Prompt: extract a PDF table without filling in what you can't read
Pull rows from a PDF with their page and line, keep unreadable cells empty and flagged, and report the printed total apart, so the gap stays visible before you import.
Works with: ChatGPT, Claude, Copilot, Gemini
The task
Prototype a PDF-to-spreadsheet extraction that keeps a trace of every row and never turns an unreadable cell into a guess.
Treat these lines as a fictional PDF transcription. Extract detail rows only, retaining SKU as text. Return JSON with rows containing sku, quantity (integer or null), page, line and exact evidence. Quantity 0 is valid. [unreadable] becomes null with review_reason="unreadable". Do not infer it from another row or the printed total. Also return known_quantity_sum, incomplete_total and printed_total. The printed total is a document claim, not an extracted detail row. Page 1 line 4: SKU 0012 | quantity 3 Page 1 line 5: SKU 0048 | quantity 0 Page 2 line 2: SKU 0091 | quantity [unreadable] Page 2 line 6: Printed total quantity: 8
How to use it
- Run it with the fictional transcription first.
- Then try a synthetic PDF in your approved extraction tool and compare every row with the page image.
- Check each null on the page itself before the spreadsheet import.
What you get
Three detail rows with SKUs kept as text (0012, 0048, 0091), quantities 3, 0 and null, a known sum of 3, the total marked incomplete, and the printed 8 reported apart.
{
"rows": [
{"sku": "0012", "quantity": 3, "page": 1, "line": 4, "evidence": "SKU 0012 | quantity 3"},
{"sku": "0048", "quantity": 0, "page": 1, "line": 5, "evidence": "SKU 0048 | quantity 0"},
{"sku": "0091", "quantity": null, "page": 2, "line": 2, "evidence": "SKU 0091 | quantity [unreadable]", "review_reason": "unreadable"}
],
"known_quantity_sum": 3,
"incomplete_total": true,
"printed_total": 8
}
SKU 0091's quantity is null and flagged for review. I didn't fill it in from the printed total. The total implies 5, but that number isn't on the row itself.Where should you start with AI?
The free assessment gives you, in 4 minutes, the 3 tasks to automate first in your business.