The document opens perfectly on your screen, but the assistant cannot find a sentence you can see. You try a more detailed prompt. The answer still misses the same section.
First determine whether the PDF contains usable text or only images. OCR creates a searchable text layer from scanned writing. Interpretation is a separate task: explaining what the text means or comparing it with another source. A readable-looking page does not establish that either step has worked.
Photo by Camilo Rueda Lopez on Unsplash.
Check one representative page
Try selecting a short passage and searching for a distinctive phrase using the document reader. Check a page containing the kind of information the task needs, not only the cover.
A file may contain a mixture of text, scans and images. Selecting the heading on page one does not establish that a table in an appendix is accessible. Record which parts matter before choosing the next step.
If an editable original exists, ask its owner whether that is the more suitable source. Reconstructing a spreadsheet from a PDF image may add unnecessary work and ambiguity.
Choose the narrowest intervention
Use this decision sequence:
Text unavailable: obtain a suitable original or use an approved recognition method.
Text present but inaccurate: correct recognition issues before analysis.
Text accurate but structure unclear: preserve headings, tables and source locations.
Meaning unclear: ask a focused interpretation question with the relevant context.
Evidence conflicting: return the conflict to the source owner.
Adobe documents OCR as a way to turn image text into selectable, searchable text, and advises checking the result. That explains the distinction; it is not a recommendation to buy a specific product or remove document protections.
An illustrative price-table problem
A scanned proposal contains a table with a base price, an optional service and an annual renewal amount. Recognition may capture the numbers but lose their relationship to the row labels.
An AI answer could then compare the wrong values even though every digit appears somewhere in the extracted text. Check the table structure and the source page before asking for a commercial comparison.
Where recognition remains uncertain, keep the original image beside the extracted result. Mark unreadable material rather than allowing a tool to complete it from context.
Specify what the reading task needs
Once the source is usable, ask for a narrow answer with page references and preserved conditions. If the question concerns an exclusion, include related notes and attachments rather than only the paragraph that looks relevant.
Keep preparation and interpretation separate in the handoff. A colleague should know whether the text was copied from an original document, recognised from a scan or manually corrected.
Stop when a better source is needed
Repeatedly asking the same question cannot restore missing pages or establish an illegible value. Request a clearer copy from the owner when the decision depends on it.
The useful finish is an answer supported by readable evidence, not proof that the assistant can produce a paragraph about any file it receives.
For related practical guidance, see preparing approved business documents and validating document fields before record updates.
Hili Consulting can help choose between text recognition and interpretation for a business PDF. Arrange a free 30-minute consultation and bring one example of the work you want to improve.
