Free PDF to Word

Worked example

Press "Try a sample". The file chip shows document-v1.pdf, a one-page PDF with a title and six short lines such as "Project: Atlas Rollout" and "Budget: $10,000". Press "Convert to Word". The caption reads "Reading the text layer...", the chip then shows "1 page", a green line says "✓ Your Word document is ready — document-v1.docx", and "Extracted text preview" lists the text, starting with "--- Page 1 ---". "Download" saves the .docx, while "Copy" and "Download .txt" sit above the preview. Tick "Add a "Page N" heading before each page" and the Word file gains a "Page 1" heading. A scan behaves differently. For a 12-page scanned PDF an amber panel says "Scanned pages found — run OCR?" and "None of the 12 pages have a text layer, so there is nothing to extract without reading the images." It adds "Reading them on your device takes about 6 minutes. Nothing is uploaded." and offers "Run OCR on 12 pages".

Which settings to use, and the limits

• Check whether the PDF has real text first. The tool reads the text layer that PDFs exported from a word processor or saved from a browser usually carry. A scan is only a picture of a page and has none. Pages with no text at all are counted one by one, so a mixed file is handled sensibly: you see "3 of 12 pages have no text layer. The rest were converted normally." and the Word file for the other nine pages is already ready to download. You choose whether to read the three scans. Nothing is downloaded or started until you press the OCR button.
• Pick the alphabet before you run OCR. Open the line "Alphabet on the scanned pages". It starts from the language of the site you are viewing, and Latin if that language has no match. A scan carries no text to detect an alphabet from, and the hint warns that a page read with the wrong alphabet "comes back as plausible-looking nonsense rather than an error". There are ten choices, from Latin and Cyrillic to Thai, Tamil and Telugu. Each shows its size, from 7.4 MB for Greek to 15.9 MB for Chinese and Japanese, and Latin is 7.7 MB. Hebrew is not available. The panel's total also includes a shared part used by every alphabet.
• Allow about 30 seconds a page, and use Cancel if needed. The estimate is 30 seconds per page. Under two minutes it is shown in seconds, in steps of 5. Longer runs use whole minutes, so 40 pages reads "about 20 minutes". Pages are read one at a time on-device, and the caption counts "Reading page 3 of 12...". Press "Cancel" and the button briefly says "Stopping...". Pages already read are kept and go into the Word file, and pages not reached are offered again. On a phone the model panel warns "On a phone this will be slow - expect several minutes, and keep this tab open." If the device cannot run the model, the OCR button stays disabled.
• Treat OCR text as a first draft. OCR output has no position information, so each line of the scan becomes its own paragraph, with no tables or headings. Every OCR page starts with an italic note, "[Read from a scanned page by on-device OCR — check it against the original.]", and the document header repeats a warning. The file name gets "-ocr" added, so a scan called letter.pdf downloads as letter-ocr.docx. A line under the result says "OCR read 3 scanned pages. Machine-read text can contain mistakes — check them against the original." Compare names, numbers and dates with the paper before you send the file on.
• Know what carries across from a text PDF. Wrapped lines are joined into paragraphs. A short line set at least 1.2 times larger than the body text becomes a Word heading, and one at least 1.6 times larger becomes Heading 1. Lines longer than 150 characters stay as body text. Columns become a Word table only when three or more rows in a row line up with the same number of columns. Images, fonts and page layout are not kept. The preview shows the first 8,000 characters, and the full text is in the download. "Add a "Page N" heading before each page" is off by default, because those headings are not in the PDF itself.
• Convert several PDFs in one go. Drop more than one PDF, or press "Add more files" with one loaded. The button then reads "Convert 3 PDFs to Word". Each file becomes its own .docx, and "Download all (3)" saves them in a ZIP called pdf-to-word. Batch mode reads only the text layer. A wholly scanned PDF fails its row with "Scanned PDF with no text layer. Open it on its own to read it with OCR." and a partly scanned one shows a note such as "2 scanned pages skipped". Files that are not PDFs are left out, with the message "Some files were skipped because they are not PDFs."

Common problems and how to fix them

It says "Something went wrong. Please try again."

The reader failed with an error the tool has no specific message for. Only a password error and an invalid-PDF error get their own sentence ("This file could not be read as a PDF — it may be corrupt."), and a document with zero pages gets the same corrupt message. Anything else, including the reader's "UnknownErrorException", lands here. Open the file in another PDF viewer. If it opens, print it to a new PDF from there, then convert that copy.

It says "No extractable text found in this PDF."

Every page is a picture and you pressed "No thanks" on the OCR offer, so there is nothing to convert. This is not a fault in the file. The text layer is empty. Press "Convert to Word" again to bring the offer back, choose the alphabet of the pages, and press "Run OCR on N pages". If you do not want OCR, the words have to come from another source, because a PDF with no text layer holds no words to copy.

It says "OCR ran, but found no readable text on the scanned pages."

Each page was drawn and read, and the recogniser returned no text. The plain case is a page that really is blank, so check the pages in a viewer. Pages that were read are not offered again, even when they came back empty. To try another alphabet, press "Convert to Word" (or "Convert again"), which restarts the check and shows the offer afresh. The hint under the picker says a wrong alphabet gives nonsense rather than an error, so also check the picker.

It says "The scanned pages could not be read. Please try again."

Something failed during the OCR pass. One page that makes no progress for 120 seconds is treated as stalled. The tool then discards that reader and starts a new one on the next try. If it repeats on a long scan, try a computer, or split the PDF so each run is shorter. A related message, "Page 4 is too large to render for OCR.", appears when a page would need a picture over 16,384 pixels on a side.

It says "This PDF is password-protected. Remove the password and try again."

The PDF is encrypted, so its text cannot be read yet. Under the message, a "Send this file to another tool" box offers "Unlock PDF". It passes the same file to the Unlock PDF tool inside this browser, and the box notes "Nothing is uploaded." Unlock it there, then add the unlocked file here. In batch mode the same message appears on that file's row.

How it compares

If you need the page to look exactly like the PDF, this is the wrong tool. It extracts the text and rebuilds paragraphs, headings and simple tables, which suits you when you want to edit or reuse the words. It does not keep images, fonts or the original layout, and it does not make a searchable PDF. Scanned pages are read by on-device OCR, so the text can have mistakes and you must proofread it. The first OCR run downloads a model once, 7.4 to 15.9 MB for the alphabet plus a shared part. Nothing needs installing, and the PDF is not uploaded.