Uploading files to AI tools, without the headaches

Why uploads to AI tools go sideways

Every few weeks someone tells us the same story. They dragged a PDF into ChatGPT, asked a question, got a confident answer, and only later discovered the number was wrong — because the PDF had a two-column layout and the model read across columns instead of down. Or they uploaded a scanned contract to Claude and it summarized what it could read, quietly ignoring the pages that were images of text. Or they pushed a 400-page deposition into Gemini, watched it work, and got a summary that skipped three chapters without ever saying so.

None of these tools are broken. They are doing exactly what they were designed to do. The problem is that the file you uploaded is not what the model actually sees. A PDF is a print-layout format. A DOCX is a box of XML. A slide deck is a pile of floating text boxes. Before any AI can reason over them, something (or someone) has to extract text, decide what is a heading versus a footer, parse tables, and OCR the scanned pages. Different tools do this differently, and each one has quirks worth knowing about.

Three things tend to go wrong:

• Format issues. Tables come out as word salad. Multi-column PDFs get read left-to-right across columns. Footnotes appear in the middle of paragraphs. Headers and footers repeat every page.
• Token waste. A messy PDF extraction can be 2–3x larger than the same document as clean markdown, eating your context window and raising the odds the model drops something important before it answers your question.
• OCR failures. Scanned PDFs are images. Some tools OCR them automatically, some partially, some not at all — and the failure mode is usually silent. You get an answer, but the model is guessing from the file name and the few pages it could read.

Converting to markdown first sidesteps most of this. Markdown is just text with a bit of structure — headings, lists, tables, code blocks. It is how language models were trained and what they read most fluently. Instead of asking ChatGPT to reverse-engineer a PDF on the fly, you hand it something it can read in one pass. Our PDF to Markdown converter handles the extraction and OCR in one shot, and the case for Markdown with AI walks through why the token savings add up.

That does not mean you should convert every file. Sometimes native upload is fine — a short Word doc, a simple receipt image, a CSV under a few thousand rows. The guides below walk through each major tool: what it handles well, what it quietly breaks, and when preprocessing with markdown is worth the extra step.

Pick your tool