How to copy a table out of a PDF without it turning to mush

You select a neat six-column table, press ⌘C, paste, and get a single column of run-together values with the headers somewhere in the middle. The problem comes from the way a PDF stores a page. Once you understand that structure, the useful fixes make sense.

Why the columns disappear

A PDF describes a page the way a printer needs it described: put this glyph at this coordinate, in this font, at this size. There is no cell, no row, no column and no table object. The grid you can see is a picture your eye assembles out of aligned text and, sometimes, a few drawn lines that have no relationship to the text at all.

So when you copy, the reader walks the text in the order it was written into the file and hands you that sequence. Sometimes that order is row by row and the result is nearly usable. Often it is column by column, or the order the typesetter happened to emit, and the result is unrecoverable. Accessible, tagged PDFs can carry real structure, but they are the minority, and the tagging is frequently wrong even when present.

The options, in order of how much they cost you

• Rectangular selection in Preview. Hold Option while dragging and Preview selects a region rather than a text run. Good enough to pull one narrow column out at a time, and useful for a two-column table. It cannot reconstruct a grid.
• Retype it. Reliable, auditable, and the correct answer for a five-row table. Stops being the correct answer at about twenty.
• Acrobat Pro export. Exports to Word or Excel with real table detection. It works, and it is a paid subscription for something you may need twice a year.
• An online PDF-to-Excel converter.Fast and free, and it means uploading the document. For a published paper, fine. For a contract, a payslip, a patient list or anything under an NDA, that is a disclosure, and the retention terms are the converter's, not yours.
• Recognise the table from the rendered page. Ignore the text layer, look at the pixels, find the grid. This is what the rest of this page is about.

Recognising the table instead of copying it

Apple's Vision framework can detect the row and column structure of a table in an image, on the Mac, with no network. Stowpad uses that: press ⌘⌃⇧T, drag over the table wherever it is on screen, and it comes back as a grid you can check before you use it.

1. Select the tableIt does not have to be a PDF. The same capture works on a table in a web page, a spreadsheet screenshot, a slide or a paused video.
2. Review the recovered gridStowpad shows the rows and columns it found so you can catch a merged header or split cell before it reaches your document.
3. Take it in the format you needMarkdown for documentation and pull requests, CSV or TSV for a spreadsheet, LaTeX or Booktabs for a paper.

Stowpad reads the rendered page and ignores the PDF text layer, so born-digital files and scans work the same way. No generative model sees the image, and the capture is deleted as soon as it has been decoded.

What to check before you trust the output

Any table recognition can be wrong, and a wrong number in a table is much harder to spot than a mangled sentence. Three things are worth a glance every time: merged header cells, which often collapse into the wrong column; empty cells, which can shift a whole row left; and numbers with thousands separators or footnote markers attached. The review step gives you a chance to catch those mistakes before the table reaches your clipboard.

Table capture is a Pro feature; plain screen text capture is in the free tier. Related: extracting text from a screenshot on a Mac.