Keep unreadable information visible when reviewing scanned records
Inspect the source image before relying on extracted text. Record unreadable fields and missing pages explicitly, and keep conclusions that depend on them open until a person resolves the evidence gap.
Ger Perdisatt, Acuity AI Advisory
2 min read
Separate the document from its extracted text
A scan may contain typed text, handwriting, stamps and faint marks on the same page. An extracted text view is a working aid. Open the source page to check the information that matters to your task. The ability to select text does not establish that the extraction is complete or accurate.
Start with a simple page inventory: filename, page count, visible page numbering and obvious quality issues. Look for cut edges, blank-looking pages and repeated pages. Describe what you can observe. A gap in printed page numbering needs investigation; it does not prove that an important page is missing.
Check the source at a useful scale
Open the original at a comfortable reading size. Compare the suspect field with nearby writing, but do not treat visual similarity as certainty. If your approved tools offer rotation or image enhancement, retain the unchanged source and compare the result with it. An apparently clearer character may still be ambiguous.
Ask for a better copy when the disputed field affects the decision. Repeating the same question to an assistant is not independent confirmation. Keep the unresolved field in the final summary and explain its practical effect. Someone reading only the summary should know that a conclusion still depends on information that has not been confirmed.
Editorial guidance informed by Acuity's work with teams. Examples are fictional and client materials stay private. To record who owns an AI use and when it is reviewed, see AI Register.