Found and fixed a scary bug: my AI resume tool was inventing fake analyses for unreadable PDFs

I built Rezycle.me a couple weeks ago — a free AI resume scan that flags automation risk and suggests career pivots.

This week I found a bug that honestly worried me. If someone uploads a resume PDF that's actually a flattened image (no real text layer — happens with a lot of visual/Canva-style CV templates), the AI was still generating a full result. Score, keywords, "evidence quotes" from the resume — except there was no resume text at all. It was making the whole thing up, confidently, for a person who doesn't exist.

Found it because a friend tested it with his real CV and got career pivot suggestions that had nothing to do with his actual background (he's in hospitality, it suggested "Data Analyst"). Dug into it, reproduced it with an empty-text payload, confirmed: any blank/near-blank input got a fully fabricated but confident-sounding analysis back.

Fixed now — if the text extraction comes back empty or too short, it tells you plainly instead of guessing, and suggests re-uploading as .docx or checking the PDF actually has selectable text.

Small thing, but it's exactly the kind of bug that quietly erodes trust if nobody catches it. Free scan's still up if you want to try it: rezycle.me