Which tool to choose
Look in the assistant for the way to attach files: often a paperclip icon or an upload button, next to the message box. For occasional use, just attaching the file to the chat is enough. If you often work on the same documents, check whether the assistant offers a project space or a collection where you can keep them always available, without re-uploading them every time. Where uploading isn't available, you can paste the text directly into the message: less convenient, but it works for short documents.
How to do it
- Open the conversation and use the button to attach or upload a file.
- Select the document and wait for the AI to have taken it in.
- Ask precise questions about the content: "summarize the key points," "what does it say about...," "find the deadlines mentioned."
- If you need the documents often, put them in a dedicated space so they stay ready without re-uploading them every time.
A concrete example
Valentina needed to understand a twelve-page rental contract full of clauses. She uploaded it into the chat and asked: explain to me in simple words what I need to know about cancellation, costs and increases. The AI answered her by citing the points of her contract, not generic rules taken from who knows where. Then she asked "is there anything unusual or unfavorable?" and got a reading targeted right at her text. In ten minutes she had understood a document she would have set aside for days on her own.
When it does NOT work (and how to fix it)
If the AI seems not to read the file
Check that the upload went through and that the format is supported. If the document is an image or a barely legible scan, the AI may struggle to extract the text from it: try a sharper version or paste in by hand the parts you need.
If the document is very long
Huge files can exceed what the AI can hold together at once. Split it into parts and work one section at a time, or first ask for a general summary and then dig into the points that interest you.
If the answers mix your file with general knowledge
Tell it explicitly: "answer only based on the document I gave you, don't add anything else." That way the AI stays anchored to your text instead of filling in with its knowledge, which might not apply to your case.
A tip from someone who really uses it
Treat the uploaded files as the source, and the AI as the one who reads and explains them to you. The value isn't that the assistant "knows" new things: it's that it applies its summarizing ability to your specific material. That's why it pays to be precise in your questions — not "tell me about the document" but "find the termination conditions" — and to ask, on the important points, that it cite where it read them, so you can verify. A document uploaded and questioned well is worth hours of reading; the same document with vague questions gives vague answers.
Frequently asked questions
What types of files can I upload?
It depends on the assistant, but usually text documents, PDFs and often images. For spreadsheets and unusual formats the support varies: if a file isn't read well, converting it to a simpler format or pasting in its text often solves it.
Do my files stay private?
The same caution as always applies: you're giving a document to an external service. For sensitive files weigh what they contain before uploading them, and where possible anonymize the delicate parts.
Does the AI remember the files I've uploaded?
As a rule they apply to the conversation in which you upload them. To have them always available you need a dedicated space to keep them in; otherwise, in a new chat, they have to be re-uploaded.
Does the AI already know my documents, is it enough to name them to it?
No, and it's the misunderstanding that leads to pointless questions. The assistant has no access to your files, your emails or your notes: it doesn't know what's in the "contract I was telling you about" until you actually give it to it. Naming a document doesn't make it read it; uploading or pasting it does. Until you pass it the text, the AI can only imagine what it contains, and imagining about a contract or a medical report is exactly what you don't want. The rule is simple: no file given, no answer based on that file.