Uploaded a 30 page PDF to Gemini, asked questions about it, and it's confidently answering with stuff that isn't in the document. Like it never read it.
Uploaded a 30 page PDF to Gemini, asked questions about it, and it's confidently answering with stuff that isn't in the document. Like it never read it.
Bet it's a scanned PDF. Open it and try to select some text with your mouse. If you can't, it's an image, not text, and Gemini has to guess at it. With Hindi or mixed-language scans, or a bad scan, the guessing goes badly and it fills the gaps with whatever sounds plausible.
Fix: run it through OCR first. Quickest way I know, upload the PDF to Google Drive, right-click, open with Google Docs. That converts it to text. Then upload the Doc version.
Other things to check, is the file too big, is it password protected, and if you're doing this inside Drive, is the file actually in your Drive or just a link someone shared once.