How to send large files to ChatGPT

ChatGPT accepts files up to 512 MB each, with a much smaller practical limit on how much text it can read at once, so large documents get truncated rather than refused. Split the file into sections, or convert it to plain text or Markdown first, and send the parts that matter instead of the whole archive.

Short answer

Two different limits get confused here, and knowing which one you have hit tells you what to do about it.

The upload limit and the reading limit

The upload limit is generous. A single file can be up to 512 MB, with tighter caps on spreadsheets and on how many files you attach at once.

The reading limit is the one that actually bites. However large the file is, only so much text fits into the model's context at a time. A three hundred page PDF uploads happily and then only partly informs the answer, which is why people get confident summaries that miss half the document. Nothing errors. That is what makes it dangerous.

What to do about a genuinely big file

Split it by section rather than by page count, so each part is self contained
Strip the parts you do not need, appendices, boilerplate, repeated headers
Convert to plain text or Markdown, which removes layout overhead and fits far more real content in the same space
Summarise each part first, then give it the summaries together for the cross cutting question
Say explicitly which section you want it to use, rather than hoping

Spreadsheets and data

A large CSV is worse than a large document, because rows are meaningless out of context and the model sees only part of the table. Filter and aggregate first in a spreadsheet, then give it the smaller result. Ask for the analysis of a summary, not of a million rows.

If the large file is your own chat archive

This is a common version of the question. People want to feed their entire ChatGPT export back in, to ask what they worked on last year. conversations.json from a heavy user runs to hundreds of megabytes of deeply nested JSON, and it is close to the worst possible input: mostly structure, not readable text, and far too big to read at once.

Convert it first. Drop the whole ZIP into the converter and you get one readable document per conversation, or a single CSV with one row per message. From there:

Search the list to find the handful of conversations that matter and attach only those as Markdown
Or use the CSV in a spreadsheet to count and filter, which answers most "what did I do" questions without the model at all

The conversion runs inside your browser tab, so your archive never leaves your device.

Pictures and screenshots

Images have their own limits and a big screenshot can be downscaled to the point where small text becomes unreadable, which is why it sometimes misreads them. Crop to the part you care about rather than sending a full desktop screenshot. If what you need is the text, read the image on your own device with the screenshot reader and paste the text, which is smaller, exact and private.

Frequently asked questions

What is the maximum file size ChatGPT accepts?

Up to 512 MB per file, with lower limits for spreadsheets and for the number of files attached at once. Reading the whole thing is a separate and much smaller limit.

Why does ChatGPT miss things in a long PDF?

Only part of the text fits in context at a time. The upload succeeds and the reading is partial, with no error to tell you.

How do I send my whole ChatGPT history back to ChatGPT?

Do not send the raw export. Convert it first, then attach the few conversations that matter as Markdown, or analyse the CSV in a spreadsheet.

Is it better to send text or PDF?

Plain text or Markdown. They carry no layout overhead, so far more real content fits into the same context window.