Cut your AI platform fees by up to 66% on company PDFs
You stop paying your AI provider to look at pictures of your pages. Attach a PDF and you're charged for an image of every page as well as the words on it. Converting to Markdown strips that out. Same analysis from your AI, up to two-thirds fewer tokens, every message.
Convert a PDF fileThree free, no account, nothing stored.
What you get
- The text, in reading order, page after page.
- One markdown file that pastes into Claude, ChatGPT, Gemini or Copilot.
- The meter, showing what that file would have cost attached and what it costs now.
Before you drop it in
- Two-column layouts can interleave. Reports and journal articles are the usual culprits. Read the first page before you paste.
- A 150-page PDF takes a few seconds. The page hasn't frozen.
- The meter's attached figure is measured on Claude. OpenAI's API renders pages the same way, so the estimate for it sits alongside.
What survives, what doesn't
| Body text | ✓ Kept | Paragraphs come through in reading order. |
| Headings | △ Partly | PDFs don't store heading levels. Headings arrive as ordinary lines, so add # marks yourself if you need them. |
| Tables | △ Partly | Rows come through as lines of text, not as markdown tables. |
| Images and charts | ✗ Lost | Dropped. The text near them stays. |
| Scanned pages | ✗ Lost | A scan with no text layer has nothing to extract. Run OCR first and we'll flag it if you haven't. |
| Headers and footers | △ Partly | They repeat on every page, exactly as the PDF repeats them. |
What it looks like
A page from an enterprise agreement comes out like this:
Enterprise Agreement 2026 1. Title This agreement is called the Northern Civil Enterprise Agreement 2026. 1.1 The agreement covers employees of the company engaged in civil construction work in New South Wales. 1.2 It starts seven days after approval by the Fair Work Commission.
Questions about PDF
Why does a PDF cost more tokens than the words in it?
Because most AI tools render every page as an image as well as extracting the text. On Claude that image is about 1,500 tokens a page. A 60-page PDF we measured cost 142,000 tokens attached and 47,000 as markdown.
Do headings come through?
As plain lines. A PDF stores where text sits on the page, not what it means, so there's no heading level to read. Word and PowerPoint files do carry that, and their headings convert properly.
My scanned PDF came back nearly empty.
It has no text layer. OCR it first with Adobe or your scanner's software, then convert. We show a warning rather than a fake 99% saving.
Does this work with ChatGPT, Gemini and Copilot?
Yes. Markdown is plain text, so every AI tool reads it. We measured the saving on Claude and estimate it for the OpenAI API, and say which is which on every number.
Do you keep my PDF?
No. It's converted in memory and sent straight back. There's no copy on our side once the response is sent.
Drop your PDF file and see what it really costs your AI.
Three free conversions. No account. Nothing stored.
Convert a PDF file