Extracting large documents (150+ and 1,000+ pages)

Large files are the most common source of timeouts, partial extractions, and "stuck at 99%" issues. Use the right approach for your file size.

Key limits

• A single extraction run handles up to 1,000 pages.
• The directive [at]parallel:false (keeps context across pages) is reliable up to roughly 100-150 pages; beyond that it can time out.
• AI-based file splitting has a 500-page limit - above that, use deterministic split instructions (syntax below).

Under ~150 pages

Extract normally. If data for one record spans multiple pages (for example payroll where the employee name appears only at the start of their section), add [at]parallel:false on its own line at the bottom of Extra Instructions.

150 - 1,000 pages

Two options:

1. Keep [at]parallel:true (the default behavior - each page processed independently). If a field like a name appears only once but applies to following pages, add: [at]parallel_extended_context: remember the employee name across pages - this remembers the value across pages while still processing in parallel.
2. Or run page-range batches: extract pages 1-250, then 251-500, and so on.

Over 1,000 pages: the Split File workflow

1. Open your sheet and click the workflow icon (in the row of icons at the top middle of the screen - hover to see labels).
2. Click into the Split File node and upload your file where it says File (click the field to change it to upload mode).
3. Hover the Manual Trigger node -> three dots -> Run Workflow.
4. The file is split into 50-page increments and results append to your sheet progressively, 50 pages at a time. This works for files up to tens of thousands of pages.

Split instruction syntax (when providing deterministic split points): write exactly

Pages 100,200,300...

• Capital P in "Pages"
• Exactly three example numbers
• Then three dots (...) - the dots continue the pattern automatically (you do not list every number)

Troubleshooting large runs

• Stuck at 99% or timed out: open Workflows, click the Data Extractor node, and press its play button to re-run just that node. Failed pages can be re-run without restarting everything.
• Duplicate records around split points: information spanning the 50-page boundaries can duplicate - switch the extractor node to [at]parallel:false and re-run that node.
• Output is blank but the file looks fine: the PDF may be corrupted. Re-download it (or re-save via Google Drive and download again) and re-upload.
• "Document exceeds 500 page limit for AI-based splitting": use the deterministic Pages syntax above instead of describing the split in plain language.

Note on billing: split and extracted pages count toward your page allowance (see "How pages and credits are counted"). Re-extracting the same document within 24 hours is free.

Updated on: 10/06/2026