Readable text
Test text selection. Scanned pages need OCR before analysis.
DeepSeek PDF analysis
A PDF may contain text, images, tables, and scans. Check that text is selectable, remove irrelevant appendices, and ask a question that requires evidence.

Test text selection. Scanned pages need OCR before analysis.
Keep useful pages and name the sections expected to answer.
Ask for a quote, section title, and “not found” when data is absent.
Workflow
Separate extraction, interpretation, and decision. This structure makes an error easier to locate.
Request the exact passages, numbers, and headings. Do not ask for a broad conclusion yet.
Explain how each passage answers the question and flag caveats or contradictions.
Return to the PDF and check each quote before making a decision.
Limits
The file or question often causes the failure. Diagnose the input before writing a longer prompt.
Missing or poor OCR creates omissions and wrong characters.
Merged cells and footnotes can break the intended meaning.
“Analyze this PDF” defines no priority, proof, or format.
A plausible sentence is not evidence until found in the source.
Control prompt
Use this for reports, tenders, and technical documentation.
Question
Evidence
Output
FAQ
Quality depends on text the model can access and how evidence is requested.
An image-only PDF needs OCR first. Check a sample of lines and numbers because OCR errors will spread into the summary.
Request a section-by-section summary with citations, then verify a sample. Explicitly forbid filling in missing information.
Anonymize parties, limit pages, and check service rules. Have a qualified professional review any legal conclusion.
The sentence may be an image, a broken table extraction, or poor OCR. Paste the excerpt as text to isolate the issue.
Traceable analysis
Prepare the document, ask a precise question, and keep a table of citations to verify.