Searchable PDF, Word or TXT: Which OCR Download to Pick

A run in PDF OCR finishes and three buttons appear. Download as Word (.docx). Download Searchable PDF. Download .txt.

They are not three wrappers around the same content. They are built from different things at different moments, and one of them deliberately ignores the corrections you just spent ten minutes typing.

The difference that catches everyone

Under the editable text box there is a line of small print: your edits are saved into the Word (.docx) and TXT downloads. The searchable PDF keeps the original scanned page image with a hidden text layer.

That is the whole thing, and it is worth spelling out.

The text box is the source for the Word and TXT files. Whatever is in it when you press the button is what lands in the download. Fix a misread name, delete a line of scanner noise, reorder a collapsed table — it all carries through.

The searchable PDF is assembled from something else entirely: the per-word recognition results captured during the OCR run. Each word came back with a bounding box, and the file is built by placing the original page image and then drawing every word invisibly at the coordinates it occupied. The text box was never consulted. Correcting a typo there does not correct it in the searchable PDF, and there is no way to make it.

If a specific word absolutely must be searchable and correct, that is a Word or TXT job.

What each file actually is

Pick by what happens next

Filing it, archiving it, or sending it to an office. Searchable PDF. It still looks like the stamped, signed, rubber-inked paper that came out of the scanner, which matters when the recipient's whole reason for wanting a scan is that it looks like the original. The difference from your input file is that six months later, a full-text search across the folder will find it.

Retyping, quoting, or reworking the content. Word. The point is to stop retyping, and a .docx opens in Microsoft Word, Google Docs and LibreOffice with every word live.

Feeding the text somewhere else. TXT. Pasting into a form field, an email, a spreadsheet column, a WhatsApp message to a colleague. No formatting to fight with.

Not sure yet. Download all three. Nothing is consumed and nothing is uploaded — the run already happened locally, and each button just serialises the result differently.

The searchable PDF's size problem

Worth planning for, because it surprises people who expected OCR to make a file smaller.

The searchable PDF does not reuse your original PDF. It is a fresh document built from JPEG snapshots of every page, rendered at the same high scale used for recognition — up to 4× the page's native size, capped at 3,500 pixels on the long edge, and encoded at high JPEG quality. A 12-page scan that arrived as 2 MB can leave as 8 MB or more.

If the portal you are uploading to caps attachments, run the result through the PDF compressor afterwards. One caution that matters here: its Low, Medium and High levels keep text selectable, while the Screen level re-renders every page as an image. Screen would flatten away the invisible text layer you just built — the file would still look right and would stop being searchable. Use one of the other three on a searchable PDF.

What the Word file is, and what it is not

The .docx is the recognized text laid out page by page, one original page per Word page. It is real, editable, selectable text.

It is not a layout reconstruction. Headings do not come back as Word heading styles. A three-column price list does not come back as a Word table. Logos, signatures and photographs are not carried over — they were never text, so they are not in the text output.

If that is what you actually needed, there is a different tool for it. PDF to Word rebuilds paragraphs, headings, lists and genuine Word tables, keeps graphics as inline images, and OCRs any scanned pages automatically as part of the conversion. It offers a few conversion modes and opens an in-browser editor on the result. Heavier, slower, and the right choice when the document has structure worth keeping.

Rule of thumb: OCR's Word export is for getting the words out. PDF to Word is for getting the document back.

The TXT quirk to know about

Every page is prefixed with a separator line:

--- Page 1 ---
Application for renewal of trade licence
...

--- Page 2 ---

Useful when you are cross-checking against the original. Unwanted when you are pasting into a single form field. Strip them with a find-and-replace on --- Page before you paste, or delete them in the editable box before downloading so the Word file loses them too.

The reset button

If you have edited the box heavily and want to compare against what the engine actually returned, a Reset to recognized text button appears next to the label as soon as your version differs from the original. It restores the raw output. It only shows up when there is something to reset, which is why you may not have noticed it.

A workflow that avoids most of the annoyance

1. Run the OCR and let all pages finish.
2. Download the searchable PDF first, before touching the text box. It ignores your edits anyway, and this way you have the archival copy safely saved.
3. Now read the text box properly and fix everything wrong.
4. Download the Word file for editing, or the TXT for pasting, or both.

Doing it in that order means you never sit there wondering whether your corrections made it into the file in your Downloads folder. They did — into two of the three.

If your source PDF already has a text layer and you skipped straight past OCR to get here, PDF text extract is the faster and more accurate route for the text itself. The rest of the document tools are listed on All Tools, and the recognition-specific ones sit under Scanners.