How to Upload PDF to Claude
You can upload PDF straight into Claude, but if your file has tables, multi-column layout, or scanned pages, convert to Markdown first. You will get cleaner retrieval and fewer parsing surprises.
Getting PDF content into Claude sounds like it should be a one-click job, and sometimes it is. Claude does accept PDF uploads directly, but "accepts" and "handles well" are two different things. This guide walks through the whole path: where to drop the file, what usually breaks, and when it is worth converting to Markdown before you upload. PDFs are the hardest case here, because two files with the same extension can behave completely differently: a born-digital export converts almost perfectly, a scan converts to nothing at all unless OCR runs. Claude takes uploads via Paperclip attachment in a chat, or the Project Knowledge panel inside a Project, caps individual files around 30MB, and works with roughly 200,000 tokens of context on its default variant. The short version: give Claude clean structure — headings, lists, tables that survived the trip — rather than a raw PDF and the hope that the parser does the right thing.
Why convert to Markdown before uploading
The case for a Markdown intermediate step, specifically for PDF in Claude.
- Markdown is token-efficient. You can fit substantially more of a PDF document inside Claude's context window (currently around 200,000 tokens, give or take depending on the plan and model variant) once formatting overhead is gone.
- Claude respects Markdown structure. Headings become navigable sections, lists stay as lists, and tables that survive conversion stay readable for the model.
- You get a predictable failure mode. When a PDF parser inside Claude silently drops a page or column, you rarely notice until the answer is wrong. With Markdown, what you see is what the model sees.
- Batch and reuse get easier. A Markdown file drops cleanly into Custom GPTs, Claude Projects, NotebookLM notebooks, or a RAG pipeline without per-tool quirks.
- Scanned PDFs are a special case — most AI tools skip OCR entirely on image-only PDFs. Running an OCR-aware conversion first turns a "blank" PDF into usable text.
Claude is especially strong at parsing Markdown structure — headings become navigable sections, and the long context window stays clean. Converted PDF files come through at around 95% text accuracy, with table fidelity we would call good. The usual catch with PDF: oCR accuracy varies with scan quality. Convert once and reuse the Markdown across Claude chats rather than re-uploading the original every time.
Step-by-step: PDF into Claude
- 1
Convert the PDF file to Markdown
Open markdownconverters.com/conversion/pdf, drop your PDF file in, and download the .md output. If the pages are scans or photos rather than born-digital text, switch on AI Vision so the text is actually extracted instead of silently coming through empty.
- 2
Skim the Markdown before you upload
Open the output in any text editor. Look for pages that came through blank — that is the signature of an image-only page the OCR missed. Check any two-column pages read in the right order. Thirty seconds here saves an hour of debugging answers that were never going to be right.
- 3
Upload to Claude
In claude.ai, open the chat or Project where you want the file. Click the attachment icon and upload the .md file. Inside a Project, use Project Knowledge so Claude keeps it in context for every new chat.
- 4
Prompt with structure in mind
Reference the file by name, then point at a section rather than asking broadly. That means quoting the heading: "In the 'Pricing' section, what is the tier for..." instead of "what does this say about pricing". Claude retrieves noticeably better when you narrow it yourself.
- 5
Verify against the source
Ask Claude to quote the passage it is citing. If the quote does not appear in the Markdown, you have either a retrieval miss or a hallucination — re-check the conversion or narrow the prompt.
Gotchas with PDF in Claude
The specific failure modes we see most often — worth knowing before you spend an afternoon debugging "bad" model answers that are really upload problems.
- Scanned PDFs return blank
Claude does not run OCR on image-only PDFs. If your PDF is a scan, the upload will "work" but return empty text. Convert with OCR enabled first.
- Claude has a 30MB-per-file cap and a total project knowledge cap around 200k tokens
big PDFs chew through it fast
- Excel files sometimes get parsed as plain CSV-like text, losing multi-sheet structure
Excel files sometimes get parsed as plain CSV-like text, losing multi-sheet structure
- PowerPoint uploads are not always consistent; Claude sees text but miss visual hierarchy
PowerPoint uploads are not always consistent; Claude sees text but miss visual hierarchy
- PDFs containing form fields or scanned pages may return partial text only
PDFs containing form fields or scanned pages may return partial text only
Getting better answers from Claude
- Upload through Paperclip attachment in a chat, or the Project Knowledge panel inside a Project — that is the path Anthropic actually supports, and the one least likely to silently truncate your file.
- Claude holds roughly 200,000 tokens, about 145k words. Converted PDF content is compact, but a long document plus a long conversation still competes for the same budget.
- Stay under about 10MB per file even though the hard cap is 30MB. Retrieval quality degrades well before the limit, particularly on dense PDF content.
- Claude has a 30MB-per-file cap and a total project knowledge cap around 200k tokens — big PDFs chew through it fast.
- PDFs containing form fields or scanned pages may return partial text only.
- Claude reads Markdown structure well, so ask it to cite the heading it answered from. If it cannot name one, your content was not retrieved.
- Check the OCR output on page one before trusting the rest. A single misread "0" for "O" poisons every answer downstream, and you will not see it happen.
Frequently asked questions
Can I upload a PDF file to Claude directly?
Yes. Claude accepts PDF uploads through its standard attachment flow. That said, direct upload often loses structure — especially tables, multi-column layouts, and anything scanned. For reliable results, convert to Markdown first.
What is the file-size limit on Claude?
As of this writing, Claude typically caps individual uploads around 30MB, with a context window of roughly 200,000 tokens on the default variant. These numbers shift with releases and differ by plan, so verify in the app before relying on a specific figure. The practical limit is lower than the hard cap anyway — retrieval quality starts to degrade well before you hit it, especially for dense technical documents. Smaller, focused files beat one enormous one.
Does converting PDF to Markdown lose information?
Markdown drops visual formatting — fonts, colors, exact page layout — because it is a semantic format. It preserves headings, lists, tables, and text content, which is what LLMs actually use. For PDF specifically, the things that matter (Scanned documents, Published papers and books) survive cleanly. Things like decorative styling do not, and usually should not.
How many PDF files can I upload to Claude at once?
Claude Projects cap Project Knowledge around 200k tokens total across all files. A single chat accepts multiple attachments, but the ~30MB per-file limit is the bottleneck for large PDFs.
Is this safe — what happens to my PDF file?
When you upload to Claude, the content is processed by Anthropic's infrastructure under their terms. If the document is sensitive, consider using the enterprise tier (where retention and training policies differ), or convert locally where possible. Markdown Converters does not retain your uploaded files beyond the conversion window.
Does Claude run OCR on scanned PDFs?
No, not in any dependable way. Claude treats image-only PDFs as if they were empty. Run OCR during the Markdown conversion so the text is actually extracted before the file ever reaches Claude.
Keep going
Drop PDF files in and download Markdown ready to upload to Claude.
Check how much of your converted PDF will actually fit inside Claude's context window.
Same workflow, tuned for ChatGPT's upload quirks and context window.
Same workflow, tuned for Gemini's upload quirks and context window.
Working with Word instead of PDF? The gotchas are different.
Background on why clean Markdown consistently outperforms native uploads across every major assistant.
Convert PDF for Claude in one step
Clean Markdown output ready to drop into Claude. Free tier, no account required to try.