benchrWhen it breaks
It says it read my PDF, but it only answered about the first few pages
Also asked as ChatGPT won't read my PDF · Claude ignores the rest of the document · the AI summarised only the beginning · it missed the table on page 200
Cause documented by a vendor or a published paper Not tested by benchr First logged Last checked
What is actually happening
The file was accepted and part of it was read. The model then answered from the part it had, without telling you the rest never arrived.
Why
- Document upload has two separate ceilings — a file-size limit and a page limit — and a file can pass one while failing the other.
- Some paths extract text only. A chart, a scan or a photographed page carries no text to extract, so it silently contributes nothing.
- Long documents can also fit the upload limit and then overflow the context window once the surrounding conversation is added.
The quick fix
Ask it one question it can only answer from the end of the document — a figure from the last table, the final section heading. If it cannot, the file did not arrive whole.
The real fix
Split the document at chapter boundaries and process each part separately, then ask for the summary across the parts. On the API, send the file as a document rather than pasted text so page images are preserved for charts and scans.
Step by step
- Check the real page count and file size against the limit for the interface you are using.
- Run the end-of-document test above before trusting any summary.
- If pages are scans or photographs, say so in the prompt — that changes which path handles them.
- Split at natural boundaries, not at page counts, so no answer straddles a cut.
- Ask for a page or section reference with every claim, so a missing region shows up as a missing citation.
If you are seeing an API error
Grounded in
- Read a 600-page PDF including its chartsThe documented page and size ceilings, read from the vendor guide.