How to extract text from a PDF offline on Windows

Choose TXT, searchable PDF or Word

Text extraction is not the same as converting a PDF to an editable document. Pick the output that matches the next task:

Choose TXT when plain content matters more than fonts, columns or page breaks. Choose Word when table structure and editing are part of the deliverable. Choose searchable PDF when the scanned page itself remains the record.

Check whether the PDF already contains text

Open a representative page and try selecting one sentence. If individual characters highlight and copied text is readable, the file contains a native text layer. If the entire page behaves like a photograph, it needs OCR. Mixed PDFs are common: a report may contain selectable pages followed by scanned exhibits.

A page can look like text but still have damaged character mappings. Copy a phrase containing punctuation, numbers and accented characters into a plain-text editor. If it becomes empty boxes or unrelated characters, the visible glyphs do not provide reliable text extraction.

Extract PDF text with PDFCore

1. Open the source PDF. Keep the original unchanged, especially for legal, financial or archival work.
2. Choose To text. Use the home-screen PDF to text tool or the Convert tab after opening the file.
3. Select the OCR language. PDFCore lists recognition languages exposed by Windows. This choice is used only when a page needs OCR, but it should match the main language of any scanned pages.
4. Choose a new TXT filename. Avoid overwriting an earlier reviewed transcription.
5. Run the extraction. Native page text is used where enough letter-or-digit content exists; image-only or nearly empty pages receive OCR fallback.
6. Review the saved file. Check page transitions, columns, tables, numbers, names and punctuation before using the text downstream.

What PDFCore 8.0.1 actually does

The released text-conversion path opens every PDF page and counts usable letters and digits. When a page contains at least 15 such characters, PDFCore writes the page's native extracted text. Otherwise it marks the page for OCR, renders the document locally, recognizes that page with the chosen Windows language and replaces the marker with the OCR result. The final file is written as UTF-8 text with a byte-order mark for compatibility with Windows editors.

This threshold is a practical rule, not proof that every native text layer is correct. A page with 15 corrupted characters can pass the count, and a diagram with a few valid labels can be sent to OCR. That is why the verification step remains necessary.

Install and choose the correct OCR language

PDFCore reads Windows.Media.Ocr.OcrEngine.AvailableRecognizerLanguages. Microsoft documents that an OCR language pack must be installed on the device before it can appear in that collection. If the required language is missing, open Windows Settings → Time & language → Language & region, add the language and install its available language features, then reopen PDFCore. See Microsoft's AvailableRecognizerLanguages documentation and Windows language-pack instructions.

Use the language spoken by most of the scanned page. Mixed-language pages, vertical text, handwriting, formulas and decorative type may need separate handling or manual correction even when all relevant language packs are installed.

Why columns and tables look different in TXT

PDF stores text fragments at page coordinates; plain text has no x/y positions, cells or floating boxes. Extraction must choose a reading order. A two-column article may be read down each column or incorrectly joined across the page. A table may become tab-separated, space-separated or a sequence of cell values with no visible grid.

• For columns, verify the last sentence of the left column and the first sentence of the right.
• For tables, compare every row containing totals, negative values or merged headings.
• For headers and footers, expect repeated text unless a structure-aware workflow removes it.
• For footnotes, verify that markers still correspond to the correct note.

If retaining real table cells matters, use PDF to editable Word tables instead of TXT.

A reproducible five-minute accuracy check

1. Select a source page containing a heading, ordinary paragraph, number, punctuation, proper name and—if relevant—a second column.
2. Create a reference transcription from the source, preserving capitalization and punctuation.
3. Run PDF to text with the intended OCR language.
4. Compare the output character by character. Record substitutions, insertions, deletions and reading-order errors separately.
5. Repeat with a native-text page and a scanned page so the two paths are both exercised.
6. Keep the source hash, PDFCore version, selected OCR language and output file with the result.

Do not publish a single “accuracy percentage” unless the corpus, reference transcription and counting method are fixed. Names, account numbers, dates, decimal separators and minus signs deserve manual review even when the surrounding prose appears correct.

Extract text from several PDFs

PDFCore's Batch tools can create one TXT file for each selected PDF. Choose PDF to text, select one installed OCR language for the batch, and choose an output folder. The same language applies to every file in that run. Group files by language first; per-file language selection is not available inside one batch.

Batch conversion writes through temporary files and then moves completed outputs into place. Keep the output folder separate from the source folder, review filename collisions, and spot-check at least one difficult page from every document type.

Privacy and limitations

PDFCore performs native extraction, page rendering and Windows OCR on the PC. The application does not send documents to an online conversion service. Local processing still creates a new text file that may be easier to search, copy or leak than the source PDF. Protect the output folder, backups and downstream tools accordingly.

Plain-text extraction does not preserve page appearance, font styling, images, digital signatures, form behavior, bookmarks or accessibility semantics. It is not a substitute for a verified transcription in legal, medical, financial or safety-critical use.

Questions about PDF to text