Do I need a paid subscription to use AI with my documents?

No. The free version of ChatGPT lets you upload files. The free version of Claude does too. You hit limits (how many uploads per day, how long the file can be) sooner than paying users, but for learning, free is fine.

Is ChatGPT the same as Claude? Can I use them interchangeably?

They’re similar products from different companies. Both are chat assistants. Both can read files. The tone of their answers is different — Claude tends to be more careful, ChatGPT more confident. Neither is right for every task. Try both, keep the one you like.

Will the AI remember my document after I close the tab?

Usually no, and this surprises people. In a basic free chat, once the conversation ends, the file is gone. If you want the AI to remember a document across sessions, you either paste it back in every time, or use features like ChatGPT "Projects" or Claude "Projects" which do keep a file on hand across conversations.

Can I ask the AI to change the document and give it back?

You can ask it to generate a revised version as text, which you then copy out. Most free-tier chats don’t let you re-upload an edited file. If you need round-trip editing, tools like Microsoft Copilot inside Word are closer to what you’re picturing.

Why does the AI sometimes just make things up?

Because underneath, it’s a pattern-matching machine, not a reference librarian. If it doesn’t have strong evidence from the file in front of it, it fills gaps with what similar documents tend to say. The next lesson covers this in detail.

What Does "AI for Documents" Actually Mean?

The one-sentence version

“Using AI with your documents” means you hand an AI assistant a file — a PDF, a Word doc, a spreadsheet — and then have a conversation with it about what’s in the file. You ask questions, it answers. You ask for a summary, it writes one. You ask it to pull out every date mentioned in a contract, and it tries.

That’s it. If that already makes sense, feel free to skim the rest. If it doesn’t, read on — because there are some quiet gotchas in that simple description that trip up almost everyone the first few times.

What is an "AI assistant"?

The AI assistants we’re talking about here are chat programs you use in a web browser or phone app. The big names right now: ChatGPT (from OpenAI), Claude(from Anthropic), Gemini (from Google), and Copilot(from Microsoft). A few smaller ones too, but those four cover most real use.

They all work roughly the same way. You open a tab, you see a chat box, you type. The assistant types back. It behaves like a person who read an enormous amount of text before your conversation started and can talk about almost any topic — not perfectly, and not always accurately, but plausibly.

The feature we care about in this path is that most of them also let you upload a file — usually by dragging it into the chat window or clicking a paperclip icon. Once the file is attached, you can ask questions about it, and the assistant will answer based on the file’s contents.

A useful analogy: the eager new intern

Here’s the mental picture that helps most new users. Imagine you hired an intern on their first day. They’re very well-read — they’ve read more books than anyone you know — and they’re eager to help. They can summarize anything, draft emails, explain concepts, write a first pass at almost any document.

But: they’ve never worked at your specific company. They don’t know your clients, your team, your files, your history. So when you hand them a folder of documents, the intern reads them carefully and gets up to speed. After that, you can ask them questions about those documents the same way you’d ask a coworker.

Two quirks of this particular intern, though:

They have an iron-clad rule against saying "I don’t know." If they read your documents and can’t quite find the answer, they will guess. Confidently. Using exactly the same tone as when they do know. This is the single most important thing to remember.
Their memory is short. When you come back tomorrow for a second conversation, they’ve forgotten everything, unless you paid extra for them to file it away. Every new chat is Day One all over again.

What AI actually can do with your documents

Once you’ve uploaded a file, here are the jobs AI tools genuinely do well:

Summarize it. "Give me a two-paragraph summary of this 40-page report." This works well. Summaries are the single most reliable thing AI does with documents.
Answer specific questions. "What’s the cancellation window in this contract?" "Does this insurance policy cover water damage?" "When is the next school board meeting?" If the answer is in the file, it’ll usually find it.
Translate. "Translate this Spanish pamphlet into plain English." Works reliably for common languages. Less so for niche dialects or highly technical jargon.
Simplify. "Rewrite this legal page so a 12-year-old could understand it." The assistant is excellent at turning stiff writing into plain speech.
Extract. "Pull out every phone number in this document and list them." Very useful for receipts, contact pages, schedules, invoices.
Compare. Upload two documents and ask "what’s different between these two versions?" It’ll walk through the differences reasonably well.
Draft a reply or follow-up. "Write me a polite email declining this invoice based on the clause on page 3." It reads the source, then writes the reply.

What AI genuinely can’t do

This is the part beginners most often get wrong, because the assistant’s tone doesn’t change when it’s out of its depth. Keep this list close:

It can’t read your mind about what matters. If you upload a 200-page document and ask "what should I know?" you’ll get a generic summary. Better to ask specific questions: what does this contract say about renewal? What fees are mentioned?
It can’t always see the whole file at once. Long documents get broken into pieces behind the scenes, and the assistant gets fed a few pieces at a time. It might miss things that happen in a part it didn’t look at for your specific question. Our page on when things go wrong covers this in detail.
It can’t reliably do math. If your file is a spreadsheet of numbers and you ask "what’s the sum of column C?", the AI will often get it wrong. Sum columns in Excel. Use AI for the interpretation, not the arithmetic.
It can’t tell you about events after its training cutoff, and won’t always know which facts are stale. If you ask about recent news based on a news PDF, it should be fine because the answer is in the file. If you ask a general question that needs recent information, it might be confidently out of date.
It can’t see what it can’t read. A scanned PDF (a photo of a page) is different from a text PDF (letters the computer can select and copy). If a scan is hard to read, or handwritten, or rotated sideways, the assistant may silently skip it. More on this in the next lesson.
It can’t keep your secrets for you. By default, what you upload to a free AI chat may be used to train the next version of the model. This matters for sensitive files. Covered in privacy basics.

The most common first-time disappointments

Here’s what usually goes wrong the first week someone starts using AI with their documents. Recognizing these ahead of time saves a lot of frustration.

"It gave me a great answer, but the numbers are wrong"

Classic. The summary sounds right, the tone is confident, but the specific figure — the price, the deadline, the percentage — doesn’t match the file. This almost always means the assistant couldn’t actually read the number cleanly (a table got mangled, a scan was fuzzy) and it filled in what similar documents usually say. Always verify specific numbers by checking the file yourself.

"I asked for a summary and it left out the most important part"

Summaries reflect what the summarizer thinks is important. If the important-to-you detail is a single sentence buried on page 37, the summary might pass it by. Ask directly: "Does this document mention anything about X?" instead of hoping a general summary surfaces it.

"It keeps saying the document doesn’t exist"

Usually means the upload didn’t attach properly, or the conversation timed out, or the file type isn’t supported. Re-attach and check — there should be a little file icon in the chat showing the document is loaded.

"It answered a question I didn’t ask"

When prompts are vague ("tell me about this document"), the assistant picks a direction and runs with it. Be specific: "Summarize the section on cancellation" beats "tell me about this contract" by a mile.

Different tools, slightly different strengths

You don’t need to pick a favorite. Most people end up using two. Broad strokes:

ChatGPT is the most general-purpose. Best default for most tasks. Its free tier handles file uploads and images.
Claude tends to be more careful and handles long documents better — worth trying when you have something genuinely big (a whole book, a huge contract).
Gemini is tightly tied to Google Drive and Docs, which is convenient if that’s where your files already live.
Microsoft Copilot is best if your documents live in Word, Excel, OneDrive, or SharePoint — it’s built into those apps.

For a more detailed take, the tool-by-tool guides in our learning hub cover each one in depth once you’re past the basics.

Ready to try it?

The next lesson, Your first upload, walks through an actual first-time upload with a couple of low-stakes examples — a restaurant menu, a utility bill — so you can practice without worrying about getting something important wrong. If you’d rather read about privacy first, jump ahead to Privacy basics.