You are trying to save a document, change the size of some text, or find a setting. The instructions say “click the button,” but the button is nowhere on your screen.
This is a useful place to try AI. A clear screenshot can help you explain what you are seeing without knowing the name of every menu. The useful result is one understandable next step, followed by a check of what actually happened.
Your first exercise: practise with a harmless task, such as saving a copy of a sample flyer. Allow about 10 minutes. You need an AI chat that accepts images, or you can type the words you see instead.
Start with a task you can undo
Choose something small: opening a display menu, finding a print preview, or locating an export option. Say what you want to achieve before showing the screen. “I want to save a PDF copy of this flyer” gives the assistant a much clearer job than “What do I do?”
A still screenshot shows one moment. After you click something, the assistant needs a new screenshot or a description of the changed screen. It cannot reliably infer what appeared next.
Image support and limits vary by service and account. OpenAI’s image-input guide, for example, explains how to attach images and notes that descriptions can be wrong. Small text and unclear controls deserve particular care.
Share the useful part of the screen
A screenshot is simply a picture of what is on your display. Capture the app window or small area needed for your question. Keep the relevant heading, button labels, and error message visible.
- On a Mac: press Shift–Command–4, then drag around the area you want. See Apple’s screenshot instructions.
- On Windows: press Windows–Shift–S, choose a capture area, then save or copy the result. See Microsoft’s screenshot instructions.
- Without an image: type the exact visible labels and message. “The buttons say Preview, Export, and Share” is useful evidence.
Before attaching anything, open the actual image you will upload. Remove unrelated tabs, account names, addresses, messages, and document contents. Never include passwords, sign-in codes, recovery codes, or payment details. If a private detail sits beside the control you need, describe the labels in text instead. A loose scribble or translucent highlighter may leave information readable.
You can practise without sharing any private information using the fictional example below. For more on choosing what to share, see personalizing AI with less personal information.
Copy this one-step prompt
Help me with one step at a time.
My goal: [what I want to finish].
My device and app: [what I know; say “not sure” if necessary].
What I see: [attach a cropped screenshot or type the visible labels].
What I have already tried: [brief description].First, tell me which relevant words or controls you can actually see. If something is unclear, ask me to read it or share a clearer image. Do not invent a button or menu.
Then suggest one next step and explain what should appear. Wait for me to report the result. Before a step deletes or replaces anything, spends money, shares information, changes account access, installs software, or changes security settings, explain the consequence and stop for my decision.
Replace the bracketed parts before sending. You do not need technical language. “A Windows laptop, using the website in Chrome” is a reasonable starting description.
Try a text-only rehearsal
This is a fictional interface, not instructions for a real app. Imagine a flyer editor called EventDesk. Your flyer is already saved. You want a separate PDF copy to print.
Add this information to the prompt: “The toolbar shows Preview, Export, and Share. I have not clicked anything yet.”
A useful first answer might say: “Export is the most likely place to find file formats. Open that menu and tell me which options appear.” That is a hypothesis to check, not proof that the app works a particular way.
Now reply: “The menu shows PNG, JPG, and PDF.” A useful next answer would ask you to select PDF, then report whether you see a download, a preview, or another dialog. It should not jump ahead and invent a save window.
If the answer tells you to click “File → Save As,” ask it to explain where those controls appear in the evidence you supplied. This catches an assistant substituting familiar instructions for your actual screen.
If the next step does not match your screen
Stop at the mismatch. More clicking rarely fixes an instruction that describes a different app or version. Use this follow-up:
I do not see [the suggested control]. The screen now says [exact words]. Reassess using only these details. Ask one clarifying question before suggesting another step.
If the assistant repeats a missing button or keeps guessing, switch to the product’s official help. Search there for the exact task or error message. When asking a trusted person for help, include your goal, the device and app, the last action, and what appeared. Those four details also make a useful support message.
Know when the job is finished
For a PDF exercise, open the downloaded file and check the page, text, and filename. Confirm that the original remains available. A cheerful “Done!” in a chat is not evidence that a file saved correctly.
Be especially careful with unexpected warnings that urge you to call a number or hand over remote access. The FTC describes these as common tech-support scam tactics. Reach support through a source you independently know is genuine.
Your reusable habit: state the goal, show the current screen, take one step, and check the result. When you are comfortable, try another small task using the same prompt.
AI disclosure: This guide was created with AI assistance. Sources checked October 5, 2026. The EventDesk exercise and sample responses are fictional teaching examples, not a live product test. See how we review our guides.
Illustrative photograph by Surface on Unsplash.
