PDF to Markdown — for pasting into ChatGPT and Claude
Converts a PDF to Markdown with its headings, tables and lists intact, ready to paste into a prompt. On supported browsers everything runs locally and the file never leaves your device. PDFs that cannot be read — scans, for instance — are reported as such instead of being converted into nothing.
When this helps
Copying out of a PDF viewer destroys the formatting and the page structure
Paste the converted Markdown straight into the prompt — tables included
The numbers you need are in a table you would otherwise retype
Get the table as a Markdown table and work from it directly
You want to see what is actually in the text layer before writing code against it
Check the extraction and the table detection before pulling in a PDF library
The text will not copy, or copies as gibberish
Convert it — and if it cannot be read, find out why rather than guessing
FAQ
- Is my file stored on your server?
- No. If your browser supports WebAssembly (WASM, a technology for running software inside the browser safely and quickly), the file never leaves your device. If it does not, the conversion runs on the server: the file is held only while it is being converted, deleted the moment it has been, and the result is deleted as soon as it reaches you. Either way nothing is kept.
- Does it extract tables?
- Yes. Tables drawn with ruled lines and tables held together only by text alignment are both detected and written out as Markdown tables. The summary after a conversion says how many pages had tables on them.
- What about scanned PDFs with no text layer?
- Not supported — but the tool will not hand you an empty file and call it done. It detects that the PDF is a scanned image, says so, and stops. Reading one would need OCR, which this tool does not do.
- What if the text in a PDF cannot be decoded?
- Some PDFs embed their fonts in a way that makes the text impossible to read back correctly, which happens with certain Japanese PDFs in particular. Rather than emitting nonsense, the tool reports that the text could not be decoded and stops. When only some pages are affected, the readable pages are converted and the summary states how many pages are missing from the result.
- How is the token estimate calculated?
- It is a rough character-based approximation, not a real tokenizer. Use it as a guide to whether the text will fit in a model's context window.
- What is the file size limit?
- 20MB per file.
Related tools
- Split a PDF into pages — When you only need certain pages
- Word to Markdown — When the source is a Word, Excel or PowerPoint file
- Personal information redaction — Strip names and contact details out of the extracted text before sharing it