Skip to content

Better questions

How to ask AI about what is on your screen

Give an AI screen companion enough context without writing a giant prompt. Start with a visible source, a clear goal, and one useful question.

··3 min read

A good screen question connects three things: what you can see, what you are trying to do, and the part that does not make sense yet. You do not need to describe every pixel or direct every annotation.

One source. One question.Illustrative diagram
From the screen to a specific questionThe highlighted line is the evidence; the small question supplies the goal. This is an illustrative document, not an app recording.Your documentWhat changes here?Context+ goal+ one questionOne next step
  1. 01
    Show the source

    Keep the relevant document, chart, or interface visible.

  2. 02
    Point to the uncertainty

    Identify the exact relationship or control that needs explaining.

  3. 03
    Name the outcome

    Say what you want to understand or achieve next.

The highlighted line is the evidence; the small question supplies the goal. This is an illustrative document, not an app recording.

Make the source easy to inspect

Before speaking, bring the relevant window forward and scroll to the part you mean. A screenshot can provide layout, labels, and visible values, but it cannot reveal a collapsed section or a document that is not open. If the question concerns a comparison, put both values in view. If it concerns a sequence, start at the first meaningful state instead of the final result.

This is useful even when your question sounds obvious to you. “Why is this wrong?” leaves several possible targets: a warning, a calculation, a heading, or a selected object. Naming the visible element lets the companion anchor its answer to evidence you can check. A short phrase such as “the total in the bottom row” often does more than another paragraph of background.

Ask for an outcome, not a production script

You should not need to specify a circle, three arrows, a color palette, and an animation timeline. Those are presentation choices. A better request names the result you need: understand why a value changes, locate a setting, or decide which visible option fits a constraint. Rayito is designed to connect an explanation with pointing and annotations when those make the relationship clearer.

Keep real constraints in the request. “Explain this chart for someone who does not know the terminology” and “help me find the export option without changing the document” describe different tasks. Neither needs a long prompt. You can add detail after the first response, once you know which assumption needs correcting.

A question to try

“What is causing the change in this chart, and which part should I look at first?”

Treat the first answer as a starting point

Check that the companion is referring to the right source before following its reasoning. If it points at the wrong row, correct the target immediately. If the screenshot lacks an important value, reveal it and continue. A screen companion is most useful as a back-and-forth conversation, with fresh context when the task changes, rather than a single answer you must accept as complete.

When an explanation reaches a decision, answer that question before adding a separate problem. This keeps the visual thread easy to follow. You can ask for a smaller step, a different example, or a return to the previous idea. For tasks that change another app, keep the intended outcome explicit and review any action that asks for confirmation.

Know where the screen stops helping

A visible page is evidence about that page, not permission to act on everything it mentions. It also does not establish whether a number is current, whether a source is reliable, or what is hidden elsewhere in an account. Ask the companion to distinguish what it can see from what it is inferring. That small habit makes explanations easier to verify and prevents a polished diagram from standing in for missing information.