OCR for searchable PDFs: settings, accuracy checks and common mistakes

OCR adds a machine-readable text layer to page images. It makes search and copy possible, but the recognized characters are a hypothesis based on pixels—not a guaranteed transcription.

Start with the scan

Straight, high-contrast pages with adequate resolution produce the best results. Crooked phone photos, bleed-through, shadows and tiny text reduce accuracy. OCR cannot reconstruct detail that was never captured.

Select only the necessary languages

Language data helps the engine decide which characters and words are plausible. Select the languages actually present in the document. Adding dozens “just in case” can slow the job and increase ambiguity.

Use DPI as a controlled trade-off

About 150–200 DPI is a useful starting point for ordinary print. Increase it for small text only when the source contains enough detail. Higher values increase processing time and memory, especially across many pages.

Verify characters that carry consequences

Check names, dates, amounts, account numbers and references. Common confusions include O and 0, I and 1, rn and m, punctuation and accented characters. Search several known phrases and compare copied text with the image.

Remember layout and accessibility

A searchable layer does not automatically create logical reading order, headings or table structure. Screen-reader accessibility requires more than raw OCR. Complex columns and forms may need dedicated remediation.

Plan for browser limits

OCR is CPU- and memory-intensive. Process smaller groups when a long scan fails, close other heavy tabs and keep the device awake. The OCR engine and language models may be downloaded before the first run.