OCR reads the shapes of letters and turns them back into text — which means how clearly those shapes were captured in the first place puts a hard ceiling on how accurate the result can be. No OCR engine recovers detail that a bad scan never captured.
| Setting | Recommendation | Why |
|---|---|---|
| Resolution | 300 DPI | Enough to tell “rn” from “m”; below 200 DPI hurts small fonts noticeably |
| Color mode | Grayscale | Black-and-white loses faint text; full color adds weight for no accuracy gain |
| Alignment | Straight, square to the frame | Skewed text measurably hurts recognition |
| Lighting | Even, flat | Shadows create brightness gradients OCR can misread |
| File format | Lossless (PNG/TIFF) until OCR runs | Heavy JPEG compression blurs letter edges |
Resolution is the single biggest factor. 300 DPI is the standard recommendation for normal printed text — enough for the engine to distinguish similar characters cleanly without producing a needlessly huge file. Below 200 DPI, accuracy drops noticeably on small fonts, footnotes, and anything condensed; above 300–400 DPI you’re mostly just paying in file size, though very small or fine print (a receipt, legal boilerplate) can benefit from 400–600 DPI. Scanning with a phone camera instead of a flatbed makes DPI moot as a setting — what matters instead is getting close enough that text fills a reasonable portion of the frame, with sharp focus, since blur has the same effect as low DPI.
Grayscale is the sweet spot for color mode on typical documents. Pure black-and-white (1-bit) thresholding can lose faint text and turn light pen marks or watermarks into noise, while full color captures information OCR doesn’t need and produces a much larger file for no benefit on plain text. The exception is a document with colored highlighting, stamps, or diagrams that matter visually — scan those in color and accept the larger file, since the accuracy trade-off for pure text is small next to what you’d lose.
Straightness matters more than people expect: a page scanned a few degrees off-square measurably hurts recognition, because OCR engines expect text to run in roughly straight horizontal lines. A flatbed scanner with the paper aligned to the edge guide avoids this almost entirely; a phone camera needs a squarer, more deliberate shot than it feels like you need. Scan to PDF corrects this automatically when scanning with your phone, cropping and straightening each page as you capture it.
Lighting comes next — even, flat lighting beats bright lighting with shadows. A single overhead light or a window behind the camera creates a brightness gradient across the page that can push OCR to misread text in the darker corners, even where a human reading it wouldn’t notice a problem. If you’re scanning with a phone, an overcast day near a window or diffuse indoor lighting beats direct sun or a single desk lamp.
Last, keep the file format lossless until OCR actually runs: PNG, TIFF, or an uncompressed PDF rather than a heavily compressed JPEG where possible. JPEG’s compression introduces small blocky artifacts around edges — including letter edges — and while modest compression is usually fine, an aggressively compressed JPEG measurably degrades OCR accuracy on fine text.
Once you’ve got a clean scan — 300 DPI, grayscale, straight, evenly lit — run it through OCR PDF to add a searchable text layer, or start with Scan to PDF if you’re capturing with a phone, since it handles cropping and straightening as part of the same pass. How to Make a Scanned PDF Searchable covers what OCR does once your scan is ready, and Why You Can’t Select or Search Text in a PDF covers the underlying reason scans aren’t searchable to begin with.
Scanning something that is not in English? Does OCR PDF Work in Languages Other Than English? covers which scripts and accented characters come through reliably.
Once the scan itself is clean, OCR PDF takes it from there — free, no registration, and the file is gone as soon as processing finishes.