AI OCR for difficult scans
Run OCR, let a vision model correct what it misread, and download a searchable PDF with better recognized text.
Drop a file 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
AI OCR, explained
AI OCR is OCR with a second pass. Standard recognition runs first; then a vision model looks at each page image alongside the recognized text and corrects words, numbers and table cells it misread. The result is the same kind of output as OCR PDF: an invisible text layer behind an unchanged page image.
Read more
It earns its cost on documents where standard OCR struggles: handwritten notes and forms, faint or low-resolution scans, faxes, dense tables of figures, and pages that mix fonts and stamps. On a clean printed letter, OCR PDF does as well for less.
AI OCR uses OCR pages for the first pass and AI credits for the correction pass; the credit cost for your pages is shown before you start. Page images and recognized text are sent to OpenAI's API under OpenAI's API data-usage terms, which by default exclude API data from model training. For crooked or noisy pages, Restore scanned PDF cleans the image first; to change text on the page afterwards, use Edit scanned PDF.
How to do it
AI OCR for difficult scans in 4 steps
1A scan: only pixels
2Recognize the text
3Text you can search and select
- 1
Upload the scan
PDF, JPG, PNG or TIFF.
- 2
Choose languages
Select the languages on the pages, as for standard OCR.
- 3
Run AI OCR
Recognition runs, then the correction pass. The cost in OCR pages and AI credits is shown first.
- 4
Check and download
Search for a few words you can see, especially numbers, then download the searchable PDF.
Example
Before and after
A faxed 2-page lab order with a handwritten patient note and a table of test codes, where standard OCR returns broken words and merged columns.
The same pages, visually unchanged, with the test codes searchable as separate values and the handwritten note recognized as text you can copy and check.
Settings & formats
What you can control
- Languages
- The languages printed or written on the pages.
Input
Output
Good to know before you start
- Corrections are model output and can still be wrong, particularly on ambiguous handwriting and digits; check important values against the image.
- Table structure improves in the text layer, but the output is a searchable PDF, not a spreadsheet; use PDF to Excel for that.
- It costs both OCR pages and AI credits, so it is more expensive than standard OCR.
If something goes wrong
If something goes wrong
- Handwriting is still misread
- Very cursive or faint writing may be beyond recognition. Restore the scan first for better contrast, and check the result word by word.
- The cost seems high for my file
- Run AI OCR only on the hard pages and standard OCR PDF on the rest.
AI OCR: questions people ask
What's the difference between AI OCR and OCR PDF?
OCR PDF uses the Tesseract engine alone. AI OCR adds a vision-model pass that corrects misreads, which helps most with handwriting, tables and poor scans. It also costs AI credits.
Does AI OCR change how the page looks?
No. Like standard OCR, it only adds an invisible text layer; the scanned image is untouched.
Is AI OCR the same as Restore scanned PDF?
No. Restoration cleans up the image itself (straightening, noise, contrast). AI OCR improves the recognized text. They work well together: restore, then OCR.
What is sent to the AI provider?
Page images and the first-pass text, to OpenAI's API, under OpenAI's API data-usage terms, which by default exclude API data from training.
Can I edit the text afterwards?
Select, copy and search it straight away. To change words on the page, open the result in Edit scanned PDF.