Convert Scanned PDF to Excel on Mac (no retyping)

The PDF on your Mac is a scan. Someone fed paper through an office copier and emailed you the result, your accountant returned last year’s statements as a scanned bundle, or you photographed a stack of invoices with your iPhone and saved them to PDF. The table is right there on the page — you can read every number — and yet every tool you point at it comes back with nothing. Copy does nothing. Preview finds no text to select, or selects a smear of OCR guesses. The web converter that handled last week’s digital invoice returns an empty sheet or a single column of noise. That is because a scanned PDF is not a document — it is a photograph of a document, and everything that normally makes “PDF to Excel” work has nothing to grab.

This guide is how to convert a scanned PDF to Excel on Mac in 2026 without retyping anything: have AI read the page image the way you do, on-device on macOS 26+, and write a real .xlsx with the columns typed and the numbers ready to sum.

Why scanned PDFs defeat the normal tools

A born-digital PDF — one exported from an invoicing system or a bank portal — contains actual text: characters placed at coordinates on the page. It contains no table structure (which is why even digital PDFs paste badly into Excel, as covered in the AI PDF-to-CSV walkthrough), but at least the characters exist. Converters work by reading that text layer and guessing where the columns are.

A scanned PDF has no text layer at all. Each page is one big image — pixels, not characters. That kills the usual approaches in order:

Copy-paste gets nothing. There is no text to copy. Selection tools in Preview either refuse or, with Live Text, hand you unstructured lines with every trace of the table’s geometry gone — quantities divorced from their descriptions, totals floating free.
Classic converters read the text layer. No layer, no output. Some silently run bolt-on OCR first, which brings the next problem.
Traditional OCR recovers characters, not tables. Off-the-shelf OCR turns the image into a stream of words with approximate positions. Rebuilding a table from that stream means guessing column boundaries from pixel gaps — and on a real scan, with skew from the copier feed, uneven lighting from a phone photo, and multi-line description cells, those guesses collapse. You get 40 invoice lines shredded into 110 rows of fragments, or every field crammed into column A.
The numbers cannot be trusted. OCR that misreads “1,240.50” as “1.240,50”, drops a digit, or fuses a parenthesized negative “(312.40)” onto the next field poisons every SUM downstream — and you will not notice until a total fails to tie.

The failure is not your scan. It is that the tools are trying to reconstruct a table from geometry, when what the job needs is something that reads the page.

What actually works: AI that reads the image

ignitai approaches a scanned page the way a person does: it looks at the rendered image, recognizes the table as a table — headers, rows, the relationship between a quantity and its line total — and writes that understanding directly into a spreadsheet. There is no separate OCR pass whose errors you inherit; reading the characters and understanding the structure happen together. A scanned statement and a born-digital statement go through the same pipeline and come out as the same clean rows.

The full Mac flow:

Drop the scan into ignitai. Drag the PDF (or a folder of them) onto the app window, or use Share from Mail or Files. Multi-page scans are fine — a 14-page scanned statement is one file, one job.
Say what you want, in plain English. For a single-table page — one invoice, one receipt, one price list — skip the prompt and let ignitai take the whole thing. For a document with several sections, one sentence picks the table:

“Extract the transaction table: date, description, amount. One row per transaction. Ignore the handwritten notes and the summary box.”

The prompt matters slightly more on scans than on digital PDFs, because paper documents accumulate clutter — stamps, handwriting in margins, a fax header — and naming what you want tells the model what to ignore.
Choose Excel (.xlsx) as the output. ignitai sets the column types during extraction: dates become dates, amounts become numbers, currency columns get currency formatting. The sheet opens in Excel ready to compute, instead of as text you coax into numbers afterward.
Click Extract. On a Mac running macOS 26 or later with Apple silicon, the extraction runs entirely on-device — relevant for scans in particular, because scanned bundles are so often exactly the documents you least want uploaded: signed forms, statements, tax paperwork. A multi-page scan finishes in seconds.
Open in Excel and check one total. Drop =SUM() under the amount column and compare it against a total printed on the page. Match, and the scan is now a working spreadsheet.

How much scan quality matters

Less than the old OCR era trained you to fear, but it is not zero. Practical guidance from real-world documents:

Flatbed and copier scans at 200–300 DPI are the easy case. Standard office-copier output converts cleanly, including light skew from the document feeder.
iPhone photos of paper work well when the page is flat and the shot is straight-ish. If you are capturing the paper yourself, use the Notes or Files document scanner rather than a bare camera photo — it deskews and normalizes lighting on capture, and the resulting PDF converts noticeably better. (Shooting receipts specifically? The receipt-scanning workflow on Mac covers the capture-to-spreadsheet path end to end.)
What genuinely hurts: text cut off at a margin, a fold or shadow through the middle of the table, and thermal-paper receipts faded to gray. If you cannot read a number when zoomed in, neither can the model. Rescan those pages rather than fighting the output.
Handwriting on the page is fine to have, mixed results to extract. A printed table with handwritten annotations converts cleanly if your prompt says to ignore the notes. Extracting the handwriting itself — a fully handwritten ledger — works more often than you would expect, but verify those numbers line by line rather than trusting a single checksum.

Batches: a banker’s box of paper into one workbook

Nobody has one scanned page; the reason you are here is a folder called scans_2025 with sixty files in it. ignitai runs the whole folder in one motion: drag it in, apply one prompt, and every document flows into a single output — one row per record, each tagged with its source filename, so a year of scanned utility bills or a shoebox of invoices becomes one Excel sheet you can pivot by file. Pages that fail — an upside-down scan, a blank separator sheet — are listed separately so you re-run only those instead of babysitting the batch. The general batch discipline is covered in the batch PDF-to-CSV guide; scans ride the same pipeline.

A worked example: scanned bank statements from your accountant

Your accountant returns January through December as one scanned PDF, 40 pages, no text layer, and you need the year’s transactions in Excel:

Drag the PDF into ignitai.
Prompt: “Extract the transaction table: date, description, amount, balance. One row per transaction. Skip the account summary boxes and page footers.”
Output: Excel. Click Extract; the pages process in one pass.
In Excel, =SUM() the amount column and reconcile: opening balance plus the sum should equal the printed closing balance on the December page.

If the reconciliation ties, twelve months of paper just became a working ledger without a single retyped row. If it is off, sort by amount and scan the extremes — a misread digit on a faded line is findable in a minute. The same job with born-digital statements is the bank-statement-to-Excel workflow on Mac; the only difference here was that your input was a photograph.

The identical pattern handles scanned tax paperwork — a W-2 that arrived on paper, scanned 1099s, receipts for a Schedule C — where the on-device requirement is least negotiable.

The verification habit that makes scans trustworthy

Extraction from a scan deserves one extra notch of skepticism than extraction from a digital file, and the check is the same single formula: reproduce a number printed on the page. Invoice total, statement closing balance, row count — anything the paper asserts, make Excel assert too. =SUM() ties out: done. It does not: sort the column, find the outlier, fix one cell against the page image. This takes a minute and converts “OCR output I should probably double-check someday” into a sheet you can actually hand to someone. A spreadsheet built from paper is only as good as the one total you verified — so verify one.

A note on privacy

Scanned documents are disproportionately the sensitive ones — signed contracts, statements, medical bills, tax forms — because sensitive workflows are the ones still running on paper. A web-based “OCR to Excel” service means uploading those exact pages to someone else’s server. On-device extraction with ignitai on macOS 26+ (Apple silicon) means the page images are read on your Mac, the Excel file is written on your Mac, and nothing leaves the machine. On older hardware, ignitai falls back to a hosted pipeline with documented zero retention. Either way, your paper trail is not training anyone’s model.

Bottom line

A scanned PDF is a photograph, so every tool that depends on a text layer — copy-paste, Preview, classic converters — returns nothing or noise. The workflow that works on a Mac in 2026 is to treat reading as the job: drop the scan into ignitai, optionally name the table you want in one plain-English sentence, export .xlsx with the types already set, and verify one printed total with =SUM(). On-device on macOS 26+, a folder of paper becomes a workbook without a page of it leaving the machine. Use it when the input is copier output, accountant bundles, or iPhone shots of paper; use the digital-PDF workflow when the file was born clean — the steps are the same, which is rather the point.

Get ignitai on the App Store and turn your first scanned PDF into a spreadsheet today.