Drop a PDF into a tool, ask it a question, get an answer with a page reference. A year ago that felt like magic. Now many products do it, ChatGPT and Claude do it, and it's roughly free. If that's the whole task, you have plenty of good options.
For a lot of work, though, one PDF and a question isn't the shape of the problem.
One document is the easy case
Single-PDF chat works because the tool has a small, contained thing to read. Scale it up and the cracks show:
- You have eight PDFs, not one, and the answer depends on comparing them.
- Half of what matters isn't in a PDF at all. It's in three web pages and a note you wrote last week.
- You don't just want the answer. You want to see how the pieces fit, which usually means laying them out, not reading them one at a time in a chat window.
What actually gets harder
Working across sources. "What does this report say that the other one doesn't" is a different retrieval problem than "summarize this report." Tools built around one file tend not to do it well.
Keeping your own thinking in the loop. Your notes, your half-formed conclusions, the reason you kept a particular passage: that is part of the material too. A tool that only reads uploaded files treats your contribution as separate from the sources, when it should be on the same surface.
Not losing the thread. A chat log is a bad place to assemble something. It scrolls away, and the structure you're building only exists in your head.
How Graphia approaches it
Graphia puts every source on one canvas: PDFs, Word and slide files, web links, images, and the notes you write yourself. The AI reads all of it together, so a question can span the whole set, and every answer points back to the card it came from. When your sources don't cover something, it says so instead of guessing.
Chatting with a single PDF is table stakes now. The question worth asking a tool is what happens when you have twelve things open and half of them are yours.