Invoice data extraction
Collect invoice and receipt fields.
Reads invoices and receipts into fields you can check: vendor, invoice number, dates, currency, subtotal, tax, total and line items.
How to use it
- Set up AI once in Chat with PDF: connect your own API key, or download a model to this device.
- Choose one or more invoices or receipts (PDF) and select Read invoices. Scanned pages are recognized first.
- Review each invoice. Select a page reference, such as p. 1, to see the page with the value outlined. Correct anything that’s wrong; the checks update as you type.
- Download invoices.csv (a row per invoice) and line-items.csv (a row per line), or JSON with each value’s page and the words that state it.
Options
| Option | Values | Default |
|---|---|---|
| Documents | Choose two or more files |
Output
For a batch, invoices.csv with a row per invoice and line-items.csv with a row per line; for one invoice or all of them, JSON. Every value keeps the page it comes from.
How values are checked
The AI is asked to copy each value from the document with the page it is on and the words that state it, and to give nothing when the document doesn’t say. superPDF then looks for every value on the page it cites. Amounts are matched however they are written (£1,234.50, 1.234,50 €, (12.00)) and dates in any common form (12 March 2026, 03/12/2026, 2026-03-12).
- Found on the page: the value is on the page the AI cited.
- Found on another page: it is on a different page of the document, and the page reference is corrected.
- Not found in the document: it isn’t on any page the AI read, so the field is left empty. The AI’s suggestion is shown beside it, and you can choose Use the suggestion anyway.
- Not stated: the AI found no value.
Text in a document that addresses an AI (such as “ignore previous instructions”) is removed before the AI reads it and doesn’t count as the document’s text when values are checked.
Arithmetic checks
Each line’s quantity × unit price is compared with its amount (within a cent, for rounding), the line amounts with the subtotal, and the subtotal plus tax with the total. When there is no subtotal, the lines plus tax are compared with the total. A failed check says what it found, such as the subtotal plus tax make 120.00, but the total is 125.00. Discounts, shipping and other charges that aren’t line items can make a check fail on a correct invoice.
Where your invoices go
With your API key, each invoice’s text goes from your browser to your provider; with a model on this device, nothing leaves it. The note under the results says which was used.
Limits
superPDF sets no file size or page limit. A small model on this device is slower and misses more, especially line items in long tables; your own key gives better results. Currencies shown only as $ or ¥ are marked for checking, since several currencies use them. Tax rates aren’t collected, only tax amounts.
Scanned pages
Scanned pages, including copier scans whose text layer is weak or garbled, are read with the quick reader first (the AI tools run OCR PDF for you), so you get a result straight away. On computers that can run it, superPDF then offers high-accuracy reading beside the result, and improves those pages in the background: in Chat and Ask multiple PDFs the pages behind your latest question come first and new questions use the improved text; elsewhere, Run again with the improved text redoes the result. It runs on this device: nothing leaves it, and it works offline once downloaded. Download: 871 MB, once, only when you choose it; delete it any time in On-device AI models. Speed: about 10 to 30 seconds a page on a recent computer (the offer shows the time for your pages); up to 20 pages are read at a time unless you choose more or pick pages. It reads English and other European languages best; check results in other scripts, and like any OCR it can misread faded or handwritten text. Running page headers, footers and page numbers are left out of the text. If anything goes wrong, the quick result stays, with a short note. See High-accuracy reading.