Scanned PDF to Word converter that gives you an editable document
A scanned contract, an old report or a document someone photocopied is just a picture of text. Normal PDF to Word tools give you a Word file with images inside, so nothing is editable. Your todo.is agent runs OCR on every page, rebuilds the paragraphs, headings and tables in Word and marks anything it could not read clearly.
The prompt
- Convert the attached scanned PDF into an editable Word document: [ATTACH THE SCANNED PDF]. The text is in [LANGUAGE]. Run OCR on every page, straighten tilted pages, and rebuild paragraphs, headings, numbered clauses and tables as real Word text and tables, not images. Keep [KEEP AS IMAGES] as images. Highlight in yellow any word the OCR is not sure about. Keep the page breaks [PAGE BREAKS]. Send the .docx and tell me which pages were hardest to read.
What to change
- [ATTACH THE SCANNED PDF]: Attach the PDF scan, up to 50 MB. Photos of pages work too.
- [LANGUAGE]: E.g. "English", "French" or "English with some Spanish names".
- [KEEP AS IMAGES]: E.g. "signatures and stamps", "the logo" or "nothing".
- [PAGE BREAKS]: E.g. "as in the original" or "let the text flow".
Example result
- Scanned PDF to Word: Lease_Agreement_2019.pdf
- 18 scanned pages · 300 dpi · English · converted to an editable .docx
- What your agent did
- • Cleaned the scans: straightened 5 tilted pages, removed gray shadows at the page edges, boosted faded text on pages 14 to 16
- • OCR: read every page, about 7,900 words
- • Rebuilt structure: the title became a Heading 1, clause numbers 1. to 24. became a numbered list, sub-clauses (a), (b), (c) indented
- • Tables: the rent schedule (page 3), the inventory list (pages 15 to 16) and the contact table became real Word tables with editable cells
- • Kept as images: the two signatures and the landlord's stamp on page 18, placed where they were
- • Page breaks: kept as in the original, so page numbers still match the paper copy
- Uncertain words (highlighted yellow)
- • Page 6: "forty [fourteen?] days" in clause 9, faded print. Please check against the paper
- • Page 11: a handwritten note in the margin was read as "agreed 2/3 – J.M." and added as a Word comment
- • Page 16: 9 inventory items with smudged text
- Limits
- • Fonts are a close standard match, not the exact original typeface
- • Handwritten notes are added as comments, not merged into the text
- • OCR quality depends on the scan; numbers and dates on faded pages are worth a check
- Next
- Ask for a tracked-changes version after you edit, a PDF copy, or a translation of the text.
How to do it with todo.is
- Copy the prompt and fill in the language and what to keep as images.
- Attach the scanned PDF in todo.is, or send it to your agent on WhatsApp straight from your scanner app.
- Your agent cleans the pages, runs OCR and rebuilds the document in Word.
- Download the .docx and check the yellow words. Reply with page numbers if anything needs another look.
Tips for a better result
- Scan at 300 dpi in grayscale or black and white. It gives clearer OCR than color photos.
- Always check numbers, dates and names in legal or financial documents. OCR errors there are small but costly.
- If the scan mixes languages, list them. The OCR uses a language model for each.
- Only need the text, not the layout? Ask for plain text or Markdown instead. It is faster to clean up.
scanned PDF to Word converter: FAQ
- Can I edit a scanned PDF in Word? Not directly, because a scan is an image. Running OCR turns the image into real text, which your agent then rebuilds as an editable Word document.
- How accurate is OCR on scanned documents? Clean, printed scans are usually read very accurately. Faded, skewed or handwritten pages have more errors, which is why uncertain words are highlighted.
- Will the Word file look like the original? Headings, paragraphs, lists and tables are rebuilt. Exact fonts and spacing can differ a little, especially with complex layouts.
- Is my document private? Your file stays in your own agent workspace and only your agent uses it.
JavaScript is required to use the todo.is app.