OCR eats clarity, not hope
Creases, shadows, and skewed receipts tank accuracy. Flatten the paper, even the light, or export a PDF page and crop — better than arguing with a blurry photo.
Open AI OCR, drop the image, copy the text, then skim amounts, dates, and tax IDs immediately. Models invent lookalike characters: 0/O, 1/l.
Tables from the screen
On-screen grids often become space-aligned pretend tables. After paste, rebuild real columns; don’t assume alignment survived.
Need one region only? Crop the table body first so OCR skips window chrome and ads. Huge images can be compressed first if the type stays readable.
English and mixed scripts
Chinese–English mixes, all-caps acronyms, handwriting — expect more mistakes. Code in screenshots needs a symbol pass; one missing } is an afternoon.
Send the text through a word count or into your doc. For IDs and full card numbers, close the tab when you’re done — don’t dump the original into a public chat.