How to shrink a file so ChatGPT will accept it

The per file cap is 512 MB, which almost nobody reaches. The caps people actually hit are the smaller ones: about 50 MB for spreadsheets, 20 MB for an image, and whatever your connection gives up on before the upload finishes.

First work out which limit you hit

Two different problems send people looking for the same fix, and only one of them responds to it.

Length is capped at 2 million tokens, and a plain text file can sail past that while weighing a couple of megabytes. If that is your problem, every technique below is wasted effort.

Export to plain text

The largest single reduction available, and usually the only one you need.

Open the document, choose Save As or Export, and pick .txt or .md. A 40 MB Word file full of screenshots commonly comes out in the tens of kilobytes as text. The words were never the weight.

What it costs you is smaller than it looks. Retrieval from documents is text only, except Enterprise, so converting throws away nothing that was going to be read anyway. Page layout and column structure go, which rarely matters for a report or a contract. The real loss is table structure, which can flatten into runs of text. If a table is central to your question, paste it separately so its rows stay unambiguous.

Markdown is worth preferring over plain text for anything with headings. It keeps the hierarchy visible at no extra size.

Compress a PDF

Whether this works depends entirely on what kind of PDF you have.

A text-based PDF is mostly text and embedded fonts. Compressing it saves very little, because there was nothing heavy in it to begin with. Export it to text instead.

A scanned PDF is a stack of page photographs, and that is where the megabytes live. Running it through a compressor that downsamples images to around 150 dpi will make a real difference, since 600 dpi pages are many times the size of the same pages scanned lower.

Strip images and embedded media

For files you want to keep in their original format.

In Word, select an image and use Compress Pictures, applying it to all images at email resolution. In PowerPoint, the same option lives under Picture Format. Embedded video in a deck is often the single largest object in the file, and removing it costs nothing on upload, because video is not among the file types OpenAI lists as supported.

This is the option to reach for when someone else needs the file to stay a DOCX. If nobody does, exporting to text is faster and shrinks it further.

Export spreadsheets to CSV

An XLSX carries formatting, formulas, chart objects, pivot caches and every sheet in the workbook. CSV carries the values.

Export only the sheets you need. The file usually drops by an order of magnitude, and spreadsheets are exempt from the token cap that truncates documents, so a large CSV often goes through whole where the same content as a document would not.

The cost is real but narrow. One sheet per file, formulas become their calculated results, and formatting disappears. For analysis that is all fine, since you wanted the numbers.

What does not shrink anything useful

Zipping. ZIP archives are not opened during upload. Smaller file, zero readable content.

Renaming the extension. Calling a .docx a .txt does not convert it. ChatGPT sees binary noise and the file size is identical.

Screenshotting pages. This turns text into images, which are both larger and unreadable. It is the exact opposite of what you want.

Compressing images below legibility. If you are shrinking an actual image for ChatGPT to look at, push it too far and the model cannot read the small print in it either.

Which one to reach for

If it is a document and you want the words, export to plain text. That covers most cases and takes about ten seconds.

If it is a scan, OCR it, which shrinks and fixes it at once. If it is a spreadsheet, export the sheets you need to CSV. If someone needs the original format preserved, strip the images inside it.

And if the file is already small and ChatGPT is still only answering about the first part of it, stop shrinking. That is length, and the answer is to send it in pieces.