Skip to content
OneFileConverter

How to prepare a PDF for ChatGPT and other AI assistants

Updated · 4 min read

Asking an AI assistant to summarise a report, explain a contract or pull figures out of a statement works best when it can read the document cleanly. A few minutes of preparation — getting the text out properly, cutting a long file into parts and taking out anything private — makes the answers more accurate and keeps details you didn't mean to share out of the chat.

Why preparation helps

An assistant can only work with the text it manages to read. Many accept PDF uploads directly, but how well they read the file varies: columns can get mixed together, headers and footers repeat on every page, and a scanned page may come through as a picture with no text at all. When you give the assistant clean text yourself, you know exactly what it sees.

There's also a size question. Assistants have limits on how much text they read and keep in mind at once. With a long document, parts of it may be skimmed, shortened or dropped without an obvious warning, so questions about page 140 can get vague answers. Smaller, focused pieces avoid that.

Step 1: check that the PDF has real text

Open the PDF and try to select a sentence. If you can highlight individual words, the file has a text layer and everything below works. If the whole page highlights like a picture, it's a scan: the words are part of an image. Our text tools can't read scans yet, so for those you'll need text recognition (OCR) first, or the original digital file from whoever made it.

Step 2: turn it into Markdown or plain text

Markdown is plain text with a few simple marks — '#' for headings, '-' for list items — that AI assistants understand well. It keeps the outline of the document, so the assistant can tell a section title from the paragraph under it.

PDF to Markdown reads each page in reading order, one column at a time on two-column pages. Lines set noticeably larger than the body text become headings, lines starting with a bullet or a number become list items, and lines wrapped by the page width are joined back into paragraphs. You can copy the result straight from the page or download a .md file. Tables come out as lines of text and images are skipped, so check any figures you'll ask about.

Convert a PDF to Markdown in your browser, ready to paste into a chat:

If you'd rather keep the text exactly line by line — for example to ask about a specific page — PDF to Text writes a plain .txt file with a '--- Page N ---' line before each page's text. Those markers let you ask questions like 'What does page 12 say about notice periods?' and check the answer against the original.

Step 3: split long documents

For anything longer than a few dozen pages, work in parts. Split PDF can cut a file by page ranges you type — such as '1-20, 21-45, 46-' for chapters — or into equal chunks with Every N pages. Convert each part, then ask about one part at a time.

  • Split where the document naturally breaks, such as chapters or sections, so each part makes sense on its own.
  • Start each new message by saying which part it is ('This is section 3 of 5, pages 46–70').
  • Ask for a short summary of each part, then ask follow-up questions using those summaries.
  • Leave out pages nobody needs — appendices, terms you already know, blank pages — before converting.

Step 4: take out personal details first

Whatever you paste or upload goes to the service running the assistant, and depending on the product and your settings it may be stored or reviewed. Read the provider's privacy terms, and when in doubt, don't share what you wouldn't want to leave your hands: names, addresses, account and ID numbers, health details, other people's information.

For pasted text, you can simply edit those details out before sending. For a PDF you plan to upload, run it through Redact PDF first: list the names and numbers to hide and switch on all email addresses and phone numbers. Pages with matches are rebuilt as images with the details blacked out, so the hidden text isn't in the file at all. Note that those redacted pages then have no text for the assistant to read — so for text you'll paste, convert the original and edit the output instead.

Tip: Replace names with roles rather than deleting them: 'the tenant' and 'the landlord' keep a contract easy for an assistant to follow, without identifying anyone.

Check what comes back

AI assistants can misread figures, mix up sections or state things the document doesn't say. For anything that matters — amounts, dates, obligations — ask the assistant which page or section its answer comes from, then look it up in the original PDF. Treat a summary of a legal or financial document as a starting point for reading it, not a replacement for advice.

Questions

Is it better to upload the PDF or paste the text?

Either can work. Pasting text you've extracted yourself means you know exactly what the assistant reads, which helps with multi-column layouts and long files. Uploading is quicker for short, simple documents.

Why does the assistant say it can't read my PDF?

The PDF is most likely a scan, with pictures of pages rather than text. Text extraction can't help there — the file needs text recognition (OCR) first, or ask the sender for the original digital document.

Does converting a PDF to Markdown upload it anywhere?

Not with PDF to Markdown: the conversion runs in your browser on your device. The text only leaves your device when you paste it into an assistant yourself.

Tools in this guide

All guides →