FILE WORKFLOW GUIDE
How to turn a scanned PDF table into Excel
Extract one scanned table page with AI, compare the cells with the source, and export editable XLSX. Beta access is required; this is not a whole-document conversion.
Choose table extraction or text OCR
A scanned PDF contains pictures of pages. To put a table into Excel, the tool must identify rows, columns and cell values in the selected page image. If you only need paragraphs, PDF OCR is a better match and exports text rather than a spreadsheet.
Prepare a readable page
Use an upright scan with the entire table visible. Avoid clipped headers, glare and tiny text. Printed English and Chinese are supported, but handwriting, merged cells and complex layouts need extra review. A clear input improves the conditions for recognition; it does not guarantee accuracy.
From source page to XLSX
- Open PDF to Excel, add one PDF, choose the page number and preview the source. PDFs are limited to 25 MB and 100 pages.
- Enter an invitation code and approve sending that page through OpenRouter. FileUnbox sends a rendered page image, not a request to process every page.
- Compare the extracted headers and every important cell against the original. Edit omissions and misread values before export.
- Download XLSX, CSV or JSON. For another page, run a separate extraction; cross-page tables are not automatically combined.
Review the values that are easy to miss
| Example source value | What to verify |
|---|---|
| 00007 | All leading zeros are present |
| 1,280.50 | Thousands and decimal separators |
| 0 / blank cell | Zero is not replaced by an empty value |
| ¥ or $ | Currency and minus signs match the source |
These are review examples, not measured recognition results. XLSX values are exported as text to preserve identifiers and leading zeros. If you need calculations, convert only the verified numeric columns in your spreadsheet application. CSV handling also depends on the application's import settings.
Beta limits and failed extractions
AI can omit content, misread cells, or return a result the tool cannot export. The October 28 launch tests included an image-table case that was rejected for inconsistent output. Do not treat this beta as an unattended workflow for important records. Keep your source and review all results.
If no usable table is returned, check the preview, try a clearer authorized source, or use text OCR if the page is mostly paragraphs. There is no searchable-PDF export, automatic multi-page batch or guarantee that a retry will succeed.