Which tool to choose
Look, next to the message box, for the way to attach an image: often the same icon you use to upload files, or an option to take or choose a photo. Check that your assistant can interpret images, because not all do. Where it exists, it's the right tool for anything you can photograph or capture with a screenshot. Where it's missing, the only alternative is to describe in words what you see, which is far more limited.
How to do it
- Attach the image to the conversation with the button to upload or photograph.
- Write a precise question: "what does this sign say?", "explain this chart to me", "what error does this screenshot show?".
- Read the answer and, if you need more, ask for a deeper look at the same element.
- For important images (documents, data), verify the critical parts, because on hard-to-read text the AI can make mistakes.
A concrete example
Marco had a screenshot of an error message from his computer, one of those incomprehensible ones. He uploaded it to the chat and asked: what does it mean and how do I fix it? The AI read the error text in the image, explained in plain words what was happening, and proposed the steps to try. Marco didn't have to retype the error code by hand or search for it by trial and error: he showed the image and received an explanation aimed at exactly what he had in front of him.
When it DOESN'T work (and how to fix it)
If the image is blurry or the text is small
The AI may read it wrong or not at all. Try again with a sharper photo, well-lit and close up, or retype by hand the part of text you need. The clearer the image, the more reliable the reading.
If the AI misinterprets what's in the image
It can happen that it describes one thing for another, especially in confused photos or ones with many details. Guide it: tell it what to focus on ("look only at the writing at the top") and correct it if it's wrong. A targeted question gives more precise readings than a generic "what do you see?".
If it's a matter of data or important documents
Use the AI to understand and get your bearings, but verify the numbers and the crucial parts against the original. On a bill, a medical report, or a table, a misread character changes everything: the convenience of automatic reading doesn't replace a check on the details that count.
A tip from someone who really uses it
Think of the screenshot as the fastest way to "show" the AI a problem instead of describing it. Explaining an error, a complicated interface, or a chart in words is slow and imprecise; showing it is immediate. When something on your screen has you stuck — a warning you don't understand, a menu where you can't find the button, a dense document — capture the image and show it. A clear photo is worth a thousand lines of description, and it spares the AI from guessing what you meant: it sees exactly what you see.
Frequently asked questions
What kind of images can I have analyzed?
Usually photos, screenshots, document scans, charts, signs, receipts. The AI can describe them, read their text, and explain them. The quality of the answer depends a lot on the sharpness of the image.
Does the AI read the text inside photos well?
On clear, legible text, yes, and it's one of the most useful uses. On small, handwritten, or blurry writing it struggles and can make mistakes: there it's worth verifying the important parts.
Do my images stay private?
The usual caution applies: you're uploading the image to an external service. For photos of sensitive or personal documents, consider what they contain before sending them and cover or omit the data that isn't needed for the analysis.
If I upload a photo, does the AI always understand what's in it?
No, and it's good to know so you don't trust it blindly. The AI interprets the image, but it can misunderstand it: mistaking one detail for another, misreading a figure, confidently describing something that isn't there. The more confused the image, the higher the risk. On inconsequential things it's fine to trust it; on a piece of data, a document, or an important decision, treat the reading as a first aid to be confirmed, not as a verdict. Taking every interpretation of a photo as certain is the most common way to build on a detail read crookedly.