For a single scanned PDF, use Adobe Acrobat's Export PDF to Excel, which runs text recognition (OCR) on a scan automatically, or Smallpdf's PDF to Excel in its OCR mode; ChatGPT also works when the table is small and you will check every cell. If the same kind of scanned document arrives every week, set up a parser such as Docparser or Docsumo instead. The free explainer How AI Reads a PDF: OCR, Parsing and Field Extraction shows what happens to a scan and where the numbers go wrong.
A scan is a picture of a page, not text. Before any tool can fill an Excel column, it must recognize the letters (OCR) and then work out where rows and columns begin and end. That second step is why scanned tables often come out with merged cells or shifted columns.
What are my options, and what decides between them?
- Adobe Acrobat. Open the PDF, choose Export PDF, then Spreadsheet and Excel Workbook. Acrobat runs OCR on scanned files during the export, and Adobe itself says the formatting may need a little cleanup in Excel. Deciding factor: you already have Acrobat and need one file done now.
- Smallpdf. Its PDF to Excel tool has a Basic mode for text PDFs and an Advanced with OCR mode for scans. Smallpdf says OCR is a Pro feature with a free trial. Deciding factor: no Acrobat, and a few files a month.
- ChatGPT. Upload the scan or a photo and ask for the table as rows you can download as a spreadsheet. Deciding factor: a short table, and you want to ask questions about it too. Check every number, because a model can misread a digit as easily as OCR can.
- Docparser. Set up a parser once for a document type, mark the table columns, and every new file is read the same way, with OCR for scans and export to Excel or CSV. Deciding factor: the same layout again and again, such as a vendor's statement.
- Docsumo. Document AI that extracts tables from scanned and digital PDFs, flags values it is unsure of for a person, and downloads to Excel. Deciding factor: bank statements and forms where you want a review step built in.
The document data extraction tools comparison lists more options.
How do I get a clean table out of a scan?
- Start with the best scan you can. Straight pages, no shadows, a decent resolution. Bad scans cause most errors.
- Try the one-off route first (Acrobat or Smallpdf) on one page.
- Compare totals. Add up a column in Excel and compare it with the total printed on the page.
- Spot-check numbers that OCR often confuses: 0 and 8, 1 and 7, 5 and 6, and commas versus periods.
- Move to a parser if you do this more than a few times a month with the same layout.
What does this look like in a real lesson?
In the OCR and parsing explainer you help a design studio with two vendor invoices: a digital PDF and a phone scan. You step through OCR (find the text, recognize the letters, produce searchable text) and then open "Spot the OCR mistakes" to see the two misreads it makes on the scan. You compare parsing rules with AI extraction, see why a blank due date beats a guessed one, and open the four checks: the math, duplicates, blanks and approval.
To see a parser reading real PDFs, Docparser: invoice data into a spreadsheet has you check the Totals rule against the PDF before downloading a CSV. And when the tables come from different vendors in different shapes, Compare Vendor Quotes Side by Side shows ChatGPT putting three quotes into one table on the same basis while you check the math yourself.
What mistakes should I avoid?
- Trusting a scanned total you did not check. Always compare one column total with the page.
- Converting a whole batch before testing one page. Settings that fail on page one fail on page fifty.
- Uploading confidential scans to a free tool without checking your company's rules on where files go.
- Fixing the same layout by hand every week. That is the moment to set up a parser.
Key takeaways
- A scan needs OCR before it becomes a table.
- One file: Acrobat or Smallpdf with OCR. Short table: ChatGPT, with checking.
- Same layout every week: Docparser or Docsumo.
- Check totals and commonly misread digits every time.
Frequently asked questions
- Can Excel read a scanned PDF by itself?
Excel's own PDF import works on PDFs that already contain text. A pure scan needs OCR first, in Acrobat, Smallpdf or another tool.
- Why do my columns come out shifted?
The tool guessed the column edges wrong, usually because of faint lines or merged cells. Try a cleaner scan, or a parser where you mark the columns yourself. The OCR and document parsing FAQ covers more fixes.
- What is document processing, in plain words?
Software that reads documents, pulls out the data you need and sends it somewhere useful. The AI document processing explainer explains the steps.