Make a scanned PDF searchable, selectable and copyable

Open the scanned file and press Ctrl+F (Cmd+F on a Mac). Search for a word you can see on the page. If the reader reports no matches, the file has no text layer: each page is one image, and there is nothing to search. Dragging across a line either selects nothing or grabs the whole picture.

OCR PDF keeps that image exactly as it is and writes the recognized words on top of it as invisible text. Each word is placed over its own box on the page, with the font sized to the box height and the text stretched to the box width, so when you later drag across a sentence the highlight lands on the printed word. The visible page does not change. A search hit or a selection is the only sign anything was added.

The output is a new file named after the original with -searchable on the end. Your original is not touched.

Many PDFs are mixed. A contract may have typed pages plus a scanned signature page, or an earlier OCR pass may have covered some pages and missed others. The "Skip pages that already have text" checkbox, ticked by default, checks each page for a text layer and leaves any page that has one exactly as it is. The note under the result says how many pages were recognized and how many were left alone because they already had text.

If every page already has text, the tool stops with a message instead of producing a file. That usually means the PDF was searchable all along and your search failed for another reason, such as a word split across a line break.

Untick the box only when you want every page read again. Existing text is not removed; the new hidden layer goes on top, so a wrong layer from an earlier pass stays and a search may match both.

The first run downloads about 5 MB of engine and English model, which the browser caches for later runs. While it works the engine uses roughly 100 MB of memory, released when the run finishes.

Each page is rendered to an image at 200 dpi, capped at 2400 pixels wide so an oversized scan is not blown up past what Tesseract handles well. The engine reads it with automatic page segmentation, which finds headings, columns and captions separately instead of treating the page as one block. Progress shows "Rendering page 3 of 12" and then "Reading page 3 of 12". Words with a confidence score below 30 on Tesseract's 0 to 100 scale are left out of the hidden layer rather than adding noise to search results.

When it finishes, a "Recognized text" box shows the plain text from every recognized page with a Copy button. That is the fastest way to judge quality before you download, with one caveat: the preview holds everything the engine read, including the low-confidence words the hidden layer drops, so a word that looks wrong there may simply be absent from search.

Tesseract reads clean scans of printed text well. Handwriting, faint or low-resolution copies and unusual fonts come out with errors or gaps. Because the added text is invisible, a mistake never shows on the page; you find it when a search for a word you can see returns nothing. Check the preview, or search for a few words from different pages.

Only English is offered right now. Words in other scripts may come back as nonsense, and any word the built-in Latin font cannot encode is skipped in the hidden layer even when it was read correctly, so non-Latin text stays unsearchable.

Pages must be upright. The tool respects a rotation flag stored in the PDF, but if the scan itself is sideways the engine sees sideways text and reads little. Fix that first with Rotate PDF, then run OCR.

The tool works on one PDF at a time, up to 100 MB. If you want one searchable document from several scans, merge them first. Once the file has a text layer, tools that read text start working on it: Summarize PDF can rank its sentences, and PDF to Markdown can infer headings from the sizes of the hidden words. Both had nothing to work with before.