Invoice OCR for scans and photos, with the math checked
Turn paper invoices you've scanned or photographed into checked, editable data without retyping them.
Drop your files hereLet go to start
or paste with Ctrl/⌘+V
Try it 5× without an account. A free account includes 20 AI credits a month. See plans
What it does
Invoice OCR, explained
Invoice OCR here is two steps. First, text recognition (OCR) puts an invisible text layer on each scanned page or photo, so the characters become text. Then the AI reads that text as an invoice: supplier, number, dates, line items, tax and totals, each with the page and quote it came from. Photos (JPG, PNG, HEIC) are turned into a one-page PDF in your browser before upload; scanned PDFs are recognised only on the pages that have no text yet.
Read more
Plain OCR is where invoice data usually goes wrong: an 8 read as a 3, a comma lost from 1,250.00. That's why every invoice is then checked with arithmetic: quantity × unit price per line, lines against the subtotal, and subtotal − discount + shipping + tax against the total. A misread digit almost always breaks one of those sums, and the page shows which one and by how much, next to the quote, so you can correct it.
Choose the language printed on the invoices before reading (English, Spanish, French, German, Japanese and many more), since recognition is tuned per language. Digital PDFs skip OCR entirely and are read as they are.
Recognition uses OCR pages (20 a month on the Free plan) and the reading uses from 1 AI credit per page read. To make a scanned invoice searchable without extracting data, OCR PDF is enough; to clean up a dark or skewed scan first, try Restore scanned PDF.
How to do it
Invoice OCR for scans and photos, with the math checked in 4 steps
1Your PDF
2Ask or summarize
3Answers that cite pages
- 1
Add scans or photos
Flat, well-lit photos or 300 dpi scans read best. Up to 10 per run.
- 2
Pick the language
The language printed on the invoices; English is added as a fallback for other languages.
- 3
Read
Text is recognised on pages that need it, then the AI reads the invoice fields and lines.
- 4
Check, fix, export
Correct anything the math check flags, then download Excel, CSV or a bill import.
Example
Before and after
A phone photo of a paper invoice, slightly angled, with a printed table of four lines and a total of 1,248.00.
Four lines and the totals read from the recognised text, with one line flagged: 3 × 112.00 is printed as 336.00 but was recognised as 386.00. Typing 336.00 back in makes the invoice add up.
Settings & formats
What you can control
- Photos
- JPG, PNG and WebP; HEIC where your browser can open it (Safari can). Each photo becomes one page.
- Scanned PDFs
- Only pages without a text layer are recognised; digital pages are read directly.
- OCR language
- Dozens of languages, with English added as a fallback when you choose another.
- Evidence
- Values whose quote can't be found in the recognised text are marked to check.
Input
Output
Good to know before you start
- Handwriting, stamps over numbers, very small print and faded thermal paper are often not recognised; type those values in.
- Photos must show one invoice, flat and in focus. Crop out the background if the page is small in the frame.
- HEIC photos only work in browsers that can open them (Safari can); elsewhere, export them as JPG first.
- OCR and reading are charged separately: OCR pages for recognition, AI credits for the pages read.
If something goes wrong
If something goes wrong
- “No text could be recognised”
- Retake the photo in better light, flatter and closer, or rescan at 300 dpi. Very dark or blurred images give the recognizer nothing to work with.
- Many values are marked “check”
- The recognised text and the AI's quote differ, which usually means a poor scan. Compare with the page (Open the PDF) and correct the values; the totals check shows which ones matter.
Invoice OCR: questions people ask
Is this OCR or AI?
Both. OCR turns the image into text; AI then picks out the invoice fields from that text. The final check that totals add up is plain arithmetic.
Which languages can it read?
Text recognition supports dozens of languages, including English, Spanish, French, German, Italian, Portuguese, Dutch, Polish and Japanese. Pick the one printed on the invoice.
Do I need to convert photos to PDF first?
No. Photos are turned into a one-page PDF in your browser automatically before they're uploaded.
Does it work on digital PDFs too?
Yes. Pages that already have text skip recognition and aren't charged OCR pages.
What does it cost?
Reading uses from 1 AI credit per page read (the Free plan includes 20 AI credits a month, Pro 500); scans and photos also use OCR pages (20 a month on Free). Exports are free and can be repeated.
Where do my scans go?
Page text is sent to our AI provider, OpenAI, under API terms that exclude it from model training. The files are saved to your PDFBetter documents, where you can delete them.