How to summarise a PDF and ask it questions without uploading it
Why uploading a confidential document to an online chatbot is a bigger step than it looks, how document question answering works, and how to check the answers you get.
By Hasnain · XLUNA™ ·
Asking an AI to summarise a long report or find the one clause that matters in a contract is genuinely useful. Doing it by uploading the file to an online service is a decision about who else gets a copy. For many documents that is fine. For a contract, a medical letter, a payslip or a client file, it often is not.
What happens when you upload a document
When you drop a file into an online chatbot, it leaves your PC and is processed on someone else's servers. What happens to it next depends on that service's settings and terms: how long it is stored, who at the company can see it, and whether it may be used to improve their models. Many services let you opt out of some of this, but the copy has still left your hands.
Your workplace may also have rules about it. Plenty of contracts and data protection policies forbid sending client or personal information to outside services, and a chatbot counts. If in doubt, ask before you upload.
How document question answering works
Whether it runs online or on your PC, the method is broadly the same. The text is pulled out of the file and split into passages. When you ask a question, the passages most likely to contain the answer are found and handed to the language model, which writes a reply based on them. A good system tells you which pages those passages came from.
Two consequences follow. First, the answer can only be as good as the text that was extracted. Second, the model is working from a handful of passages, not the whole document at once, so a question whose answer is spread across many pages is harder than one with an answer in a single place.
When the file has no text
A PDF made from a scanner or a phone photo is often just a picture of each page. There is no text inside it to extract, so a document tool sees nothing. You can tell by trying to select a sentence in a PDF reader: if you cannot, the page is an image. Text recognition, which turns a picture of text back into text, fixes this, though it struggles with handwriting, faint copies and unusual layouts.
Asking good questions
- Be specific. "What is the notice period for ending this contract?" beats "Tell me about the contract".
- Start with a summary. It shows you how the document is organised, and what to ask next.
- Ask for the source. A question that names a section or a page is easier to answer well.
- Split big questions. Ask about each part separately rather than one question covering the whole file.
Checking the answers
Language models can state something confidently that the document does not say, especially numbers, dates and names. For anything that matters, treat the answer as a pointer and check it.
- Look at the page the answer cites and read the passage yourself.
- Be most careful with figures and tables. Columns can come out of a PDF in the wrong order, so a number may be attached to the wrong row.
- If an answer has no page reference, or the page does not support it, ask again more narrowly.
- For legal, medical or financial decisions, use the summary to find your way around, then read the original or ask a qualified person.
Doing it offline
An assistant that runs entirely on your PC does all of the steps above locally: the text extraction, the search and the writing of the answer. The document never leaves the machine, so the question of where it is stored and who can see it does not arise. The trade off is that it uses your own processor and memory, so a long document takes longer on a modest laptop than on a powerful desktop.
How XLUNA AI Assistant Pro handles this
In XLUNA AI Assistant Pro you drop in a PDF or a Word file, get a summary, and ask what you need to know. Answers point to the pages they came from, so you can check them in seconds. Text can also be pulled out of screenshots and scans, and charts explained in plain words. The AI model ships with the app and runs on your PC, and the app makes no internet connections of its own, so your documents, questions and answers stay on your machine.