Separate PDF pages by recurring phrase and partition variable-length invoices
Batch exports from accounting software, payroll systems, and banking portals often combine hundreds of separate records into one continuous PDF. Partitioning variable-length invoices automatically by keyword saves hours whenever individual records span between one and three pages long.
Detecting split phrases across visual text lines
When you select Split by text, the browser loads the file locally and reconstructs visual text lines on every page in sequence. It checks each page against the phrase entered in Start a new file at every page containing rather than relying on fixed page counts.
Every page that contains the target phrase begins a new group, while subsequent pages without the phrase append to the active group. Under the Library of Congress PDF 1.7 (ISO 32000-1) format description, the copied pages retain their original vector text, embedded fonts, hyperlinks, and compressed object streams without any lossy rasterization or re-encoding.
Handling pages before the first match
If page 1 does not contain your trigger phrase, the splitter still opens an initial group at index 1 so no cover sheet or summary table is dropped. Once the phrase appears on a later page, a second group begins immediately.
At least 2 groups must be formed for splitting to proceed. If your document is a flat scan with zero embedded text, run OCR on scanned pages first so the splitter can read the headings.
How does automatic file naming sanitize generated PDF filenames?
Automatic naming cleans each matched text line with Unicode compatibility normalization, strips characters blocked by operating systems, and caps generated PDF filenames at 60 characters. Adding zero-padded numeric prefixes before the cleaned line guarantees every split file sorts in its original document order inside your file manager.
Sanitizing forbidden characters and 60-character caps
Invoice lines frequently include slashes, colons, or quotation marks such as Invoice: 2026/04-108 that Windows and macOS forbid in filenames. Leaving Name files after the matching line enabled triggers a sanitizer based on Unicode Standard Annex #15 on Normalization Forms using NFKD decomposition.
The sanitizer replaces reserved symbols and control bytes with single spaces, trims whitespace, and truncates strings at 60 characters. If a matched line consists solely of stripped punctuation, the filename falls back cleanly to from page N.
Preserving sort order with zero-padded prefixes
When a batch produces 120 split files, alphabetical sorting without padding places 10 ahead of 2. The output builder pads each sequential number to match the total digit width, producing 001 through 120 automatically.
If you prefer generic numbered names instead of document titles, uncheck Name files after the matching line. Each file then receives a clean 01 from page 1.pdf style label ready for downstream bookmark-based chapter splitting workflows or manual archiving.
Keyword splitting vs. fixed page interval division and bookmarks
Choosing the right splitting method depends on how your source PDF is structured internally. Documents exported from ERP systems rarely include outline trees, whereas published manuals often lack identical header strings on every chapter opener.
| Split method | Trigger rule | Best document type | Requires text layer | Output naming |
|---|---|---|---|---|
| Text phrase split | Recurring word or line | Invoices, payslips, statements | Yes | Zero-padded matched line |
| Bookmark split | Outline tree depth | Books, technical manuals | No | Numbered chapter title |
| Page range split | Explicit page numbers | Fixed-length handouts | No | Custom range suffix |
| File size split | Target MB threshold | Email attachments | No | Sequential part index |
| Half-page cut | Vertical or horizontal bisection | Two-page book scans | No | Single combined PDF |
As outlined in the PDF Association ISO 32000 standard overview, page trees store page objects independently of outline dictionaries. When an exported report lacks bookmarks altogether, text-based matching provides reliable boundaries without manual inspection.
For documents where every section already has a known start and end page, you can separate fixed page ranges manually instead of searching text lines. Both methods copy page dictionaries losslessly without re-encoding embedded fonts or high-resolution images, keeping file sizes predictable across large archives.
What happens with scanned receipts without a searchable text layer?
Image-only receipts store flat raster bitmaps rather than character operators, so the extractor finds zero text to match. Mistyped phrases or strict capitalization settings can also stop a split before producing output.
Diagnosing image-only pages and OCR requirements
Flatbed scanners and mobile camera apps save pages as embedded raster bitmaps with 0 bytes of selectable text. When every page returns empty text, the splitter halts immediately and prompts you to add an OCR layer before retrying.
Tip: Always test your trigger phrase on a 5-page sample when splitting a 500-page archive. Verifying that your matched line produces clean 60-character filenames prevents re-running large batches unnecessarily.
Fixing case sensitivity and partial line matches
Case-insensitive matching is active by default so invoice number matches INVOICE NUMBER on uppercase headers. Tick Match upper and lower case only when a lowercase body word would otherwise trigger false splits mid-document.
Choose a phrase unique to the start of each record, such as Account Summary rather than Total, so multi-page statements do not split on every sub-page. After splitting, you can also remove surplus separator pages from any individual output file.