OCR
Scanned PDFs
What to expect from OCR quality on scanned and image-heavy documents.
OCR runs for every document
ChatPDF sends every upload through Mistral OCR 4 — not only scanned files. For clean, text-based PDFs this is transparent. For scanned or image-based PDFs, OCR quality becomes the main factor in how well retrieval and citations work.
Quality expectations
- Clean scans produce better chunks and stronger retrieval.
- Skewed, low-resolution, handwritten, or table-heavy pages may need manual review.
- OCR output should be treated as extracted evidence, not a perfect copy of the source.
Required configuration
OCR depends on a Mistral API key. Uploaded PDFs and OCR JSON artifacts are stored in Convex file storage. See Configuration.