A scanned PDF is a photograph of a document. Without OCR (Optical Character Recognition), the text in the scan is as searchable as text in a photograph: not at all. OCR analyses the image of each page and generates a text layer beneath the visible image, allowing the document to be searched, copied, and indexed by search engines and document management systems.
How OCR works
OCR software analyses the pixel patterns in an image and attempts to match them to known character shapes. Modern OCR systems use neural networks trained on millions of character examples and achieve very high accuracy on clearly printed text in standard fonts. The recognition process runs character by character across each line of text identified in the image.
The resulting text is embedded in the PDF as a hidden layer beneath the image. When you search the document for a word, the PDF reader searches this hidden text layer. The visible content remains the original scan image, so the document looks exactly the same as before. What changes is that it becomes searchable and that text can be selected and copied.
When OCR is worth doing
Long documents that you or colleagues will need to search frequently benefit most from OCR. A 200-page scanned reference manual or a thick archive of correspondence becomes dramatically more useful when specific words and names can be found instantly rather than by page-by-page reading.
Documents submitted to digital filing and archiving systems often require searchable PDFs. Law firms, hospitals, and government agencies increasingly require that all filed documents be text-searchable so they can be retrieved by content rather than just by filename or metadata.
Academic papers, theses, and research documents are more citeable and useful when searchable. Supervisors and reviewers who receive a searchable PDF can navigate to specific sections and search for particular terms without scrolling through the entire document.
OCR accuracy limitations
OCR accuracy depends heavily on the quality of the scan, the font style, and the language. Clearly printed standard fonts on clean white paper at 200 DPI or higher typically achieve accuracy rates above 99 percent. One error per hundred characters means approximately five errors per page of normal text, which is acceptable for most purposes.
Handwritten text, unusual fonts, low-contrast scanning, and damaged documents significantly reduce accuracy. OCR on handwriting is a separate specialised task with lower accuracy than printed text recognition. Do not rely on OCR for handwritten content where exact text retrieval matters.
Languages using non-Latin scripts require OCR models trained on those scripts. Arabic, Chinese, Devanagari, and other scripts have their own OCR models with varying accuracy levels. Most online OCR tools support a range of languages but accuracy varies more than for Latin-script languages.
How OCR affects file size
Adding a text layer to a scanned PDF adds a small amount of data to the file. The text itself is compact: a full page of text as a string of characters is typically a few kilobytes. The overhead of embedding this in the PDF structure adds a modest amount. Overall, the text layer adds between 1 and 5 percent to the total file size, which is negligible.
What can significantly affect file size during OCR processing is image compression applied during the process. Many OCR tools re-compress the page images as part of their processing pipeline. Depending on the settings, this re-compression can either increase or decrease the file size of the resulting document.
An OCR tool that compresses page images to grayscale at 150 DPI during processing may produce a file that is 50 to 70 percent smaller than the original colour scan, even though it is adding a text layer. The image compression during OCR is the size change, not the text layer itself.
Making a scanned PDF both searchable and smaller
The combination of OCR with image optimisation during processing is the most efficient way to handle scanned documents. If your OCR tool offers settings for output resolution and colour mode, selecting grayscale at 150 DPI produces a document that is significantly smaller than the original scan while adding the searchable text layer.
If your OCR tool does not offer image compression options, run the PDF through a compression step after OCR processing. The text layer will be preserved through a compression pass that only targets the image data.
Checking OCR quality
After OCR processing, open the PDF and try to select text by clicking and dragging. If text selection works, the OCR has produced a text layer. Search for a word you know appears in the document. If the search finds it and highlights the correct location, the OCR quality is adequate for search purposes.
For documents where the OCR text will be used for processing beyond search, such as data extraction or automated classification, sample the accuracy by copying a paragraph of text and comparing it to the original scan. Any systematic errors in character recognition will be visible through this spot-check.