match parsing? Did scanned pages produce blank text?
· Key content: sample the beginning, middle, end and decisive tables, footnotes and appendices; request concrete content and locations.
· Task coverage: for line-by-line review, all exceptions or all-record counts, require chapter, page or data-range coverage records. A few retrieved passages are not a complete review.
This file has 48 pages. Check readable coverage and content on pages 1, 24 and 48 and in the appendix table. Then review by chapter, recording processed ranges and missing, blank or truncated content. Do not describe retrieved excerpts as a full-document inspection.
Failing a spot check can mean parsing omissions or failure to use supplied content. *Lost in the Middle* found position alone affected multi-document QA and key-value retrieval. Check extraction before deciding to reupload, split processing or provide key pages. Original study
References
1. OpenAI, File inputs, official API documentation, How it works and How spreadsheet augmentation works
2. Liu et al. (2024), Lost in the Middle: How Language Models Use Long Contexts, TACL