Working with documents: PDFs, spreadsheets and images

Nine pages in thirty seconds

Joana is going to rent a two-bedroom flat (a "T2"). The landlord sent her the contract as a PDF: nine pages of small print. She adds the file to a conversation with the AI and writes:

Read this rental contract. Make a list of all the deadlines and amounts (rent, deposit, payment day, notice to leave, rent adjustments), with the page and the exact sentence where each one appears. Then tell me what questions I should ask the landlord before signing. In plain English.

In thirty seconds she has the list. She goes to page 6 to check the notice period and sees that the AI wrote 60 days. The contract says 120. A mistake like this, in a contract, is expensive.

The whole lesson is here: the AI saves you hours of reading documents, and you check what matters against the original.

What happens when you attach

When you add a file, the AI reads the text in it (or "looks at" the image) and works on it, as if you had pasted it. This has an advantage: working on what you give it, it makes up far less than when it answers from memory (Level 1, lesson 3). But it can misread a number, skip a page or merge two clauses into one.

And attaching is sending. The file goes to the computers of the company that runs the service, just like pasted text.

PDFs: summary, questions and deadlines

With a PDF in front of it, there are three requests that work almost every time:

• Summary: "Summarise in 10 lines what these rules say about who can apply."
• Questions: "What doubts should I take to the meeting with the landlord?" or "What does this document not explain?"
• Deadlines and amounts: "Make a list of all the dates, deadlines and amounts, with the page for each one."

It works for the letter from the Finanças (the tax office), the rules of a competition run by the câmara (town council), the rental contract or the washing machine manual.

Tables: calculations and a clean version

Paulo is the treasurer of the parents' association. The expenses for the Christmas party are scattered across emails and notes. He pastes everything in and asks:

Organise these expenses in a table with the columns: date, description, category (food, decorations, gifts, other) and amount. Then work out the total for each category. At the end, give me the formulas so that I can do the same totals in a spreadsheet. Expenses: [paste here, without parents' or pupils' names]

The clean table saves him an hour. But he checks the totals in the spreadsheet.

Photographs: documents and labels

Not everything is in a PDF. You can take a photograph and ask:

• The residents' association notice posted in the lift: "Summarise this in 3 lines."
• The label on a garden product, written in a language you do not read well: "What does it say about gloves, children and animals?"
• The label on a food package: "Does this contain nuts?"

For the photograph to be read well: good light, phone held straight, one page at a time and the whole document inside the picture. For a medicine or a dangerous product, the AI helps you understand, but you confirm with the leaflet or with the pharmacist.

The limits

• Very long documents. The AI may summarise the beginning and the end well and skip over the middle (this is the context window from Level 1, lesson 3). Ask by parts: "Now only chapter 4."
• Complex tables. Merged cells, several headers, scanned tables: the columns get swapped. First ask "Show me the table as you read it" and check before asking for calculations.
• Handwriting. It reads part of it and guesses the rest. Ask: "Mark with [?] everything you could not read with certainty."

How it is done today

This is how it looks on 6 October 2026; your screen may differ.

All the main tools have a button for attaching, usually next to the box where you type: a paperclip or a "+" sign. On a computer you can also drag the file onto the conversation. On a phone, the same button usually lets you choose between a file, a photograph from the gallery or the camera. Free versions may limit how many files you send or their size: see the Current state page.