A newer reason to convert a PDF
For twenty years, converting a PDF meant wanting to edit it. A growing share of the reason now is different: people want to give a document to a language model, and they want to know exactly what it is receiving.
That is a different job with different requirements. You do not need the layout. You do not need pagination. You need the words, in the right order, with enough structure that a heading is recognisable as a heading, and you want to be able to read the result yourself before you send it.
Markdown is close to ideal for that, and Word is close to the worst option.
Why Word is the wrong target here
A .docx is a ZIP of XML. Handing one to a model means either the model unpacks it, or a library converts it back into text first, and if the answer is the second, you have converted twice and lost control of what happened in between.
Plain text is the opposite failure. It carries the words and discards every signal about how they relate. A section heading is indistinguishable from a sentence fragment, so the model has to infer the document's shape from prose alone, and it frequently infers wrongly.
Markdown sits between them. It is text, so nothing needs unpacking. But a leading # costs one character and preserves the outline that tells a reader, human or otherwise, how the document is organised.
Reading order matters more than it seems
On a two-column PDF, a naive extractor reads across the page and interleaves the columns. A person spots this instantly, because the result is gibberish.
A language model does not. It receives a document that looks like prose and quietly means something else, and it will answer questions about the scrambled version with complete confidence. Bad extraction does not produce a visible error; it produces a plausible wrong answer.
This converter detects columns geometrically and reads each one in turn, and tells you when it found them, so you know which kind of document you are dealing with.
What to strip before sending
Conversion is a good moment to look at what you are about to hand over, because it is the last point at which the document is entirely under your control.
Personal data you do not need. Names, addresses, account numbers and dates of birth are frequently irrelevant to the question you are asking, and easy to delete from Markdown before pasting.
Boilerplate. Repeated page headers, footers, confidentiality notices and "page 4 of 37" markers add tokens and no meaning. Page numbers and headers repeated at the same place on every page are left out for you; anything else is worth a look.
Anything you are not entitled to share. A document being convenient to convert does not make it yours to send to a third-party service. That judgement is unavoidably yours.
If you want to edit it instead
The Word converter uses the same document model and writes real paragraphs with heading styles, rather than the text boxes most converters produce. Same reading-order and paragraph work, different output format.
And if nothing at all came out, your PDF is a scan, see what that means, and how to check in ten seconds.