Anyone who has converted a PDF to Word knows the result. The document looks right until you start typing. Then every line turns out to be a separate text box, nothing wraps, and fixing one word means nudging half the page. Scanned documents are worse still: usually there is no text at all, only a picture of the page.
We have reworked how conved.com converts PDF to DOCX, DOC, ODT and RTF. You now get a document you can actually edit, even when the PDF is a scan.
Why a PDF is not a Word document in disguise
PDF was designed to make a document look the same everywhere. So it does not store text the way you wrote it. It stores what gets printed: characters and fragments of lines, each at an exact position on the page. That three lines form one paragraph, that a bold line is a heading, or that the lines around some numbers make a table is usually not recorded anywhere in the file.
PDF can carry the logical structure of a document ("tagged PDF"), but only if the program that creates the file adds it. The Library of Congress format description notes that many PDFs made by "printing to PDF" have no such structure. A converter then has to work it out from the page layout.
Many converters skip that step and drop each line into a text box at its original position. The output looks like the original and is close to useless for editing. The opposite shortcut is plain text, which loses paragraphs, tables and images.
What the conversion does now
- Lines are joined back into paragraphs, so text reflows when you edit it.
- Tables drawn with ruled lines become real Word tables, and images stay in place.
- If a PDF cannot be rebuilt this way, you still get a document that looks like the original (lines in text boxes) instead of an error.
Scanned documents: read first, then rebuild
In a scanned PDF the page is just an image. The converter detects this and runs text recognition (OCR) first. For Latin script it looks for English, German, Czech and Slovak. Based on the script of the first page it also handles Cyrillic, Greek, Arabic, Hebrew, Japanese, Chinese and Korean.
Recognised text is not a document yet. OCR returns individual words with their positions, and in justified text with wide gaps it is easy to end up with one word per line. That is exactly what we saw in a scanned set of meeting minutes: more than half of the "paragraphs" had one or two words. So the text is rebuilt from the page layout:
- lines are joined into paragraphs, and words hyphenated at the end of a line are put back together,
- headings and list items stay on their own,
- a paragraph that continues on the next page is not split,
- page numbers in headers and footers are removed.
What to expect
A PDF exported from Word or another word processor usually converts close to the original. If you still have the original DOCX, though, it is always the better starting point.
From a scan you get clean text in paragraphs, without images, stamps or font formatting. Text recognition covers up to 30 pages; a longer scan is converted without it. Accuracy depends on the source: handwriting, stamps and skewed pages can introduce errors, so read the document through before you rely on it.
Try it on the PDF to DOCX page. Conversion to DOC, ODT and RTF works the same way, and the PDF format page has more background.