To get AI to walk you through a computer task one step at a time, tell it explicitly to give you only one instruction at a time and wait for you to confirm before continuing — then show it your screen, either with a screenshot or live screen share, so it can see exactly what you’re looking at instead of guessing. This combination of a clear “one step at a time” instruction plus visual context is what turns a wall-of-text answer into something you can actually follow.
If you’ve ever asked ChatGPT how to do something on your computer and gotten back eleven steps in a single block of text, you know the problem. You read step three, look up, forget what step four said, and now you’re scrolling back and forth between your screen and the chat window like you’re reading sheet music you can’t play. It doesn’t have to be like that. AI can absolutely act like a patient person sitting next to you, pointing at the screen and saying “click that, now click this.” You just have to ask for it that way.
Why AI defaults to dumping all the steps at once
Left alone, most AI models will answer a “how do I…” question the way a manual would: front to back, all at once. That’s not a flaw exactly — it’s just the default behavior for a text-based question. Research on AI prompting consistently points out that ChatGPT thrives when tasks are broken down into steps, and if your request is too messy or complex, AI might give a confusing or incomplete answer. The fix isn’t a smarter AI. It’s a clearer instruction from you.
Tell it explicitly to slow down
The simplest fix is to say so directly. You can tell the AI something like: “Give me one step at a time. Don’t move to the next step until I tell you I’ve done it.” That framing works because you’re not just asking a question — you’re setting a rule for how the whole conversation should run. One prompting guide frames this well: “Act as an interviewer and ask me project manager interview questions one at a time. After each answer, give me feedback” — the same “one at a time, then check in” structure works just as well for a computer task as it does for interview practice.
If you want the AI to actually understand what it’s guiding you through before it starts firing off instructions, you can also ask it to repeat the goal back to you first. As one 2026 prompting guide puts it, telling the model to “repeat my goal, create a step plan and report live” is perfect when you want to ensure it fully understands your intent before starting — it restates your goal, builds a step-by-step plan, and reports progress live. That confirmation step catches misunderstandings before they turn into ten minutes of wrong clicks.
Show it your screen instead of describing it
Here’s where things get genuinely useful: you don’t have to describe your screen in words anymore. You can just show it.
Screenshots
The easiest option is a plain screenshot. Snap what’s on your screen, drop it into the chat, and ask your question. As one breakdown of ChatGPT’s vision features puts it, ChatGPT comes with the power of “vision” — it’s multimodal, meaning it can process images and doesn’t just extract text but can understand intent and respond with context. That means you can literally screenshot a confusing settings menu and ask, “What do I click next to turn this off?” A troubleshooting-focused guide describes exactly this use case: you can snap a photo and let ChatGPT troubleshoot your broken PC step by step, which beats digging through forum threads hoping someone had your exact error message.
Live screen sharing
For anything with multiple steps, live screen sharing is the better tool, because you don’t have to keep taking new screenshots every time something changes. Both ChatGPT and Google’s Gemini now support this.
In ChatGPT’s app, you start this from Advanced Voice Mode. OpenAI’s own help documentation explains that video and screen sharing are available to eligible subscribers in the ChatGPT iOS and Android apps — to share your screen, select the more-options menu, then select Share Screen and follow your device’s prompts. Once it’s running, you talk out loud and ChatGPT talks back, watching your screen the whole time. OpenAI has described this in action with a mundane but relatable example: if you need to tweak settings on your phone or computer, you can share your screen with ChatGPT to enable the AI to walk you through the menus and buttons to get things sorted, no more digging through endless tech forums.
Google’s Gemini Live works the same way and is built into the Gemini app for free. According to Google’s own rollout notes, when screen sharing is active, a clickable window called “Share screen with Live” lets Gemini view the user’s screen in real time and respond to questions based on what is shown, while users can also use their device’s camera or scroll through any app while talking. One independent guide sums up why this beats screenshotting: share your screen with Gemini Live and it will analyse what you’re looking at in real time — ask for help with an app you’re confused by or get step-by-step guidance navigating a website.
What a good “walk me through it” prompt actually looks like
Once you’ve got the screen sharing going, the words you use still matter. A few things to include:
- State the end goal first. “I’m trying to turn off notifications from this one app” gives the AI a target, instead of you narrating each click and hoping it catches up.
- Ask for one instruction at a time. Literally say “one step at a time” or “wait for me to say done before continuing.”
- Tell it what device and app you’re using. Windows and Mac menus look nothing alike, and neither do old and new versions of the same software.
- Ask it to check in. Something like “after each step, ask me if I see what you expect me to see” catches the moment things go sideways instead of five steps later.
This lines up with general prompting advice too — one library guide on working with AI notes that you should break large topics into smaller steps or questions and ask the AI step by step, starting a new chat for each unrelated topic so the AI works with a clean slate. The “clean slate” part matters more than people think. If your chat is cluttered with three unrelated topics from earlier in the day, the AI has more noise to wade through before it focuses on your actual screen.
When text is still the better option
Screen sharing and voice mode feel impressive, but they’re not always necessary. If the task is quick — “how do I change my desktop background” — typing it out and asking for numbered steps works fine. Save the live screen share for the messier stuff: settings buried three menus deep, an error message that doesn’t match anything you can find online, or software you’ve genuinely never opened before. That’s when having something watch over your shoulder and correct you in real time actually saves you time instead of adding a setup step.
The bottom line
Getting AI to walk you through a computer task one step at a time isn’t about finding a secret tool. It’s about asking for the pace you actually need and giving the AI eyes on what you’re doing, whether that’s a screenshot dropped into a chat or a live screen share in ChatGPT’s Advanced Voice Mode or Gemini Live. Say what you’re trying to do, ask for one step at a time, and show it your screen. That’s the whole trick.
Hi! I use AI to help research and write posts on this site. I do my best to keep things accurate, but please double-check anything important — and nothing here replaces advice from a licensed or certified professional.
