Scanned PDF to Excel

A PDF exported from accounting software or Word contains its text, and a classical converter simply lifts it out. A scanned PDF contains only an image of each page. Copy-paste selects nothing, and a text-based converter returns empty cells or a single column of noise.

Here each page is read from the image, the way a person would read it, and every grid on the page becomes its own table named after its page. A 12-page scanned report with one table per page gives you a workbook with 12 sheets, in order.

Only the pages you are entitled to are ever sent to be read. A preview of a 200-page scan costs the same as a preview of a three-page one: your free pages are read, and the rest are counted and listed without being opened. A job reads up to 40 pages, so run a longer document as several jobs.

Scans from office copiers usually read cleanly. The trouble comes from fax-quality scans at low resolution, pages fed in crooked, and faint dot-matrix print. If a page comes back marked as hard to read, rescan that page at 300 dpi and run it on its own.

Upload a picture of a table and it comes back as a grid of rows and columns on this page. Check it against the original, then download it as an Excel file or a CSV, or copy it straight into a sheet you already have open.

That covers a price list someone sent as a photo, a screenshot of a dashboard that will not export, a scanned invoice, a timetable pinned to a wall, a stock sheet filled in by hand, and a table inside a PDF that refuses to copy cleanly.

Every grid found becomes its own table, and each keeps the page it came from, so ten photos of a ledger give you ten labelled tables rather than one tangled one. Nothing you upload becomes a public page: the file is deleted when the job finishes.

Cell values, row order, column order, the header row and empty cells all survive. The first row of each table is treated as its header.

Numbers are written as numbers, which is the whole point of a spreadsheet. "1,234.50" becomes 1234.5, "$12" becomes 12, and an accounting negative like "(45.00)" becomes −45, so a column adds up the moment it opens. Values that only look like numbers stay as text: a part code like 00123 keeps its leading zeros, and a phone number is not turned into arithmetic.

Formatting does not come across, by design. There are no colours, borders, column widths or merged cells. A merged heading arrives as its value in the first cell with empty cells beside it, which is the flat shape that sorting, filters and pivot tables need anyway.

Two ambiguous cases are left alone rather than guessed: European decimals like "1.234,56" and percentages. Both arrive as text, and Excel's Text to Columns or a find-and-replace converts a whole column in one step.

The biggest factor is how the picture was taken, not which software reads it.

1. Shoot square-on. Perspective squeezes the far side of the page, and the right-hand columns are the first thing to fail.
2. Fill the frame with the table, not the desk it sits on. Small text is where digits get confused.
3. Keep the rows level. A photo rotated a few degrees can split one row into two.
4. Use even light and no flash. A shadow across a column costs you that column.

A screenshot beats a photo every time, because the text was drawn rather than photographed. If the table is on a screen, take a screenshot rather than photographing the monitor.

Tables with ruled lines read best. Tables held together only by spacing usually work, but occasionally two columns merge into one. When that happens, crop to the table and run it again, or split the column afterwards in Excel.

Three pages a day, across up to three uploads, cost nothing and need no account. You see every table on the page and can copy any of them. The copy button puts a table on the clipboard as tab-separated rows, so it pastes into Excel or Google Sheets cell by cell.

The downloads are what cost credits: 3 for XLSX and 2 for CSV, once per job. Downloading the same file again later is free. Pages beyond your free three are 1 credit each, so a ten-page scanned statement is 7 credits plus 3 for the spreadsheet — about 43 cents from the middle pack.

Upload more pages than you have room for and the ones inside your allowance are read in full, while the rest are listed as unread. Nothing is charged for pages that were never read.

Unlike Image to Text, a PDF that already contains text is still read page by page here. Rebuilding a grid needs the layout, not just the words, so every page costs the same.

A misread letter is obvious. A misread digit in a price column is not, and those are exactly the values people use without looking.

The fastest check is a total. If the original has one, sum the column in Excel and compare. A match means every figure in that column is almost certainly right, and a mismatch tells you which column to look at.

Cells that could not be read come back as [illegible] rather than a guess. Search for the word after you open the file, fix those cells from the original, and you are done. That is a couple of minutes per page, against the half hour it takes to retype one.