Reading order
Automatic by default, with a single-column or two-column override for the pages where geometry alone gets it wrong. This is the control that fixes scrambled output.
Convert PDF to Markdown or Text
Extract clean, properly structured Markdown or plain text — headings, paragraphs, lists, tables and reading order preserved. No copy-pasting.
Structure-awareRebuilds headings, lists & tables — not raw text dumps.
Private & secureProcessed in memory and removed after your session.
Markdown or plain textSwitch format anytime — edit before you download.
No signupFree, no watermark, no account needed.
Structure, not just text
A PDF records where each glyph sits on the page — not what the document means. That is why copy-and-paste gives you broken lines, repeated headers and interleaved columns. This converter reconstructs the structure from that layout and writes it out as Markdown or plain text, then shows you the result so you can check it before you download.
What carries over, and what to check
Markdown preserves logical structure — headings, lists, links, emphasis, tables — but not exact visual layout, fonts or positioning. Here is the honest per-element picture.
| Element | How it is reconstructed | Worth a look |
|---|---|---|
| Headings | Inferred from font size, weight and spacing together, so a bold sentence mid-paragraph is not promoted to a section title. | Documents that style headings by colour alone. |
| Paragraphs | Wrapped lines are re-joined and hyphenated line breaks are repaired, so you get sentences rather than one line per printed line. | Deliberate line breaks in poetry or addresses. |
| Lists | Bullets and numbering are detected and written as Markdown lists. | Deeply nested lists, which can flatten a level. |
| Tables | Detected tables become GFM pipe tables, aligned text or TSV — your choice, or ignore them entirely. | Merged cells and borderless tables. Spot-check every one. |
| Links | Link annotations are carried through as Markdown links. | URLs printed as plain text were never links to begin with. |
| Reading order | Reconstructed from geometry, with an explicit single-column or two-column override when the guess is wrong. | Sidebars, pull quotes and mixed-column pages. |
| Page furniture | Repeated headers, footers and standalone page numbers are recognised across pages and removed. | Irregular running heads that change every chapter. |
| Page boundaries | Optional page markers as a horizontal rule, a heading or an HTML comment — useful for citing a source page later. | Nothing; this one is exact. |
| Emphasis and code | Bold, italic and monospace runs are carried over where the PDF distinguishes them by font. | Code indented by spacing rather than set in a mono font. |
| Images and equations | Not extracted. Text around them is kept; the figures themselves are not written to the output. | Anything whose meaning lives in a diagram. |
No converter is exact, and any tool claiming otherwise has not met your PDF yet. What this one does instead is put the output beside the original so the review takes a minute rather than a re-read.
See the difference
The same document, before and after. Structure recovered, page furniture gone.
Before (PDF)
After (Markdown)
# Project Proposal ## 1. Introduction This proposal outlines the scope, timeline, and costs associated with the construction of the new office complex. ## 2. Project Details | Item | Description | Estimated Cost | | --- | --- | --- | | Design | Architectural & Structural | $12,500 | | Construction | Main Building | $250,000 | | Electrical | Wiring & Installation | $18,000 | ## 3. Timeline - Design Phase: 2 weeks - Construction Phase: 16 weeks - Final Inspection: 1 week
How to convert a PDF to Markdown
Three steps, and an editor in the middle so nothing ships unreviewed.
Upload your PDF
Drag and drop your PDF or choose a file — up to 10MB and 100 pages. Only PDF files are accepted.
Extract the content
The tool rebuilds headings, paragraphs, lists, tables, links and reading order into clean Markdown, and removes repeated page furniture.
Review and download
Compare against the original, switch between Markdown and plain text, edit anything that needs it, then download the .md or .txt file.
Jump back to the converter when you are ready.
Why PDF columns come out in the wrong order
The most common complaint about every PDF-to-text tool ever written, and the reason this one has a manual override.
A PDF stores each character’s position, not its logical reading order. In a two-column layout the characters may be stored so that a naive reader runs left to right across both columns, interleaving two separate sentences line by line. Tools infer the intended order from geometry, and that inference can fail on sidebars, pull quotes and mixed layouts.Tagged PDFs encode reading order explicitly, which is why they convert far more reliably.
Automatic detection handles ordinary documents well. When it does not — and on a journal paper or a newsletter it sometimes will not — setting Reading order to Two columns tells the converter what the geometry could not, and the page comes back in one clean flow. That override is a one-click fix for the failure mode that most tools leave you to solve by hand.
Curious whether your file is tagged at all? The PDF object explorer reports it, along with the structure the converter is reading.
The controls that fix the usual problems
Eight settings, each aimed at a specific way PDF extraction normally goes wrong.
Table output
Markdown pipe tables, aligned text, tab-separated values, or ignore tables entirely — because a table that will not reconstruct cleanly is often better left out than left wrong.
Headers and footers
Repeated running heads and footers are detected across pages and stripped, so the same line does not appear forty times in your output.
Page numbers
Standalone page numbers are removed separately, since they survive header removal in documents that put them somewhere unusual.
Hyphenation and line joins
Words split across a line break are rejoined and wrapped lines are merged into real paragraphs, which is what makes the output readable rather than ragged.
Headings and lists
Heading detection and list preservation can each be turned off when a document defeats them and you would rather have flat text than wrong structure.
Links
Keep link annotations as Markdown links, or drop them for a cleaner text file.
Page separators
Mark page boundaries with a rule, a heading or an HTML comment — the simplest way to keep a source-page reference through to a citation.
What people convert PDFs for
Six jobs where a PDF is the wrong container for content you need to work with.
RAG and LLM pipelines
Markdown chunks on heading boundaries, which raw extraction cannot do. Turn on page separators and each chunk keeps a source page you can cite in an answer.
Obsidian and Notion
Get a real note instead of a rough paste: headings become headings, lists stay lists, and the text is editable and linkable straight away.
GitHub and static sites
Move PDF documentation into a repo or a Hugo, Docusaurus or MkDocs site as version-controllable text with a diffable history.
Research papers
Two-column academic PDFs are the hardest case and the one the reading-order override exists for. Section hierarchy and tables carry through; equations do not.
Searchable archives
Plain text outlives every proprietary format. Convert once and the content is greppable, indexable and readable in fifty years.
Study and lecture notes
Slides and handouts become notes you can annotate, reorganise and search, without retyping anything.
Markdown for RAG and LLM pipelines
Retrieval quality is downstream of extraction quality. Scrambled input produces embeddings of scrambled text, and no amount of tuning recovers from that.
Chunk on headings
Heading levels give a splitter real boundaries to cut on, so a chunk is a section rather than an arbitrary window of characters.
Less positional noise
Running heads, footers and page numbers repeated across every chunk dilute the embedding. Removing them is a retrieval improvement, not a cosmetic one.
Citable page markers
Page separators survive chunking, so an answer can point at a source page instead of a document. Use the HTML-comment style to keep them out of rendered output.
It fixes structure, not meaning
Conversion does not correct OCR errors, resolve ambiguous tables or repair a badly written source. Garbage in still applies — cleaner, better-organised garbage.
If the output is not right
Six symptoms, what causes each, and the setting or edit that resolves it.
| Symptom | Why it happens | What to do |
|---|---|---|
| Columns interleaved, footer before the body | Reading order inferred from geometry read across the page instead of down one column. | Set Reading order to Two columns and re-run. |
| Paragraphs broken into single lines | Each printed line is a separate text run in the file. | Line joining and hyphen repair are on by default; check they were not disabled. |
| Table cells in the wrong column | Merged or borderless cells give the reconstruction nothing to align to. | Switch Table output to aligned text or TSV, or ignore tables and paste them separately. |
| The same line repeats on every page | A running head or footer that varies slightly, so it was not recognised as repeated. | Delete it once in the editor before downloading. |
| Headings came out as normal text | The document distinguishes headings by colour or letterspacing rather than size or weight. | Add the # levels in the editor; the structure is otherwise intact. |
| Output is blank or nonsense | The page is a scanned image with no text layer, or the fonts have no usable character mapping. | The tool flags this rather than returning silence. Run OCR elsewhere first, then convert. |
Markdown, plain text, or neither
Three different outputs for three different jobs. You can switch between the first two after conversion without re-uploading.
Markdown (.md)
Structure kept with lightweight symbols: # for headings, - for lists, pipes for tables. Renders on GitHub, in notes apps, wikis and static-site generators, and chunks well for retrieval.
Plain text (.txt)
The same reconstruction with the symbols dropped. Useful when you only want the words — a diff, a word count, a search index, a script that does not care about formatting.
Not raw PDF-to-text
A flat text dump returns one undifferentiated string: headings, lists and tables all become ordinary lines. That is what makes it cheap, and what makes it nearly useless for anything downstream.
Limits, and what happens to your file
Stated up front, because finding out afterwards is worse.
Scanned pages need OCR first
This converter extracts real, selectable text. A scanned page is an image, so there is nothing to extract — the tool detects that and says so rather than handing back blank output. Add a text layer elsewhere, then convert.
Text only, one file at a time
Images, figures and equations are not written to the output, and there is no batch mode or API. Up to 10MB and 100 pages per document; password-protected files must be unlocked first.
Processed, then removed
Your file is sent over an encrypted connection, parsed to extract its text, and removed after your session. It is not stored, shared, or used to train AI models.
Full detail in file handling and the privacy policy.
PDF to Markdown FAQ
Yes. Upload your PDF and it is converted to clean Markdown with headings, paragraphs, lists, tables and links rebuilt from the document structure. You can edit the result and download it as a .md file.
It detects headings from font size, weight and spacing, and reconstructs bullet and numbered lists — including nested items. You can toggle heading and list detection off if you prefer plain paragraphs.
Detected tables are rebuilt as Markdown tables, aligned plain text, or tab-separated values — your choice. Very complex tables may need a quick manual review, which the editor lets you do before you download.
The converter extracts real, selectable text. If a PDF is a scanned image with no text layer, it will tell you that OCR is required rather than returning empty output. OCR is not part of this tool yet.
Yes. The extracted content opens in an editor next to a preview of your PDF, so you can compare, fix anything, switch between Markdown and plain text, and rename the file before downloading.
Your file is processed to extract its text and is removed automatically after your session. We do not use uploaded documents to train AI models.
Unicode text is preserved, and the editor supports right-to-left content with automatic direction. As with any PDF tool, extraction quality depends on how the original document embedded its fonts.
Markdown keeps structure using lightweight symbols — # for headings, - for lists, pipes for tables — so it renders nicely on GitHub, in notes apps and static sites. Plain text drops the symbols for the simplest possible output.
You can convert PDFs up to 10MB and 100 pages. If your file is larger, compress it first and convert the smaller version.
PDFs describe where glyphs sit on a page, not the logical document. Multi-column layouts, unusual fonts and heavy tables can be ambiguous, so the tool shows confidence-based warnings and lets you correct anything before downloading.
Partly. Markdown preserves logical structure — headings, lists, links, emphasis and tables — but not exact visual layout, fonts, colours or positioning. PDF is fixed-layout; Markdown is lightweight structured text. Meaning and hierarchy carry over, appearance does not.
No conversion between these two formats can be. A PDF stores drawing instructions rather than semantics, so structure has to be inferred, and inference can be wrong on unusual layouts. That is why the output opens in an editor beside your original instead of downloading straight away.
PDF-to-text returns a flat string in which headings, lists and tables all become ordinary lines. PDF-to-Markdown reconstructs the structure: headings become # levels, lists become bullets, tables become pipe tables and links are kept — which is what makes the output useful for editing, publishing and retrieval.
Because a PDF stores each character’s position, not its logical reading order, so a two-column page can be read across both columns instead of down one. Set Reading order to Two columns and re-run — that override exists precisely for this failure.
Each printed line is a separate text run inside the PDF, so a naive extraction returns one line per printed line. This converter re-joins wrapped lines and repairs hyphenated line breaks by default; check those options are still enabled if you see ragged output.
Yes, and both are on by default. Running heads and footers that repeat across pages are detected and stripped, and standalone page numbers are removed by a separate rule so they do not survive in documents that place them unusually.
Usually yes. Markdown gives clean structured text that chunks on heading boundaries, so an embedding represents a section rather than an arbitrary window. Raw PDF extraction often scrambles reading order and mixes footers into body text, which measurably hurts retrieval.
Yes. Turn on page separators and choose a horizontal rule, a page heading or an HTML comment. The markers survive chunking, so a retrieved passage can still point at the page it came from.
No. It reconstructs structure from the text the PDF contains. If that text is already wrong — bad OCR, a broken character map, a poorly written source — the output is better organised but no more accurate.
Yes, that is a common use. Markdown is compact, keeps the heading hierarchy a model can follow, and avoids the positional noise that raw PDF text carries. Spot-check tables first, since those are the least reliable part of any conversion.
Convert here, download the .md file, and drop it into your vault or import it into Notion. You get real headings, lists and links rather than the rough result of copying and pasting from a PDF viewer.
Yes, and two-column papers are the case the reading-order override exists for. Section hierarchy, lists and tables carry through; equations and figures do not, so mathematical papers need those added back by hand.
No. This converter produces text — the words around a figure are kept, but the figure itself is not written to the output. If the meaning of a page lives in a diagram, Markdown is not going to carry it.
No. Equations set as images are skipped entirely, and equations set as text extract as the characters they contain rather than as valid maths. Dedicated maths-OCR tools are the right choice for equation-heavy documents.
Not currently — the converter handles one document per run, and there is no batch mode or API. Convert each file separately and combine the Markdown afterwards if you need a single document.
The output uses common syntax that renders correctly in both: ATX headings, hyphen bullets, and GFM pipe tables when you choose Markdown tables. It displays properly on GitHub, in notes apps, wikis and static-site generators.
Get the content out of the PDF
Free, no signup, no watermark, no daily limit. Review the output before you download it.
Convert a PDF to MarkdownAll PDF toolsPDF object explorerPDF font checkerExtract linksEdit PDF