This is how to show AI instead of telling it

This is how to show AI instead of telling it.

People spend ten minutes typing a description of an appliance display, an assembly diagram, or a public product page when the useful thing is already in front of them. Describing is lossy: you leave out details you did not notice, and one of those details may change the answer.

The skill is a two-part reflex: ask whether showing the thing would help, then ask whether a safe, narrow version of that thing exists. If the useful input is itself private or sensitive, this is not the technique to use.

What "showing" covers

• Ordinary objects: the appliance display, plant leaf, furniture part, paint wall, product label, or assembly step that makes no sense. Frame the object, not the room around it.
• Public documents: a manufacturer manual, public form, product specification, museum guide, or instructions anyone can open. Use the exact page the question needs.
• Narrow screen captures: an error message or one harmless settings panel, captured after closing unrelated tabs and notifications. A fresh tight capture is safer than a full desktop screenshot.
• Fictional or permitted data: a small made-up table for learning a technique, or a non-confidential file you are explicitly allowed to share. Work data is not ordinary merely because it is in a spreadsheet.

Before you attach anything

Do not start with the whole file and plan to redact it later. Start with the safe subset. Move private objects before taking a photo. Physically cover identifiers before photographing a few needed rows. Close unrelated tabs and notifications before taking a screenshot. Inspect the newly created image once. If the task still needs a medical record, financial statement, identity document, private message, child’s information, confidential work, or a contract whose sensitive contents are central, stop instead of trying to sanitize your way through it.

Watch it on a public assembly diagram

The ask:

I attached page 8 from the manufacturer’s public bookcase manual and a tight photo of the two ordinary hardware pieces on my table. Match each piece to the diagram only when the shape and label agree. Quote the diagram label you used, tell me what you cannot see clearly, and point me to the next numbered step. Do not invent a part number or tell me to force anything.

The version where you type "I have a little silver bracket and a round plastic thing" leaves the important shape and labels behind. The public diagram plus a tight photo gives the AI something concrete to compare. You still confirm the label on the original manual page and stop if the fit requires force or the manufacturer’s instructions disagree.

How to ask about what it’s looking at

1. Say what you need, not just "explain this": "match these two pieces to the labels on page 8" beats "help with my bookcase."
2. Tell it what you already know, so it skips the obvious.
3. Ask "what should I look at more closely that you can’t read well?" It knows where the photo was blurry or the scan was cut off, but only tells you if asked.

The limits, honestly

• It misreads. Numbers especially: confirm any figure it quotes against the document with your own eyes before acting on it.
• Files carry more than the subject. A screenshot can include tabs and notifications; a photo can include mail, faces, keys, or a room layout. Create and inspect a narrow capture instead of uploading the original.
• Some inputs should not enter the chat. Medical records, identity documents, full financial records, private messages, confidential work, and sensitive contracts do not become safe because they are images. When the sensitive content is the useful content, stop.

Once this reflex installs, more of the physical world becomes askable. The second reflex matters just as much: show only what the task needs.

• Share only: An ordinary object, public document, or tight capture whose useful content is visible without identity, account, medical, child, private-message, or confidential-work information.
• Verify: Every quoted word, number, clause, label, and image detail against the original file, then use an accountable professional for medical, legal, financial, or safety decisions.

This is how to use AI.