Limits by plan
Every ChatGPT attachment limit: 20 a message, 3 a day on Free
5 attachment caps stack: 20 per message, 3 a day on Free, 80 every 3 hours, per project, per GPT. A 30 second diagnosis for which one stopped you.
102 guides, sorted by when we last checked the facts in them. Browsing by subject instead? Start here.
Limits by plan
5 attachment caps stack: 20 per message, 3 a day on Free, 80 every 3 hours, per project, per GPT. A 30 second diagnosis for which one stopped you.
Limits by plan
The oldest part of the conversation stops being sent, with no error and no marker in the reply. What survives, what goes first, and the test that proves it.
Working with documents
12 tips, and the first is a 30 second check, because long files are truncated with no warning. Then how to force quotes and stop invented answers.
Limits by plan
Only Enterprise gets visual retrieval for PDFs. Free, Go, Plus, Pro and Business extract text and discard images, which is why charts and scans return nothing.
What ChatGPT accepts
ChatGPT reads 4 file families and OpenAI publishes no extension list. 7 uploads go through cleanly and give it nothing to read, with the fix for each.
Limits by plan
ChatGPT Free: 3 file uploads a day, 500 MB Library storage, 5 files per project. What still works when you are out, and what Go at $8 changes.
Getting past the limits
The limit cannot be raised, so all 5 fixes send less at a time. Paste in parts under 10,000 characters, plus the 4 tricks that only look like they work.
Error messages explained
Hit the limit having uploaded 1 file? Failed attempts count toward the 80 per 3 hours cap. Every ChatGPT upload error here, with the fix that clears it.
Error messages explained
Nothing deleted the file. The chat outgrew its context window, so the oldest content stopped being sent. Re-attach it, and the setup that stops it recurring.
Comparisons
2 of the 8 upload caps move when you pay. The 2 million token cap that truncates long documents does not. Work out which wall you hit before spending $20.
Working with documents
Ten files in one conversation blur into one vague source. Label each on upload, order them deliberately, and know when to open a second chat.
Limits by plan
No published character cap, but paste over 10,000 characters and ChatGPT turns it into a file upload, spending an upload you never meant to spend.
Working with documents
ChatGPT does not open archives, and a Notion export is a zip of markdown files. Unzip it, join every page into 1 text file with headers, upload that.
Error messages explained
20 files per message is the web cap since February 2026, so this is a per message block, not your account or your quota. Send 1 per message, or bundle them.
Limits by plan
Plus gives 80 files every 3 hours instead of 3 a day, plus 25 files per project. The 2M token cap is identical on every plan, so long files still get cut.
Working with documents
Free gets 5 project files, Plus 25, Pro 40. What belongs in project files, what to leave in a single chat, and why a smaller set gives sharper answers.
Limits by plan
Free 5, Go and Plus 25, Pro and Business 40 files per project. OpenAI's own FAQ still prints 20 for Plus, and the project allowance is not your chat one.
Comparisons
Both hold files that persist. Projects hold more and stay personal, custom GPTs hold ten for life and can be shared. Which fits your documents.
Limits by plan
25 GB per user, and a separate 500 MB Library cap on Free that bites first. Deleting old chats no longer frees either one, so here is what still does.
Limits by plan
512 MB per file, 2M tokens, 25 GB per user: identical on Team, Business and Enterprise. Only Enterprise changes what ChatGPT can read inside a PDF.
Limits by plan
No. 512 MB per file and 2M tokens are checked at upload, before any model sees the file. The 1 row in 9 that the model does change, and what to change instead.
Error messages explained
2 documented causes: failed uploads count toward the 80 per 3 hours cap, and 25 GB of storage never clears on its own. Deleting chats no longer frees it.
Limits by plan
OpenAI publishes no reset time for the Free cap of 3 a day. The general cap of 80 files rolls over 3 hours, freeing slots gradually, so an hour gets you moving.
Working with documents
Consumer plans: yes, unless you switch it off in data controls. API and Enterprise are excluded. Files go in 30 days, but deleting the chat no longer does it.
Error messages explained
A negative count is a display fault, not a real limit: your allowance minus whatever is already counted. 5 things push it under zero, and the fixes in order.
Limits by plan
10 files per GPT, and that is a lifetime cap rather than per conversation. They also eat your 25 GB storage, and GPT creation has left personal accounts.
Working with documents
Files past 2 million tokens are truncated with no warning. One question about the final page tells you how much ChatGPT really read, in 30 seconds.
Getting past the limits
Copy the text, do not upload the .eml or .msg. Paste over 10,000 characters and ChatGPT makes it an attachment anyway. What to strip out of the thread first.
Limits by plan
20 per message on web, 3 a day on Free, 80 every 3 hours, 10 per custom GPT. Six caps that stack, plus the one that never resets however long you wait.
Limits by plan
How many PDFs you can upload on Free, Plus, Pro, in Projects and to a custom GPT, plus the length cap that decides how much of a long PDF gets read.
Getting past the limits
Free gives 3 uploads a day and the cap cannot be raised. 5 ways to spend fewer of them, including the one route that does not count as an upload.
Getting past the limits
Paperclip on desktop, plus button on mobile, question in the same message. Then the one check that tells you whether ChatGPT read the whole PDF or just page 1.
Getting past the limits
Keep each part under 10,000 characters, cut at section boundaries, and send the wait-for-all-parts instruction first. Past 10,000 a paste becomes an upload.
Getting past the limits
512 MB per file, 2M tokens per document, 3 a day on Free. Four separate limits stop a big file, and the one that stops most people shows no error.
Working with documents
There is no pin or lock for an attachment, so documents drop out of long chats silently. How to set the chat up, re-anchor when answers slip, when to restart.
Getting past the limits
Pasting is a message, not an upload, so no daily, rolling or storage cap touches it. But paste over 10,000 characters and ChatGPT converts it to a file anyway.
Getting past the limits
Concatenate the 3 to 6 files that matter into 1 text file with path headers. ChatGPT never unpacks a zip. The one line command, and what to strip out first.
Getting past the limits
512 MB is the ceiling, but a 3 MB PDF can still be too long: text caps at 2 million tokens. How to tell truncation from rejection, and the 4 fixes.
Getting past the limits
20 files per message on web since February 2026, not 10. Projects cap at 25 on Plus and a custom GPT at 10, plus the trick that sends 12 files as one.
What ChatGPT accepts
Plain text wins, because there is no parsing step to go wrong. All 9 common formats ranked, plus the 2 second test that spots a PDF ChatGPT cannot read.
What ChatGPT accepts
Not usefully. A .pages, .numbers or .key file stores its text as compressed binary. Export to Word, Excel or PowerPoint first, and here is which keeps most.
What ChatGPT accepts
EML reads fine, MSG does not: it is Outlook binary with no text layer. Save it as Text Only from Outlook, and why attachments never come out of either.
What ChatGPT accepts
Yes. An EPUB is a zip of HTML, so the text extracts cleanly. DRM stops you dead and book length breaks it quietly. How to convert one and split it by chapter.
What ChatGPT accepts
No. ChatGPT holds no Google session, so a Docs or Drive link returns a sign in page. The 30 second export route, and the format that keeps your tables intact.
What ChatGPT accepts
Yes. Logs are plain text and upload fine up to 512 MB, but answer quality collapses far below that. The jq and grep lines that cut a log to what matters.
What ChatGPT accepts
No. OpenAI lists 4 upload families: text, spreadsheets, presentations, documents. Audio is in none of them. Transcribe it, and keep the speaker labels.
What ChatGPT accepts
Yes, ODT uploads and reads fine. OpenAI publishes no extension list, so here is the test for any unusual format, and when exporting beats trying it.
What ChatGPT accepts
No. RAR, 7z, tar and tar.gz are never unpacked, so 1 upload buys nothing. How to extract each format, handle split .part1 sets, and what to send instead.
What ChatGPT accepts
Yes. RTF is a document format and uploads like any other, inside the 512 MB and 2 million token caps. The 1 case that fails, and the 10 second fix for it.
What ChatGPT accepts
Yes. SRT and VTT are plain text and upload like any other file. Strip the timestamps on anything feature length, since they take more room than the dialogue.
What ChatGPT accepts
Yes, up to 512 MB per file, and DOCX is among the most reliable formats. Every plan but Enterprise discards the images, and how to spot what was dropped.
What ChatGPT accepts
No. ChatGPT never unpacks archives, so a ZIP costs 1 upload slot and returns nothing readable. Extract it, then bundle the files into 1 text file instead.
Comparisons
3 approaches to a document too long for one upload: retrieval, a bigger context window, or splitting it. Which one fits depends on the question you ask.
Comparisons
The web app extracts text, retrieves the relevant parts and runs a Python sandbox, all inside the subscription. The API gives you none of it. What to rebuild.
Working with documents
Text extraction strips page breaks, so ChatGPT rarely knows the page. Ask it to quote the heading and the exact sentence, then search the PDF for that line.
Working with documents
Uploading both files at once blends them into mush. Summarise each in its own chat, then compare the 2 summaries in a third, and attribution stays intact.
Error messages explained
The chat is full, not the file. Messages, replies and uploaded files share one total, so the same file uploads fine in a new chat. How to move across cleanly.
What ChatGPT accepts
Spreadsheets cap around 50 MB and are exempt from the 2 million token limit that truncates documents. Why a 200,000 row CSV goes in when a report does not.
Working with documents
Ask for a review and you get praise. Ask it to argue for rejecting you, or to sit with 60 applications and 10 minutes, and the criticism finally arrives.
Limits by plan
512 MB per file and 3 uploads a day on Free apply identically in the Mac and Windows apps: one account, one pool. The one thing only a browser can do.
Working with documents
3 causes: silent truncation, sampling randomness and context eviction. How to tell which one you have in under a minute, and the question that separates them.
Working with documents
Ask for a tab indented outline rather than markdown and PowerPoint imports it straight. One idea per slide, notes kept separate, and the prompt for both.
Working with documents
Name every field, name the format, demand the source: "Return a CSV with columns date, party, amount". The 3 rules that stop ChatGPT filling gaps for you.
Working with documents
Ask for CSV, 1 row per source row, values copied exactly, empty fields for empty cells. Then check it: a misread table comes back looking perfectly clean.
Working with documents
ChatGPT has no source to consult, so its verdicts are guesses. Use it to extract and rank the claims that need checking, then the prompt that keeps it honest.
Error messages explained
4 caps produce this message: 512 MB per file, 50 MB spreadsheets, 20 MB images, 2M tokens of text. Under 10 MB means length, and compressing will not help.
Limits by plan
Yes. The ChatGPT mobile apps use the same 512MB and 2 million token caps as the web, since limits belong to your account. What a phone cannot do, and the fix.
Error messages explained
An upload that spins forever has 4 common causes. A 1 line text file in incognito tells you which one, in under a minute. Stop retrying: failed attempts count.
Comparisons
These extensions split a document into pieces ChatGPT accepts and send them in order. They cannot raise the 512MB cap or the Free 3 a day. How to judge one.
Working with documents
Export the statements to CSV rather than uploading the PDF report. Spreadsheets are exempt from the 2 million token cap. Then check every figure it hands back.
Working with documents
Extract the outline first, then hand it back as a template. Output arrives under your own section headings, in order, instead of one summary that flattens them.
Limits by plan
20 MB per image is the size cap, and rarely what stopped you. The limit on how many is, and PDF pages count toward it. What resets it, and how to send less.
Working with documents
JSON hits the 2 million token cap fast. Send the schema and a short sample, agree the approach, then convert tabular JSON to CSV, which is exempt from it.
Working with documents
Summarise a contract, judgment or policy clause by clause, never in one sweep. What ChatGPT locates reliably, and why 'no such clause' is never proof.
What ChatGPT accepts
Markdown, JSON, XML and code are plain text, so ChatGPT reads them cleanly up to 2 million tokens and 512MB. Why .md beats PDF and DOCX for a long document.
Limits by plan
512 MB is the hard ceiling and almost nobody reaches it. A 4 MB scanned PDF fails anyway, because a scan holds no text layer to extract. How to tell them apart.
Working with documents
Do not ask for a summary. Ask for 4 things: decisions, commitments with owners, open questions, disagreements. Keep the speaker labels or you lose every name.
Limits by plan
OpenAI publishes no page limit. Size stops at 512 MB, length at 2 million tokens, and length truncates a PDF silently. The question that reveals how far it got.
What ChatGPT accepts
Yes, PPTX uploads work. But every plan except Enterprise strips the images out, so charts and diagrams never reach the model. Speaker notes usually do.
Error messages explained
Every plan except Enterprise extracts digital text and discards images, and a scan is all image. A 2 second test to confirm it, then the OCR route that works.
Error messages explained
The analysis step failed, not the upload, so file size is not the cause. 3 things explain almost every case. Retry once, test a 2 line file, then check yours.
Working with documents
A chapter summary is the least useful thing to ask for. Upload the chapter and be tested on it instead, with worked examples and questions in exam format.
Working with documents
ChatGPT cannot open a YouTube link or the video file, so use the transcript. 4 steps to pull it out and paste it in, plus what a text only view misses.
What ChatGPT accepts
PNG, JPEG, WEBP and GIF all work, capped at 20 MB each. PSD, AI and RAW do not. Which to pick for a screenshot, and when text beats an image entirely.
Working with documents
A translation runs as long as its source, so it needs 2x the room a summary does. Work section by section, and keep the glossary that stops 1 term becoming 3.
Error messages explained
ChatGPT rejects the format before anything uploads, so renaming the extension never helps. Convert to PDF, DOCX, TXT or CSV, and tell 3 refusals apart.
Comparisons
ChatGPT caps document files at 2 million tokens and 512MB. Gemini publishes context windows per model, not per-file caps. Which suits your document, and why.
Comparisons
NotebookLM answers only from the sources you gave it and cites them. ChatGPT blends your document with what it already knows. Where each one lets you down.
Comparisons
Perplexity searches the web and cites it. ChatGPT reasons over the file you hand it, up to 512MB and 2 million tokens. Which fits depends on where files live.
Working with documents
Pasting a URL gets you 1 page, and only when browsing works. Save each page as text, flatten them into one upload with URL headers, and quotes stay traceable.
What ChatGPT accepts
Video is not among the 4 file families OpenAI lists for uploads. Transcribe it and upload the .txt instead, plus what to do when the visuals are the point.
Working with documents
Text files cap at 2 million tokens, so a full book will not fit in one go. Work chapter by chapter with a running notes file, and know when to stop.
Getting past the limits
Export to plain text first: a 40 MB Word file of screenshots lands in tens of KB. Then compression, image stripping, CSV, and the case where none of it helps.
Getting past the limits
Excel is capped at about 50MB but exempt from the 2 million token cap. Export to CSV, cut the columns you do not need, and send a sample if it is too big.
Getting past the limits
The cap is 512 MB per file, and a heavy deck is heavy because of images ChatGPT discards anyway. Export the slide text and speaker notes, and see what goes.
Getting past the limits
ChatGPT caps any single file at 512 MB, and a 700 MB archive is never 700 MB of text. See what is inside, pick the few files that matter, send them as one.
Getting past the limits
ChatGPT extracts no text from a scan, so run OCR first: Preview on a Mac, PowerToys on Windows, OCRmyPDF for anything long. The 2 checks that catch bad OCR.
Comparisons
Most ChatGPT extensions do 1 of 5 things, so name your problem first. The 15 second permission check rules out the risky ones before you read a feature list.
Working with documents
3 things it does reliably: plain-English translation, spotting missing clauses, finding one-sided terms. 4 it cannot. Plus the data setting to check first.
Working with documents
Upload a paper and ChatGPT praises it. The prompts that turn it on the methodology instead, the statistics it gets wrong, and where to stop trusting it.
Working with documents
Anything past 2 million tokens is cut off silently, so run the last section test before you read the summary. Then the prompts that stop it coming back bland.
Comparisons
ChatGPT caps a document at 2 million tokens and Claude holds more at once. Where ChatGPT still wins is spreadsheets, and where it loses, no workaround helps.