How a table is actually detected
The detector works on alignment rather than on ruled lines, because plenty of real tables have no lines at all and plenty of ruled boxes are not tables. It looks for runs of text whose left edges repeat down the page at the same x-positions, groups those into candidate columns, and then applies two sanity checks before calling it a table.
The first check is fill: in a real table most cells are short. If the average cell is nearly as wide as its column and long, this is prose in columns — a two-column newsletter, say — and turning it into a table would be worse than leaving it as text. The second is size: fewer than three rows or fewer than two columns is not enough evidence, and forcing a table on that basis produces false positives far more often than it rescues a real one.
This is why a table with one very long free-text column sometimes comes through as paragraphs. It is a deliberate trade: the detector would rather miss an unusual table than turn every two-column page in your document into a grid.
Which mode for which document
Pick flowing when you intend to keep writing. A report you are updating, a contract you are amending, an article you are re-editing — anything where paragraphs need to re-wrap when you type. Flowing mode gives you ordinary Word text: headings are headings, paragraphs are paragraphs, and the document behaves like something authored in Word.
Pick layout when the arrangement is the information. An invoice where the total has to sit under the column of line items. A statement. A form. A certificate. Layout mode uses positioned text frames and page-anchored tables to put things where they were, and the cost of that fidelity is that the document is stiff — inserting a sentence will not push everything below it down, because nothing is in a flow to be pushed.
If you are unsure, convert twice. It takes seconds, nothing is uploaded either time, and comparing the two side by side answers the question faster than reasoning about it.
When the PDF is a scan
If the converter reports very little text, or none, the pages are probably images. A scanned or photographed document has no text layer to extract, and no converter can produce editable text from it without optical character recognition, which this site does not yet do. PDF to JPG will get you the page images if that is useful, but there is no honest route from a scan to an editable Word file here today.
A note on what "keeps the layout" can mean
No converter reproduces a PDF perfectly in Word, and any that claims to is describing a narrower set of documents than yours. The two formats disagree at the foundations: one is a fixed arrangement of glyphs on a fixed page, the other is a flowing document that decides where things go when it draws them. Layout mode gets close by giving up flow. Flowing mode keeps flow by giving up exact position. Choosing between them consciously is the whole technique.
