Optical character recognition works by analyzing the shape of pixels on a page and matching them against known letterforms, which means its accuracy is entirely dependent on how clean and legible those pixel shapes are. When OCR misreads a document, the underlying cause is almost always something about the scan itself, not a limitation of the recognition algorithm, and understanding which factor is causing errors is the fastest way to fix them.
Resolution is the single biggest factor
OCR engines need enough pixels per character to distinguish similar-looking letters, an 'e' from a 'c', a '1' from an 'l' or 'I'. Scans below roughly 200 DPI often don't have enough detail for reliable recognition, especially at smaller font sizes, while 300 DPI is the generally recommended minimum for body text and 400-600 DPI helps significantly with small print, footnotes, or dense tables. If a document was scanned with Scan to PDF at a low resolution setting to save space, that's frequently the root cause of poor OCR results, not the OCR step itself.
Skew, rotation, and contrast
A page scanned even slightly crooked forces the OCR engine to work with tilted text lines, which dramatically increases misreads, since letterform recognition assumes roughly horizontal baselines. Rotating a misaligned scan to true with Rotate PDF before running OCR often improves accuracy more than any other single fix. Low contrast, faded ink, colored paper, a slight gray cast from a poor scanner bulb, is the second major culprit, since OCR relies on a clear boundary between text and background to segment characters correctly.
Handwriting, decorative or unusual fonts, and dense multi-column layouts also degrade accuracy, since most OCR engines are trained primarily on standard printed fonts in simple layouts. A scanned page with a complex table or a two-column academic paper will produce more errors than a plain single-column letter, even at identical resolution and contrast.
Getting the best results in practice
Before running OCR PDF, it's worth checking that pages are upright and straight, scanned at 300 DPI or higher, and reasonably high contrast, black text on white or light backgrounds performs best. If a scan came out crooked or the original document had physical damage or fading, repairing the source scan first produces a much better outcome than trying to compensate for a bad scan by re-running OCR repeatedly on the same file. For documents where the underlying structure matters as much as the text (tables specifically), it's often more reliable to pull structured data out directly with the PDF Table Extractor rather than relying on OCR text alone to reconstruct rows and columns.