How to turn a PDF back into an editable document, without uploading it

Drop the PDF into REFLOW. It reads the file with the same engine a browser uses to display a PDF, rebuilds the paragraphs, headings, lists, tables and pictures, and writes a .docx that Word, LibreOffice and Pages open — or Markdown, or the document straight into Docs. Nothing is uploaded, and the page works with the network off. Before you save it, read the summary of what it decided: which lines it took to be headings, how many running heads it dropped, whether it found a table. A converter that shows you none of that is asking you to take its reconstruction on faith.

A PDF does not contain paragraphs

This is the thing that explains every frustration people have with PDF conversion, and it is almost never said plainly.

A PDF is a set of drawing instructions. "Put this glyph at this coordinate in this font at this size." That is the whole of it. There is no record that these six lines are one paragraph and the next four are another; no record that this line is a heading; no record that these five lines are a bulleted list. When you read a page, your eye supplies all of that from the spacing and the sizes, instantly and without noticing.

So a converter is not extracting structure. It is inferring it, from geometry, with the same clues your eye uses:

• Lines from glyphs that share a baseline.
• Words from the gaps between them — a gap of about a quarter of the type size is a space, a smaller one is kerning.
• Paragraphs from a bigger-than-usual vertical gap, an indent, or a line that stops short of the right margin when every other line reaches it.
• Headings from the type size, ranked against whatever size most of the document is set in.
• Bold and italic from the names of the embedded fonts, which is why a document that embeds "Helvetica-Bold" converts better than one that fakes bold by drawing the text twice.
• Lists from a marker at the start of a line and an indent on the lines that continue it.

Every one of those is a judgement, and every converter makes them slightly differently. That is why the same PDF comes out differently from three tools, and why the right question to ask a converter is not "is it accurate" but "what did it decide".

The parts nobody mentions

Three things separate a document you can work with from a wall of lines, and they are all invisible when they work.

• Running heads and page numbers. The title at the top of every page and the number at the bottom are furniture. Carried into a document they land as a stray line every page or two, and removing forty of them by hand is how people give up. Lines that repeat at the same height across most of the pages are dropped, and the count is reported so you know it happened.
• Two-column pages. A page set in two columns has lines whose left half and right half share a baseline. Read the baselines straight across and every sentence is spliced to an unrelated one — this is why converting an academic paper usually produces nonsense. The gutter has to be found in the glyph positions, and each column read to the bottom before the next begins.
• Hyphenation. A justified document breaks words across lines. Join the lines naively and you get "unpre- pared"; strip every line-ending hyphen and "long-standing" becomes "longstanding". The document itself answers this: one that hyphenates does it on many lines, one that does not has real hyphens. The decision is made from that evidence and the count is reported either way.

What comes back reliably

Why not reproduce the layout exactly

Some converters aim for a Word file that looks identical to the PDF. They achieve it by putting every line in its own text box pinned to a coordinate, or every page in a table. It looks perfect until you edit one word, at which point nothing reflows, the box overflows, and the document is less use than the PDF was.

The alternative is to produce the document — paragraphs that reflow, headings that behave like headings, lists you can add an item to — and accept that the fonts and the page furniture are gone. That is the choice made here, and it is made on the view that somebody converting a PDF wants to change it. If what you want is the page rather than the text, you do not want a conversion at all.

A scan gives you nothing, and should say so

Half the PDFs people try to convert are scans: a photograph of each page, with no text in the file anywhere. No converter can extract text that is not there, and a tool that hands back an empty document rather than saying why is wasting your afternoon. Read the scan with OPTIC first, which recognises the words and can save the file back as a PDF with the text behind the picture. Then convert that.

Why it matters that it stays on your machine

The PDFs people convert are the ones they did not write and cannot retype: a contract to mark up, a report to quote from, a statement to reformat, a paper under embargo, a client's file. Every free converter takes a copy onto a machine you do not control, and the file sits there under a retention policy you did not read. For a document covered by a duty of confidence — a solicitor's, an accountant's, a doctor's — that upload is the breach, not a risk of one.

A browser can do the whole job. Load the page, turn the network off, and convert. That it still works is the only assurance that does not depend on trusting anybody.

The short version