Give ChatGPT the text of your PDF, not the raw file

Attaching a PDF works, and for a short document it is the right thing to do. Converting it first earns its keep on longer ones: you see the text the model will see, cut what does not matter, and keep the result for next time.

Drop your PDF or Word file here

or choose a file from your computer

PDF or Word (.docx) · up to 50 MB

How do I get a PDF into ChatGPT?

Convert the PDF to Markdown at mypdf2md.com, then paste the text into a conversation, add the file to a custom GPT's knowledge, or send it through the API. Conversion runs in your browser, so nothing is uploaded.

Does converting a PDF use less context than attaching it?

Usually. A PDF is read as the extracted text of each page plus a picture of that page, so every page arrives twice over. Markdown sends the words alone — though those page pictures are also how charts reach the model.

What does a PDF lose when it is converted for ChatGPT?

Figures are dropped, and equations come through as whatever characters the PDF stored rather than as LaTeX. Headings, lists, tables, code blocks and emphasis all survive. A scan holds no text to convert and needs the OCR converter first.

How to convert a PDF for ChatGPT

  1. 1

    Convert the file

    Drop the PDF onto the converter above. PDF.js parses it in a Web Worker on your own machine.

  2. 2

    Check the output

    Read the Markdown against the preview: heading levels, table rows, and the order of anything in columns.

  3. 3

    Trim it

    Cut cover pages, revision tables and appendices; the counter under the editor shows what is left.

  4. 4

    Hand it over

    Copy the Markdown into a conversation, or use Download .md for a custom GPT's knowledge or an API request.

Why the text alone goes further

A PDF is not read the way a text file is. OpenAI's documentation describes the process: on models that can read images, the API extracts the text of each page and sends a picture of that page alongside it. Every page arrives twice, once as words and once as pixels.

That is why a long PDF fills a context window faster than the same words as Markdown. Markdown carries the text alone: headings stay headings, tables stay tables rather than collapsing into loose numbers, and repeated page furniture is gone. What structure buys a model sets out why that helps beyond the saving.

The page picture is not waste, though: it is how a chart or diagram reaches the model at all, and this converter extracts text, not images, so a figure leaves its caption behind and nothing else. When the picture carries the argument, attach the original PDF alongside the Markdown.

  • The text is yours to read and edit before the model sees it
  • Running headers, footers and page numbers are dropped
  • Tables come out as GitHub Flavored Markdown pipe tables
  • Nothing is uploaded: the conversion runs in a Web Worker on your machine

Getting the Markdown into ChatGPT

In a conversation, paste it straight into the message box. ChatGPT reads Markdown natively, so it needs no wrapper. Put the document first and your question after it.

For a custom GPT, use Download .md and add the file to its knowledge, so every conversation with it can draw on the document instead of you pasting again. Markdown is plain text, so if an upload refuses .md, saving it as .txt changes nothing about the contents.

Through the API, OpenAI's file search lists .md among the formats it accepts, as text/markdown, so a converted document drops straight into a vector store. For retrieval rather than a single send, it has to be cut up first, and where you cut it decides what comes back.

How much to send

A converted document is text, so its size is easy to judge: the editor counts words and characters as you trim — a better measure of what you are about to spend than a page count, since pages vary wildly in what they hold.

Trimming is the cheapest quality improvement available. A long specification usually holds a dozen pages bearing on your actual question; the rest is cover sheets, revision history and boilerplate the model would otherwise weigh against what matters. Cut on the heading ladder, so each piece keeps a title.

The free converter on this page reads the first 30 pages of a document, and says so when one runs longer. Password-protected files are accepted, and the password is used by the PDF reader in your browser and never sent anywhere.

Documents that convert badly

A scan is the clean failure: a picture of text and no text, so nothing can be extracted and the result is nearly empty. It needs optical character recognition — a separate converter here, because it is the only one that uploads, with its own things it quietly drops.

Two-column papers are the subtler one. The converter finds the vertical gutter and reads left before right, but a full-width table or wide equation crossing the middle defeats it, and the damage shows within a few lines — putting the order back is quick once spotted.

Tables usually survive and still deserve a glance, especially where the original merged cells or split one across a page; repairing the crooked ones takes a few pipe characters. Equations arrive as stored characters rather than LaTeX, so dense mathematics converts poorly.

Footnotes give no warning at all: they are ordinary text at the foot of a page, so they convert into ordinary paragraphs, landing between the sentence that cited them and what followed.

Frequently asked questions

Does ChatGPT accept a .md file?
OpenAI's file search documentation lists .md among its supported formats, as text/markdown. For ordinary uploads Markdown is plain text either way, so saving the file as .txt changes nothing about what is in it.
How many pages does the free converter handle?
It converts the first 30 pages of a document, and tells you when one runs longer. There is no cap on how many documents you convert, and no account is needed.
Is anything sent to OpenAI while the PDF is converted?
No. The conversion happens in your browser, and no request carries your document to OpenAI, to us, or to anyone else. What you do with the Markdown afterwards is a separate step.
My scanned PDF converted to almost nothing. Why?
A scan is a picture of a page, so it carries no text layer and there is nothing to read. Use the scanned PDF converter instead: it recognises the text on our server, says so first, and is the only page that uploads.

Related converters

Further reading