# PDF to Markdown Reading Order Problems: Find the First Wrong Turn

Diagnose mixed paragraphs, repeated headers, and misplaced figures in PDF-to-Markdown output before editing or converting again.

If converted Markdown jumps between topics, stop before rewriting sentences to make them fit. The wording may be correct while the paragraphs are out of order. Editing the prose first can disguise the original extraction problem.

Work from a specific bad passage back to its page. A small, repeatable diagnosis is more useful than describing the entire result as “garbled.”

## Match the passage to its original page

Find a distinctive phrase in the Markdown and search for it in the PDF if its text layer allows that. Otherwise use the nearest section heading, figure number, or printed page label to locate the scan. Distinguish the PDF viewer's page position from the page number printed on the document.

Read the source paragraph and its immediate neighbors. Mark where the Markdown agrees, where it diverges, and where it rejoins the source. For example, a caption may have been inserted between two halves of one paragraph. That needs a block move, not a new translation of the whole section.

If you cannot locate the original passage, do not let an AI tool invent the missing connection. Preserve an uncertainty note until the source is found. Fluent repair is not evidence of correct order.

## Separate headers and footers from the argument

Repeated journal names, chapter titles, author names, or page numbers can appear inside sentences after extraction. Check whether the same fragment returns at a regular page interval.

Copy one example into a review note with its page location. Then inspect another occurrence before using a global search-and-replace. A word used in a running header may also appear legitimately in the text. Numbers deserve particular care: a printed page number can resemble a reference number, year, or table value.

Remove a confirmed running header from a working copy and reread across the join. If the sentence still fails, there may also be a missing paragraph, a column switch, or a broken word. Do not assume one cleanup fixes every boundary.

## Look for figures, tables, and sidebars in the wrong place

A paragraph interrupted by a table is different from a paragraph whose words were interleaved with table cells. In the first case, moving a complete block may restore the flow. In the second, identify the text and table separately against the PDF before making changes.

Use the caption and body reference to decide where a figure belongs. A figure should remain interpretable with its caption, number, and surrounding explanation. If an image is missing entirely, inspect the downloaded resources rather than replacing it with a written guess about what it showed.

For sidebars, check whether the author intended a separate reading path. A boxed definition may not belong between the main paragraph's first and second sentence. The [two-column review guide](https://pdfcraft.ai/en/blog/two-column-pdf-reading-order/) covers input layouts that commonly cause these transitions.

## Record enough detail to choose a remedy

A useful issue record is short but reproducible:

| Field          | Illustrative entry                                                   |
| -------------- | -------------------------------------------------------------------- |
| Source         | PDF viewer page 18, printed page 12                                  |
| Output locator | Section “Evaluation,” paragraph beginning “The second trial”         |
| Problem        | Right-column paragraph appears before left-column continuation       |
| Correction     | Restore the source sequence; retain the table after its introduction |
| Recheck        | Confirm the next page starts at the same point                       |

This is a sample record format, not a reported PDF Craft result. Use your document's actual locations and observations.

A few isolated mistakes may be faster to correct in a Markdown editor. Repeated column interleaving or extensive omissions may justify another conversion route or a better source. Before paying to process the same file again, establish what you expect to change: scan quality, source file, processing settings, or converter. Repeating identical inputs is not a reliable repair strategy.

## Check the repaired sequence before using the text

In [PDF Craft's Markdown workflow](https://pdfcraft.ai/en/pdf-craft/pdf-to-markdown/), upload the PDF, choose Markdown, and download the result from the completed task. Document processing aims to recover useful structure, including complex layouts; it does not remove the need to review the reading order.

Keep the untouched export, your working copy, and the source PDF. Compare a paragraph before the edit, the edited block, and a paragraph after it. Open the Markdown in its destination editor as well: an editor's rendering problem and a source-order problem need different fixes.

Only then extract quotations, split research notes, or submit passages to another tool. The [PDF-to-Markdown research guide](https://pdfcraft.ai/en/blog/pdf-to-markdown-for-research/) explains how to keep those extracts tied to their source. A repaired paragraph should preserve the author's sequence, not merely sound plausible on its own.