Which file format does ChatGPT handle best?

Ranked, with the reasoning

Why plain text wins

Every other format has to be parsed before ChatGPT sees any words. That step is where things go wrong: a PDF with an unusual layout, a Word file with corrupted styles, a spreadsheet whose header row is merged across four cells.

Plain text has no such step. The file is the words.

It is also tiny. A Word document full of images can be forty megabytes; its text is a few dozen kilobytes. Given that images are discarded on every plan except Enterprise, converting throws away nothing that was going to be read.

What you lose by converting

Being fair about it, because the answer is not always convert.

Layout. Column structure and page design go. For a report or contract this rarely matters, since you wanted the words.

Table structure. Tables can flatten into runs of text. If a table is central to your question, keep the original format, or paste the table separately so its structure is unambiguous.

Images and charts. Discarded either way on every plan except Enterprise, so converting loses nothing extra here.

Headings. Plain text keeps the heading text but not its heading status. Markdown solves this neatly, which is why .md is worth preferring over .txt for structured documents.

The scanned PDF trap

Worth calling out because it is the most common wasted upload.

A text-based PDF and a scanned PDF look identical. One works perfectly and the other returns nothing at all.

Test it in two seconds: try to select a sentence. Words highlight, and it is real text. A rectangle draws over the page, and it is an image, which means OCR before ChatGPT can do anything with it.

What to do with each format

Long Word document. Upload as-is unless it is large. If it is, save as plain text, since the weight is images.

Scanned PDF. OCR first. Nothing else works.

Spreadsheet. Export the sheets you need to CSV. Far smaller and exempt from the token cap.

Slide deck. Upload it, and expect the text only. If your question is about a chart, screenshot the slide and upload it as an image alongside.

Many code files. Concatenate into one text file with a header line naming each file, so the structure survives and it costs one upload rather than many.

The limit format does not fix

Converting solves size problems. It does not solve length.

If your document is past 2 million tokens, plain text is exactly as long as the original, because the words are the same. You will still be truncated, and still without a warning. That needs sending less at a time rather than a different file extension.