Can't Copy Text from Your PDF? Scanned vs Text PDFs, and How OCR Accuracy Works

You open a PDF, try to select a sentence to copy it, and nothing highlights — or the whole page selects as one block. That's the tell: the PDF has no real text inside it, just a picture of text. Which tool gets you usable text out depends on which kind of PDF you actually have, and how these three tools decide that differs enough to matter.

Two fundamentally different kinds of PDF

A text PDF — exported from Word, generated by software, most invoices and reports — stores each character as actual character data, positioned on the page. Your browser or PDF reader can select, search, and copy it directly, instantly, because the text is already there as data.

A scanned (image) PDF — a photographed or scanned page — has no character data at all. Each page is a picture, the same as a JPEG, just wrapped in a PDF file. Nothing on the page is selectable, because as far as the file format is concerned there is no text, only pixels arranged to look like text. Getting real text out of one requires OCR — software that looks at the image and recognizes what the shapes probably say.

To check which you have: try selecting text with your cursor. If individual words and lines highlight, it's a text PDF. If the whole page highlights as a single image, or nothing highlights, it's scanned. A PDF can also be mixed — some pages with real text, others that are scanned insertions — which matters for which tool to reach for, below.

Three tools, three different jobs

PDF to Text reads a PDF's existing text layer directly — fast and exact, because it's reading data rather than recognizing an image. It also has an automatic fallback: if the whole file has almost no extractable text (about 20 characters or fewer, a strong sign the pages are scans), it switches to OCR automatically, in English only. That fallback only triggers when there's essentially no text anywhere in the file — a mixed PDF with real text on most pages and one scanned page won't trigger it, and that one image-only page will come out blank.

Scanned PDF to Text always runs OCR, on every page, in a language you choose (English, Spanish, French, German, Hindi, or Portuguese). This is the right tool for a PDF you already know is scanned, for a scan that isn't in English, or for a mixed PDF where you specifically need the image pages recognized rather than skipped.

Image to Text runs the same OCR engine, but on a standalone image file (JPG, PNG, WebP) rather than a PDF — a photo of a whiteboard, a screenshot, a single scanned page you haven't turned into a PDF yet. It offers the same six languages and shows a confidence score for the result. It doesn't open PDFs at all; for those, use one of the two tools above instead.

What decides how accurate OCR actually is

All three tools use the same underlying OCR engine (Tesseract), so accuracy depends less on which tool you picked and more on the input:

• Resolution. Low-resolution or heavily compressed scans give the engine less detail to work with — a crisp 300 DPI scan recognizes far better than a blurry phone photo of the same page.
• Skew and orientation. A page scanned crooked, or a photo taken at an angle, confuses character recognition; straightening it first (or rescanning) usually helps more than any setting in the tool.
• Language match. Picking the wrong language for the text tells the engine to expect a different alphabet or letter frequencies than what's actually there, which measurably hurts accuracy — always match the language selector to the text.
• Font and print quality. Clean, standard printed fonts recognize best. Handwriting, decorative or stylized fonts, low-contrast text (light gray on white), and dense small print are all meaningfully harder for OCR and should be treated as unreliable rather than fixed by any setting.
• What it can't do at all: tables and layout. All three tools return plain text in reading order — column structure, table cells, and formatting aren't reconstructed, whether the source is a text PDF or a scan.

Because none of this is guaranteed, proofread OCR output before relying on it, especially for anything with numbers, names, or other details where a misread character matters. Image to Text's confidence score is a useful signal for this — a low score on a page is a cue to check it more carefully, not proof that it's wrong.

Privacy: what actually happens to your file

All three tools run the recognition engine as WebAssembly directly in your browser tab — your PDF or image is never uploaded to a server, on any of them. The one thing that does come from the network is the OCR engine itself and its language data (a few megabytes), downloaded from the jsDelivr CDN the first time you use OCR in a session and cached by your browser after that — later runs start faster because that download is skipped. That first-run download is the only external request the OCR features make; the file you're recognizing text from stays local.

Which tool for which situation

If what you actually need out the other end is an editable Word document rather than plain text: PDF to Word only works from a real text layer, the same way PDF to Text does — it doesn't run OCR at all. Point it at a scanned PDF and it tells you so and redirects you to Scanned PDF to Text instead, which will get you the words as plain text, not a formatted Word document with the original layout.