PDFCellsConvert a scanned PDF to Excel with OCR
Open the scan in PDFCells: pages without a text layer are read with free OCR in your browser, cells the OCR was unsure of are flagged, and the totals are checked before you download Excel. For skewed, faint or photographed pages you can choose AI instead.
Steps
- Go to pdfcells.com and choose the PDF, or drop it on the page. Drop several files to open them as a batch. A password-protected PDF asks for its password. Images open the same way: JPG, PNG, WebP, GIF (first frame), HEIC from an iPhone and multi-page TIFF. Paste a screenshot with Cmd/Ctrl+V.
- The pre-check marks scanned pages and recommends OCR. A page whose text layer is garbled is marked “broken text” and read with OCR too.
- Review the grid. Cells OCR read with under 70% confidence are flagged, as are totals and balances that don’t add up.
- If a page looks weak (no table found, a failing check, or many unsure cells), PDFCells offers to read it again with a second, experimental OCR engine, and you can switch between the two results.
- Download Excel.
Which reading method to use
| Method | Runs | Cost | Best for |
|---|---|---|---|
| PDF text | In your browser | Free | Born-digital PDFs with a text layer: exact characters, no OCR errors |
| OCR (tesseract.js) | In your browser | Free | Clean scans, screenshots, straight photos |
| Experimental OCR (PP-OCRv6) | In your browser | Free | Offered as a retry on weak pages |
| AI (Claude) | One page at a time on our server and Anthropic | 25 free pages a day signed in, then credits | Skewed, low-quality or photographed pages, irregular tables |
On our own synthetic test corpus, the default OCR read 94.4% of table cells exactly across 44 pages; re-reading every page with the experimental engine’s retry gave 96.1%. Scans are the hardest part: see the accuracy page for the breakdown and what those numbers don’t tell you.
Tips for better scans
- Scan at 300 dpi in greyscale rather than a low-resolution colour photo.
- Photograph the page flat and square-on; small tilts are levelled automatically.
- Mark the table area if the page has handwriting or stamps around it.
Free OCR is not perfect on faxes and very poor scans. The checks catch many misreads (a wrong digit breaks the running balance or a line total), but not every one, so look at flagged cells before you rely on the numbers.
Questions
Is the OCR done online?
No. The OCR engines run in your browser, so a scanned page isn’t uploaded. Only pages you send to AI leave your device.
Can I convert a photo of a table to Excel?
Yes. Open the photo (JPG, PNG, HEIC and others) or paste a screenshot. It is read like a scanned page.
What languages does the OCR read?
The in-browser OCR uses an English model, which reads numbers and English text well. For accented or non-Latin text, use AI for that page.