Overview
PDF work typically falls into three categories:
- Reading and Q&A (understanding, summarizing, locating clauses)
- Structured extraction (text, tables, images, metadata, OCR)
- Conversion and editing (Word/Markdown output, merge/split, watermark/encrypt, batch operations)
Before choosing a Skill, sample the PDF to check: scanned vs. native text, two-column layout, tables spanning pages, page count and language, and presence of sensitive content. These factors determine whether you need OCR, reading-order reconstruction, or manual review.
How to choose by task (key points)
- Long-document reading / Q&A: pick Skills that support page-range reading and preserve page references so outputs can be traced back to source pages.
- Text & table extraction: prefer Skills that detect native text vs. scanned pages, provide table exports (CSV/Excel), and preserve heading hierarchy; be especially careful with two-column layouts and multi-page tables.
- Conversion / batch editing: for format changes, merging, splitting, or signing, choose Skills that support batch runs and output format controls; always test on a copy to see how well formatting is preserved.
(When applying these rules, verify each Skill's published capabilities for OCR, two-column handling, and batch behavior.)
Worked example: 20-page scanned two-column research report → CSV tables + Markdown chapters
Goal: Extract tables to CSV and convert “Methods” and “Conclusion” sections into Markdown, keeping page-number references and image placeholders.
Step-by-step (recommended order): 1) Sample-check pages manually - Open pages 1, 5, 10, 15, 20 to confirm scan quality, two-column layout, cross-page tables, image clarity, and language. Note problematic pages (e.g., blurred scans or rotated pages).
2) Make backups and split by chapter locally - Create a copy of the original file. Extract page ranges that contain the target chapters into separate files to simplify focused processing.
3) OCR and page-type detection (use an extraction-capable Skill) - Run a Skill that distinguishes native-text pages from scanned images, then OCR scanned pages. Test on one page first and proofread critical numbers and table borders. - If the document is two-column, use a Skill that outputs a correct reading order or supports two-column merging; otherwise you may need manual reordering later.
4) Detect and export tables - Use the table-extraction feature and export to CSV/Excel. Inspect column counts, merged cells, and multi-page tables. - Manually merge table fragments caused by cross-page continuation or fix column misalignment before analysis.
5) Convert chapters to Markdown - Export text for the “Methods” and “Conclusion” pages into Markdown, preferring Skills that keep heading structure. - Manually fix issues caused by two-column order (reorder columns into natural reading flow), combine broken sentences, fix formulas or reference numbers, and insert page references (e.g., “(p.12)”) and image placeholders.
6) Package outputs - Save the CSV tables, Markdown files, and a page-reference map. Include a short QA note listing known issues (e.g., "Table 2 spans p.7–8; merged manually").
Verification checklist (post-processing):
- OCR quality: compare critical numbers (amounts, percentages, statistics) against the original images.
- Table integrity: confirm headers, merged cells, and cross-page continuations are correct; mark any uncertain cells for review.
- Reading order: ensure two-column text is stitched in correct left-to-right order and paragraphs aren’t split incorrectly.
- Page references: exported text should include original page numbers for traceability.
- Output format: verify CSV encoding (UTF-8) and delimiter; check Markdown image placeholders and table formatting.
Sampling & batch strategy
- Always run the full pipeline on a small sample (2–3 pages) first. If results meet quality requirements, run the pipeline in batches and keep processing logs (which pages processed, when, and by which Skill) for auditability.
Privacy, permissions, and limitations
- Privacy & uploads: for confidential documents, prefer Skills that operate locally or explicitly state no external upload. Only provide necessary page ranges. If Skill behavior is unclear, test in an isolated environment.
- Typical limitations: OCR struggles with low-resolution scans, skewed pages, complex equations, and vertical text. Two-column layouts and irregular formatting may break sentence order. Tables spanning pages may be split and need manual merging.
- Verification requirement: always sample-check automated outputs before using numerical or legal content for decisions.
Summary & quick checklist
- Identify PDF type (scan vs. native, two-column, tables, sensitivity).
- Test OCR and table extraction on a small sample; correct two-column order and cross-page tables as needed.
- Preserve page-number mapping and processing logs; include a QA note with delivered files.
Limitations statement: This guide provides practical steps and checks based on common PDF processing patterns. Before using any specific Skill, read that Skill’s documentation and privacy details, and run tests on copies to confirm suitability.