PDF to Markdown: what can and cannot be preserved

A PDF describes where things appear on a page. It does not always describe paragraphs, list nesting, table headers, or reading order. A converter has to infer those structures from positions, fonts, spacing, and text runs.

That is why two PDFs that look similar can convert very differently. One may contain normal text with font information. Another may contain scanned images, embedded subsets, or characters without reliable Unicode mapping.

Simple reports, contracts, guides, and statements with extractable text usually produce useful Markdown. Headings can often be inferred from font size and placement. Lists can be detected from numbering, bullets, and indentation.

Tables convert most reliably when the rows and columns are visibly aligned and the text is not split into many tiny fragments. Very wide tables may still need manual editing because Markdown tables are less flexible than PDF layouts.

Scanned PDFs need OCR before a text converter can help. Complex forms, multi-column newsletters, watermarks, rotated text, and decorative typography should be reviewed carefully.

For legal, financial, or customer-facing content, always compare the Markdown result with the original PDF before publishing or sending it to an AI system.

A PDF is closer to a drawing of a page than a normal document file. It may store a paragraph as many separate text fragments placed at exact coordinates. It may store columns, sidebars, form labels, watermarks, and page numbers as independent pieces with no clear reading order. A human sees the layout and understands it. A converter has to infer structure from position, font size, spacing, and repetition.

That is why a PDF can look clean on screen and still convert poorly. The visible page is not the same as the text layer. Some PDFs contain good Unicode text. Some contain embedded fonts with weak character mapping. Some contain scanned images. Some mix text and images on the same page. The converter can only work with what the file exposes.

For this reason, PDF to Markdown should be treated as extraction plus review. It can save time by pulling out headings, paragraphs, lists, and tables, but it cannot guarantee the authorial structure that was never stored in the file. This is especially important for contracts, statements, forms, and multi-column guides.

Try selecting a sentence in the PDF viewer. If selection grabs normal words in the right order, the file probably has a usable text layer. If selection jumps across columns, grabs one letter at a time, or selects a whole image, expect more cleanup. Copy a small paragraph into a plain text editor. Strange characters, missing spaces, or words in the wrong order are early warnings.

Look for repeated page furniture: headers, footers, page numbers, confidentiality labels, and watermarks. A converter may include these if they appear as normal text. Decide whether they should remain. For a legal document, page numbers may help review. For a guide or article, repeated headers may just pollute the Markdown.

Check tables before trusting them. PDF tables may be drawn with lines, spaces, or separate text blocks rather than stored as tables. If the table has merged cells, footnotes inside cells, or numbers aligned by visual spacing, compare every row after conversion. A single shifted number can change the meaning of a statement.

Lists are one of the hardest parts of PDF conversion because the marker and the body may be separate text objects. A numbered item may look like `17.` followed by a paragraph, but the file may store the marker far away from the sentence. Nested lists add another layer because indentation has to be inferred from coordinates rather than document semantics.

Multi-column pages create a different problem. A reader naturally finishes the left column before moving to the right. A PDF text layer may not store that order. If the Markdown jumps between columns, review the original page and move paragraphs into the correct sequence. This is not a cosmetic fix; reading order affects meaning.

Page breaks should rarely become hard paragraph breaks in Markdown. If a sentence continues from the bottom of one page to the top of the next, join it. If a section ends at a page boundary, keep the paragraph break. The decision depends on language, punctuation, and context, so build a review habit around page transitions.

If the PDF is scanned, OCR is the first step. A text converter cannot extract words that are only pixels. Even after OCR, proofread names, dates, numbers, and uncommon terms. OCR errors often look plausible enough to pass a quick skim, especially in financial or legal documents.

If you can get the original Word, HTML, spreadsheet, or presentation file, use that instead of the PDF. The original format usually contains more structure. PDF conversion is useful when the PDF is the only available source, but it is rarely the cleanest source when an editable document exists.

Use a risk-based review. A classroom handout can tolerate minor spacing cleanup. A contract, account statement, medical record, or compliance document needs careful comparison. The more the output will influence a decision, the closer the Markdown should be checked against the PDF.