How OCR (reading scanned documents) works

Our reader handles three flavours of document, and you don't have to tell us which is which:

1. Text-based PDFs — the kind your bank emails you. Easy.

2. Scanned PDFs — image-only PDFs from a flatbed or office scanner. We read the image directly.

3. Photos — JPG, PNG, HEIC images from your phone camera. Snap, upload, done.

Why it works: instead of using old-school OCR (which struggles with crumpled receipts, glare, and faded thermal paper), we use a vision model — the same kind of AI that recognizes things in photos. It reads receipts the way a human does.

When it might struggle: very low-light photos, receipts where the ink has rubbed off, partially-cropped images. We surface anything we're not 100% confident about in the Review queue rather than silently miscategorize.

Our promise: if we can't read your document, we don't charge you for it.