🤖 Part 8: Uploading Files & Long Documents

So far you've been typing everything into the box. But some of the best work an assistant does starts with a file — a PDF you don't have time to read, a photo of a receipt, a contract you need explained. This part shows you how to hand a document over, ask smart questions about it, and stay realistic about what the assistant can and can't see once you do.

📚 What You'll Learn

By the end of this part, you'll be able to:

Attach a file — PDF, Word, text, or image — using the paperclip or upload button, and ask questions about what's inside
Summarize a long PDF and pull specific facts out of it without reading every page yourself
Extract text from an image, like a photo of a receipt, a screenshot, or a handwritten note
Recognize the real limits — file size and page caps that vary by tool and tier, and why very long files can get quietly truncated

In This Part

Why Upload a File at All

Back in Part 1 we noted one of the assistant's real weaknesses: it doesn't know your facts — the specific document on your desk, the invoice in your inbox, the report your boss just sent. It only knows the general patterns it learned during training. Uploading a file closes that gap. Instead of hoping the assistant already knows something, you give it the exact material and ask your question against that.

The difference is night and day. Ask "what's a fair late fee on an invoice?" and you'll get a generic answer. Upload your contract and ask "what late fee does this agreement actually specify?" and you get an answer grounded in the real document — the clause, the number, the page. That shift, from general knowledge to your material, is what makes an assistant feel less like a search engine and more like a sharp assistant who just read your paperwork.

🧠 Mindset

Think of an upload as handing someone a stack of paper across a desk. They can only answer from what's on those pages plus what they generally know — they can't see the pages you didn't hand over, and they can't peek at the rest of the file cabinet. The clearer and more complete the stack you hand over, the better the help you get back.

Finding the Paperclip: What You Can Attach

In every one of the big assistants, uploading works the same way: look for a paperclip icon (📎) or a small plus (+) button near the message box, click it, and pick a file from your computer or phone. On mobile you'll usually also get options to snap a photo or choose one from your camera roll. Once the file appears attached above the box, type your question and send — the assistant reads the file and your question together.

What you can attach varies a little by tool and by whether you're on a free or paid plan, but the common types look like this:

File type	Examples	Good to know
PDF	Reports, contracts, manuals, statements, forms	The most common upload; works well when the PDF has real text (not just a scanned picture)
Word & text	.docx, .txt, notes, drafts, transcripts	Read cleanly; great for "summarize" or "rewrite" tasks on your own writing
Images	Photos, screenshots, receipts, handwritten notes, diagrams	The assistant "reads" the picture and can pull out text or describe what it sees
Spreadsheets (sometimes)	.csv, .xlsx tables of data	Support varies by tool and tier; Part 10 covers data work in depth

⚠️ Watch Out — a scanned PDF is really a picture

Not all PDFs are equal. If you can highlight and copy the words in a PDF, it has real text and the assistant will read it easily. If it's a scan — a photo of a page saved as a PDF — the words are just an image, and the assistant has to "read" them the way it reads a photo. That usually works, but it's less reliable, especially for small print, tables, or messy scans. If results look off, mention that the file is a scan so the assistant knows to look harder.

Asking Good Questions About a Document

Uploading is only half the job — the other half is asking well. The same prompt habits from Part 7 apply here, just pointed at the file. Vague questions get vague answers; specific questions get useful ones. Here's the flow from a file to a trustworthy answer:

graph LR A["📎 Attach your file
PDF, Word, image"] --> B["❓ Ask a specific question
⟨summarize / find / explain⟩"] B --> C["💬 Grounded answer
drawn from the document"] C --> D["🔎 Spot-check it
against the real pages"]

A few kinds of question earn their keep again and again:

Summarize: "Summarize this report in five bullet points a busy manager could read in a minute."
Find: "What does this lease say about pets? Quote the exact sentence and give the page."
Explain: "Explain section 4 of this contract in plain English — what am I agreeing to?"
Compare: "Here are two quotes. Which is cheaper overall, and what's different besides price?"
Extract into a list: "Pull every deadline and its date from this document into a table."

✅ Tip — ask it to point to the page

Whenever the answer matters, add "quote the exact wording and tell me where it appears." This does two things at once: it nudges the assistant to answer from the document rather than from memory, and it hands you a quick way to check — you can flip to that spot and confirm. We'll go deep on this "grounding" habit in Part 9; for now, just get in the routine of asking for the receipt.

⚠️ Important: The assistant answers from what you actually gave it — not from the file's title, not from what you meant to attach, and not from a newer version you have open elsewhere. If you upload last month's draft and ask about "the latest terms," it will confidently describe last month's terms. Before you trust an answer, make sure the file in the chat is the one you think it is.

Pulling Text Out of an Image

One of the quietly magical things a modern assistant can do is read a picture. Snap a photo or take a screenshot, upload it, and ask the assistant to type out what it says or answer a question about it. This turns a whole category of annoying chores into a ten-second task.

You have a photo of…	Ask something like…
A paper receipt	"List every item and price from this receipt, then total them so I can check the math."
A screenshot of an error message	"What is this error telling me, and what should I try first?"
A handwritten note or recipe card	"Type out this handwritten note as clean text; flag anything you can't read clearly."
A menu, sign, or label in another language	"Translate this menu into English and tell me which dishes are vegetarian."

The results are genuinely useful, but reading an image is harder than reading typed text, so this is exactly the place to keep the skeptic's hat on. Clear, well-lit, straight-on photos read best. Faint handwriting, glare, crumpled receipts, or tiny print can trip it up — and when it can't quite make out a word, it may guess rather than tell you it's unsure. That's why the prompts above ask it to flag what it couldn't read and why you sanity-check any number that matters.

⚠️ Watch Out — verify the numbers, always

A misread 7 as a 1, or a decimal point in the wrong place, can quietly change a total. For anything you'll act on — an expense report, a dosage, an account number — treat the extracted text as a fast first pass and confirm the important digits against the original yourself.

The Limits: Size, Pages & the Missing Middle

Uploads are powerful, but they aren't unlimited — and the exact limits are one of the most-changed, most-varied things in this whole field. How big a file you can attach, how many pages it will really read, and how many files you can add at once all depend on which assistant you're using and whether you're on the free or a paid tier. Free tiers are more limited; paid plans generally allow larger files and more of them.

⚠️ Important: Any specific number we printed here — "20 MB," "100 pages," "10 files" — would be stale within months, and would differ across ChatGPT, Claude, and Gemini anyway. So we won't quote them. When a limit matters to you, check the official help pages linked at the end of this part, or just try the upload: the assistant will usually tell you plainly if the file is too big or if it could only read part of it.

There's a subtler limit worth understanding, because it can bite you without any error message at all. An assistant can only hold so much text in view at once — its context window. Feed it a very long document and it may truncate — quietly read the beginning and maybe the end, while skimming or skipping the middle. The reply can still sound complete and confident, yet miss a clause buried on page 40. It's the same reason a very long chat can start to "forget" things you said early on — the oldest turns slide out of view as the conversation grows.

📖 Definition

Context window: the amount of text — your file plus the whole conversation — an assistant can hold "in mind" at one time. Context windows are large and have been growing fast, but they are not unlimited: exceed one and the oldest or middle material quietly falls out of view, which is why a very long document can miss the middle and a very long chat can start to forget earlier parts. (The exact size varies by tool and plan and changes often.)

📖 Definition

Truncation: when a file or conversation is too long to fit, so the tool uses only part of it — often the start and end — and leaves out the rest. Nothing warns you visually; the answer just silently rests on incomplete material. The fix is to work in smaller chunks so the whole thing actually gets read.

The practical response is simple: for a long document where every part matters, don't ask one sweeping question across the whole thing. Break it up. Ask about one chapter or section at a time, or upload the relevant portion on its own. You'll get answers that actually cover the material instead of answers that only look like they do. And a note on privacy — what happens to a file after you upload it is a real question, and we cover it properly in Part 13; until then, don't upload anything truly sensitive.

🛠️ How To: Upload a PDF and Ask Three Questions

What you'll do: attach a PDF you actually have — a report, a manual, a statement, a lease — and get a summary, a specific fact, and a plain-English explanation, all grounded in the document.

Step by step

Pick a safe, useful file. Choose a PDF that isn't sensitive (no account numbers or private data until you've read Part 13). A product manual, a public report, or a sample contract is perfect.
Click the paperclip (📎) or plus (+) next to the message box, select your PDF, and wait for it to finish attaching — you'll see it listed above the box.
Ask question one — summarize: "Summarize this document in five bullet points, then give it a one-sentence takeaway."
Ask question two — find a fact: "What does it say about [a specific topic]? Quote the exact wording and tell me the page or section."
Ask question three — explain: "Explain [a confusing part] in plain English, as if to someone who's never seen this kind of document."
Spot-check one answer. Open the PDF, flip to the page it cited, and confirm the quote is really there and really says that. This is the habit that makes uploads trustworthy.

💡 Tip

Keep all three questions in the same conversation. The assistant remembers the file across your follow-ups (Part 5's memory), so once it's uploaded you can keep asking — "now compare that to section 3," "put those dates in a table" — without re-attaching. If it ever seems to have lost track of the file, just upload it again and carry on.

Best Practices

✅ Do's

Confirm the right file is attached before you trust any answer — the assistant reads what you gave it, not what you meant to give it.
Ask it to quote and cite the page. This grounds the answer in the document and hands you an easy way to check.
Break long documents into parts. For anything where the middle matters, ask section by section so nothing gets skipped.
Sanity-check extracted numbers from receipts, statements, and photos against the original yourself.

❌ Don'ts

Don't assume it read the whole thing. Long files can be truncated silently — a confident answer isn't proof it saw page 40.
Don't trust a scanned or blurry image blindly — misread digits and dropped words are common; verify the parts that count.
Don't upload sensitive files — anything with private, medical, or financial details — until you've read Part 13 on privacy.
Don't rely on exact size or page limits you read somewhere — they change often and vary by tool and tier; check the official page or just try it.

💡 Pro Tips

If a PDF is a scan, say so — "this is a scanned document, read it carefully" — so the assistant knows to work harder on the image.
For a big report, first ask "give me the table of contents or main sections," then dig into the section you actually need. It's faster and dodges the truncation trap.

📓 Learning Journal

Keep a journal as you work through this guide — digital or paper. After each part, jot down:

Key ideas you learned
Things that clicked for you
Questions or confusion points to revisit
Ideas you want to try
Your progress and how you feel about it

✍️ This part's prompt: Find one document or image you've been avoiding — a long PDF, a dense manual, a pile of receipts, a form you don't understand. Upload it (as long as it isn't sensitive) and ask it your three questions: summarize, find, explain. Then check one answer against the original. How much time did it save you, and where did you catch it being vague or slightly off? Write down what you'd upload next.

📝 Part Summary

🎓 Key Takeaways

Uploading a file closes the assistant's biggest gap: it answers from your material instead of general memory — click the paperclip (📎) or plus (+) to attach.
You can attach PDFs, Word and text files, and images (and sometimes spreadsheets), then summarize, find facts, explain, or pull text out of a photo.
Ask it to quote the wording and cite the page — it grounds the answer and gives you a fast way to verify.
The real limits — file size, page count, and silent truncation of long files — vary by tool and tier and change often; break big documents into parts and check the official pages.

🎉 What You Can Now Do

You can now put a document into the conversation and get answers rooted in it — a summary of something you'd never find time to read, the one clause that actually applies to you, the text typed out of a photo in seconds. Just as importantly, you know where the edges are: which files read cleanly, when a long document gets skimmed, and why the numbers off a receipt always deserve a second look. That combination — reach for it, then check it — is exactly the habit that makes uploads pay off.

❓ Common Questions

Why did the assistant miss something that's clearly in my file?

Most often the file was too long and got truncated — it read the beginning and maybe the end but skimmed the middle, where your fact lived. Try asking specifically about that section, or upload just that portion on its own. A scanned or low-quality PDF can also cause it, since the text is really an image it has to decipher.

Can I upload several files at once and ask about all of them?

Often yes — many assistants let you attach multiple files and compare them — but how many, and how large, depends on the tool and your plan. If it struggles, add them one or two at a time, or combine them into a single document first. When it matters, check the tool's official help page for current limits rather than guessing.

Is it safe to upload private documents?

That deserves a real answer, not a shrug, so we give it a whole part. Until you've read Part 13 on privacy, treat anything with personal, medical, financial, or confidential details as off-limits, and practice on documents you wouldn't mind a stranger seeing.

🔭 Up Next

In Part 9: Grounding Answers in Your Material (and Its Limits), we'll turn the "quote the page" habit into a real skill — making the assistant answer only from the sources you provide, asking it to cite the exact passage it used, and understanding why even a grounded answer still needs your eyes on it.

📚 Additional Resources

🌟 Keep Going

You've just unlocked one of the assistant's most practical powers — turning a document or a photo into answers instead of a chore. The trick from here is trust with a seatbelt: lean on it, then check what matters. Next, let's make those answers even more trustworthy by grounding them firmly in your sources. 🤖