PDF to Text OCR for Scans, Forms, and Docs
How to get editable text out of a scanned PDF
Extract text from scanned PDF pages with OCR and export it as plain text or Markdown. Select the language and page range to focus on the relevant part of the document. Clear, upright scans usually produce more reliable text than blurred or skewed pages. Review names, numbers and table structure before reusing the result; OCR does not guarantee an exact transcription.
Cost: Free options are available. Pro options or use beyond the free allowance may require credits; see the guide below.
How to convert a scanned PDF to text
Upload one PDF
Click Choose PDF or drop a PDF of up to 200 MB onto the preview. PDFs you already uploaded in other PDF tools are listed under Uploads. Each job reads one PDF.
Find the pages you need
Page through the preview. The counter shows the position in the file, for example
4 / 12. Enter those positions in Pages (optional), such as1-3,7, or leave the field empty to read every page.Pick the document language
Document Language starts on Auto-detect, which suits most scans, including pages that mix languages. Choose a specific language when a first run misreads characters.
Choose the output format
Plain Text (.txt) is the default. Switch Output format to Markdown (.md) when headings, lists and tables should stay recognisable as structure.
Check the units and start
The box under the settings shows your selection, for example
Selected pages 12/12 · units 2. Click Extract text and keep the page open until the result appears. A job can run for up to 30 minutes.Proofread, then copy or download
Read the Extracted text panel. Copy puts the text on your clipboard, Download text saves
pdf-ocr.txtorpdf-ocr.mdin UTF-8, and DOCX saves a Word document you can edit. These results are also saved in the shared file manager.
Language, output format and pages
The Settings panel has three options. None of them changes your PDF.
- Document Language: Auto-detect
- The default. Your pages are read by GLM-OCR, a dedicated OCR model from Z.ai that works out the language and script by itself. It takes no language hint, which makes it a good fit for documents that switch between languages.
- Document Language: a specific language
- Choose English, German, Spanish, French, Italian, Portuguese, Russian, Japanese, Chinese, Arabic or Hindi. The pages are then read by a Google Gemini model that uses the language, its script and its reading direction as a hint for ambiguous characters and word boundaries, such as Han characters shared by Chinese and Japanese or right-to-left Arabic. Text in other languages is kept as written.
- Output format: Plain Text (.txt)
- Returns text with line and paragraph breaks and no Markdown markup. Bullets become
-lines. Tables the engine recognises become one line per row with tabs between the cells, so a pasted table splits into spreadsheet columns. Good for emails, forms and search indexes. - Output format: Markdown (.md)
- Keeps the structure as markup:
#headings, lists, emphasis and tables. Choose it for reports and forms you want to move into a wiki, a note app or a Markdown editor. A plain text editor shows the raw syntax. - Pages (optional)
- Empty means all pages. Enter single pages and ranges separated by commas, for example
5-10or1-3,7. Only the selected pages are recognised and counted. Numbers are positions in the file, not the page numbers printed on the paper. Pages are read in document order, and a page listed twice counts once. Write ranges with both ends, such as5-40. If you enter a page the file doesn't have, the tool names the actual page count before anything is processed.
How pages count as OCR units
Every job is measured in OCR units. One unit covers up to 10 selected pages, and each job rounds up. A 1-page letter and a 10-page form both use 1 unit, 11 pages use 2, and a 37-page contract uses 4. Only the pages you select count, so 1-3,7 is 4 pages and 1 unit.
Because rounding happens per job, splitting a 60-page report into 1-20, 21-40 and 41-60 costs the same 6 units as one job, while separate jobs for 1-5 and 6-10 cost 2 units instead of 1. Split long documents at multiples of 10 pages; smaller jobs are also easier to proofread.
Units are counted per account, or per IP address for guests, against rolling 1-hour and 24-hour allowances. The free units available are the smaller remaining allowance. If your selection exceeds that amount, signed-in users can cover each extra unit with 1 credit. Without sufficient allowance or credits, select fewer pages or wait for earlier usage to leave those windows.
Getting cleaner text from scans
Text recognition can only be as good as the page images inside your PDF.
Scan at a sensible resolution
As a rule of thumb for scanning printed text, start at 300 dpi and check that small letters are sharp. This is scan-preparation guidance, not a required file setting. Low-resolution scans and faxes can blur small print, making characters harder to distinguish.
Straighten sideways pages
Upright, level lines read most reliably. Turn pages that were scanned sideways or upside down with Rotate PDF before you extract.
Need the gist, not every word?
For long reports where a summary is enough, PDF Summary condenses the document instead of transcribing it.
Formats and limits
- Input format
- PDF, scanned or digital
- Output formats
- Plain text (.txt) or Markdown (.md) in UTF-8, a Word document (.docx), and copy to clipboard
- Maximum file size
- 200 MB per PDF
- Files per job
- 1 PDF, all pages or a page selection
- Usage
- 1 OCR unit per 10 selected pages, rounded up per job
- Processing
- On our servers, with recognition by GLM-OCR (Z.ai) or Google Gemini. A job can run for up to 30 minutes.
- Watermark
- None
- Storage
- Uploads and results are deleted automatically after 48 hours for guests and after 30 days for free accounts. Paid plans can keep files longer.
Frequently asked questions
Does PDF OCR make my PDF searchable?
No. It extracts the text and returns it as plain text, Markdown or a Word document. Your PDF itself is not changed, so keep it if you need the original layout.
Is PDF to text OCR free?
Yes, within a free allowance of OCR units, and without a watermark. Each job uses 1 unit per 10 selected pages, and the count is shown before you click Extract text. Beyond the allowance, select fewer pages, wait for units to become available again, or log in and pay 1 credit per extra unit.
Why are some words or numbers wrong?
Blurry, skewed or low-contrast scans can cause mistakes. Look-alikes such as 0 and O, 1 and l, or rn and m are common OCR confusions, so check names, amounts, dates and reference numbers against the preview. To compare recognitions, switch Document Language between Auto-detect and the document's specific language: this changes between GLM-OCR and Gemini. Changing from one specific language to another keeps Gemini. Each run uses units again.
Is my PDF sent to a third party?
Yes. OCR runs on our servers, which pass the PDF to an external AI service: GLM-OCR from Z.ai with Auto-detect, or Google Gemini with a specific language. If Gemini is rate-limited, the same recognition request goes through OpenRouter. On our side, uploads and results follow the storage period above; that period describes our storage rather than the external providers' policies.
Does it translate the text?
No. Text comes back in the language it is written in. A language chosen under Document Language only helps the engine read ambiguous characters; passages in other languages on the same page are kept as written.
Why did the job fail?
The server has to read the page count before recognition starts. A PDF that asks for a password when you open it cannot be read, so remove the password with Unlock PDF first. If the selected pages contain no readable text, for example blank pages, the job ends with an error saying no text was extracted.