This article lists which document types the indexer can extract text from, plus notes on quality, speed and known limitations for each.
Native Text Extraction (Fast)
- PDF (text-based): Excellent. Most PDFs created from Word/Pages/InDesign are text-based.
- DOCX: Excellent. Includes headings, lists, and table contents.
- XLSX: Good. Each sheet's cells extracted; formulas not evaluated.
- PPTX: Good. Slide titles and body text; speaker notes optionally.
- RTF: Good. Older Word-compatible format.
- HTML: Excellent. Tags stripped, text retained.
- Markdown / plain text: Trivial — read as UTF-8.
- EPUB: Good. Book chapters extracted.
- ODT / ODS / ODP: Good. OpenDocument equivalents handled.
- CSV / TSV: Good. Each row indexed.
OCR Required (Slow)
- Image PDFs: Scanned documents need OCR; Tesseract is reasonable for English print, less so for handwriting or low-resolution scans.
- PNG / JPG with text: OCR on the image directly.
- TIFF: Multi-page TIFF (common for fax archives) supported.
Not Supported (Skip with Warning)
- Encrypted / password-protected files.
- Proprietary formats without an open extractor (older AutoCAD, custom binary).
- Audio/video transcripts (use the AV captioning pipeline instead).
- Database backups, executable binaries.
Worked Examples
- Modern contracts library: All PDFs Word-generated; native extraction works perfectly. No OCR needed.
- 1990s archive: Many scanned PDFs from old fax machines; enable Tesseract, accept some misreads.
- Engineering knowledge base: PPTX from training materials plus DOCX from procedures; both well-handled.
- Mixed corporate file dump: Use the per-file status indicator to spot what failed and either re-index after enabling OCR or skip permanently.