Can ChatGPT read .eml and .msg email files?

OpenAI describes what it accepts as "All common file extensions for text files, spreadsheets, presentations, and documents." Email messages are not in that sentence in either direction. Nothing says they are supported and nothing says they are excluded, which is why the answer comes down to what the format actually is.

Why one works and the other does not

Open an .eml in Notepad or TextEdit. You will see a block of headers, then the message body in readable words. That is all an .eml is: a MIME message saved to disk as text. The same test that decides every format applies here. If you can open it in a text editor and read it, so can ChatGPT.

Now open a .msg the same way. You get binary noise with occasional fragments of text scattered through it. Outlook stores messages in a compound binary container of its own, and reading it takes software written for that specific format. ChatGPT has no reason to have that, so a .msg either gets rejected on upload or comes back as an unusable mess.

Turning a .msg into something readable

Three routes, in order of how little you lose.

1. Save as Text Only from Outlook

   In the classic desktop app, open the message, then File, Save As, and pick Text Only for the file type. You get a .txt with the headers and the body. Nothing that ChatGPT would have used is lost.

2. Forward it to yourself and save as .eml

   Most webmail clients let you save a message as .eml. That keeps the structure and stays readable, which .msg never was.

3. Print to PDF

   Worth it only when the visual layout matters, for instance a formatted invoice in the body. Otherwise it is a heavier file carrying the same words.

If you do not have Outlook at all, ask the sender to forward the message rather than send the .msg again. A forwarded message you can save as text in any client.

Attachments inside an email do not come out

This is the part that catches people, and it applies to .eml files that upload perfectly well.

Attachments are stored inside the message as base64 encoded text, which is roughly a third larger than the original file. That encoded block is not detached and it is not decoded. A 4 MB PDF attached to a two line email becomes several megabytes of characters that mean nothing, sitting in a file ChatGPT otherwise reads fine.

So the email body gets read and the attachment does not, and nothing tells you which happened. If the attachment is the point, save it out of the email and upload the PDF or the spreadsheet on its own. If the covering message matters too, upload both as separate files.

Whole mailboxes are a different problem

Gmail Takeout hands you .mbox, which is readable text in principle. Size is what stops it. A real inbox exported this way runs to hundreds of megabytes, and even under the 512 MB file cap you are far past 2 million tokens, so most of it is silently ignored.

For anything at mailbox scale, extract the fields you actually want, usually date, sender, subject and body, into a CSV. A year of email reduced to those four columns is a manageable file, and it is far easier to ask questions about than a raw dump.

Strip the noise before uploading

Email is mostly repetition. In a thread of thirty replies, the same quoted trail appears thirty times, along with thirty signatures and thirty legal footers. That can be the majority of the file, and every character of it competes with the content you care about.

Before you upload a thread, delete the quoted history under each message and keep only what was new in that reply. Delete the routing headers as well, the Received and DKIM lines, since they say nothing about what the email means.

One more reason to bundle: sixty separate .eml files is sixty uploads, against 3 file uploads per day on Free and 80 files every 3 hours on any plan. The same sixty messages pasted into one text file in date order is a single upload, and the chronology survives, which it does not when files arrive one at a time.

What to do, in one line

For a single email, save it as .txt. For a thread, export the whole thread into one text file with the quoted trails removed. For a mailbox, pull the fields you need into a CSV. And for attachments, always save them out and upload them separately, whatever the email format was. Fighting a .msg is never worth it when saving it as text takes ten seconds.