How to upload large files to ChatGPT: 4 limits, 4 fixes

"Large" turns out to mean four different things in ChatGPT, and picking the wrong fix wastes an afternoon. Here is how to tell them apart and what to do about each.

The four limits, side by side

Each one carries its own number and fails in its own way. Find the row that matches what you are seeing, then jump to the fix in the last column.

Spreadsheets are the exception worth knowing. CSV and spreadsheet files are capped at about 50 MB instead, and they are exempt from the token limit entirely, so a very long spreadsheet often goes through where a document of the same length would not.

The second row is the one worth internalising. It is not a size limit, it is a length limit, and because it fails silently it is the reason people think ChatGPT read their document when it did not.

Step 1: convert it to text

Try this before anything more elaborate. It is free and it takes a minute.

Most of a document's weight is not its words. A Word file carries fonts, styles, and revision history. A PDF carries page geometry and any embedded images. Strip that away and what is left is text, which is small.

1. Open it in whatever created it

   Word, Google Docs, Pages, anything.

2. Export as plain text

   Save As in Word. File, then Download, then Plain Text in Google Docs.

3. Compare the size

   A Word document full of images routinely goes from tens of megabytes to tens of kilobytes.

Two things this does not fix. It will not help a scanned PDF, because there is no text in it to extract. And if the file is long rather than heavy, the word count is unchanged, so you may still be over the token cap.

Step 2: send only what you need

Consistently underrated. You rarely need the whole document read. You need the part your question is about.

Pull out that section and send that. Answers usually improve, because there is less irrelevant material competing for attention.

Step 3: split it and send the parts in order

When the document really does need reading in full, this is the method that works, and it is the core of every approach to bypassing the ChatGPT file upload limit.

Cut it into pieces small enough to be accepted, then send them one after another. Keep each part under 10,000 characters, because past that ChatGPT converts the paste into an attachment rather than a message, and an attachment spends one of the upload slots you were trying to protect. The critical part is not the cutting, it is what you say first:

I am going to paste a long document in several parts. Do not summarise or answer anything until I tell you the last part has arrived. Reply "received" after each one.

Skip that and ChatGPT answers the first chunk as though it were the whole document, then keeps adjusting. What you end up with is a summary weighted toward the opening that reads as though it covers everything.

Doing this by hand is fine for three or four parts. Past that the failure mode is a mistake you do not notice: a chunk pasted twice, one skipped, one silently truncated by the paste.

Step 4: know when to use something else

For a book, a full codebase, or a year of transcripts, chunking stops making sense. Even in order, the earliest parts fall out of memory before the last arrive.

At that size, use a tool built for it. Assistants with much larger context windows exist, and retrieval tools that search a document and surface the relevant passages will beat chunking every time.

Things that do not work

That last one is worth remembering. OpenAI's FAQ confirms failed attempts count against the rolling cap, so twenty retries makes your next hour worse.

Checking it actually worked

The habit that saves you from a confidently wrong answer: after uploading, ask something that can only be answered from the end of the document.

"What is the last section called?" If it cannot tell you, it did not read that far, and everything it says about the document is based on part of it.