Copilot Vision in Edge is unlike anything I’ve seen in a web browser and takes the concept of generative AI in a new direction. Instead of creating text or images based on prompts, the browsing companion analyzes everything visual and textual on a web page and verbally converses with you about it, providing context and explanations. Yes, the Google Lens feature in Chrome bears a slight resemblance, letting you highlight objects on a page and get search results in a side panel, but it’s not at all conversational.
I tested the preview version of Copilot Vision in early 2025, but the release version is now available to all via the desktop versions of the Edge web browser. Its main features are as impressive as ever, though some changes to how it works on web sites with private data give me pause. I’m here to walk you through how to get the most out of Copilot Vision and understand its downsides.
How to Get Copilot Vision in Edge
Originally, Copilot Vision was available only to select Copilot Pro subscribers ($20 per month), but now it’s free for anyone with the desktop version of the Edge browser. Copilot Vision is entirely opt-in.
To get started, first make sure your browser is up to date by choosing the three-dot menu at the top right, then Help & Feedback, and then About Microsoft Edge. The top section on this page checks if you’re running the latest version. You must also sign in to a Microsoft account to use Copilot Vision, so do that next. Finally, open Edge, choose the Copilot icon at the top right to open the AI’s side panel, and click the microphone button at the bottom.
(Credit: Microsoft/PCMag)
You then see a dialog box at the bottom center of the browser window where you can give the tool permission to enable the feature. The box explains it as “Get help from Copilot by sharing what’s on your screen and talking about it. Copilot can hear your voice and respond like a friend in conversation.” Hit the Accept button.
(Credit: Microsoft/PCMag)
Next, you see Copilot’s toolbar across the bottom center of the windows, with an eyeglasses icon and a mic icon. Those are for stopping the AI from viewing your screen and muting your mic.
(Credit: Microsoft/PCMag)
Since the feature is speech-based (even though it’s called Vision), you need to enable your mic for it. After you tap the microphone icon, you hear a piano lick, and then the lifelike AI speech of Copilot Voice welcomes you. You can choose among four voice personalities: Canyon, Grove, Meadow, or Wave. I picked Wave (which has a British accent), though I use Canyon on my phone.
Copilot Vision can describe and give you info on what you’re seeing on a site, but also just chat about anything. Here’s a small taste of what it looks and sounds like:
As you can see, the interface for Copilot Vision is completely different from the Edge browser’s Copilot sidebar, which is a standard ChatGPT-like chatbot interface. The Vision toolbar even disappears when you aren’t using it.
Getting Started With Copilot Vision
When I first clicked the Copilot Vision screen-share icon, the borders of the browser window changed color, in my case from blue to tan. The mic and eyeglasses icons turn red when they’re active. A friendly voice said, “Hey Michael, how are you doing today? What’s on your mind? Or should I surprise you with something fun?”
(Credit: Microsoft/PCMag)
The main Copilot Vision for Edge page provides sample things to ask. For example, it showed a web page with four cute dogs and suggested I ask it to “tell me more about these breeds.” The next suggestion was “Summarize these articles,” which showed that the tool works with both images and text on web pages. Then, it demonstrated that it knows about geography by showing four cityscapes and prompting, “Which of these cities has the oldest buildings?” Finally, it said, “Now it’s your turn,” and suggested some sites like Amazon, Target, Tripadvisor, and Wikipedia to get started.
One of the sites Copilot proposed was GeoGuessr, which has its own World Cup. I told it that I wasn’t interested in the soccer World Cup, and it assured me that this wasn’t related.
When I stopped speaking with it for a while, I got a message saying, “Sorry, nodded off for a second! Try reconnecting.” That’s actually a good thing, since you don’t want it to keep listening if you accidentally leave it on. When I asked if I could provide feedback to Copilot Vision directly with my voice, I was impressed by its reply: “Your feedback will be passed on to my developers.” If you interrupt Copilot Vision, it politely stops speaking.
What’s It Like to Use Copilot Vision?
Interestingly, Copilot Vision told me that it sees only the part of the browser that I can see, and not anything beyond the visible window. But when I went to a page on Everyday Health, it summarized content far below what I could see. Since the standard Copilot sidebar in Edge can summarize entire web pages, it’s possible that Vision was tapping into it for this info.
Get Our Best Stories!
Your Daily Dose of Our Top Tech News
By clicking Sign Me Up, you confirm you are 16+ and agree to our Terms of Use and Privacy Policy.
Thanks for signing up!
Your subscription has been confirmed. Keep an eye on your inbox!
(Credit: Microsoft/PCMag)
When I navigated to my photos on OneDrive, Copilot Vision described a bird photo there. It treated OneDrive content as private in the preview version, so that could now be a potential privacy concern. It also knew that a highly distorted photo on Flickr was of a spider, which I couldn’t recognize at first. It can now see content on Instagram and other social networks, too.
The mute button makes the AI stop listening to you, but I wish there were a button to silence it when it goes on a bit too long about background information on a page or in response to a query. Instead, you can say, “Quiet!” This will stop it from speaking. That said, it remains active when you tell it to stop watching. You can just hit the X in its control panel at the bottom of the screen to close it in that case.
I tripped up Copilot occasionally, especially when I asked it to do things it couldn’t, like turn itself off. In those cases, it usually went back to describing what was on the page and adding more context and background info.
Co-Gaming
Don’t let this section get your hopes up too much: Copilot Vision can be active while you play games on the web, but not as a competitor or partner. Instead, it provides strategy tips or commentary about what’s on the screen. When I started playing Mr. Mine on CrazyGames.com, Copilot Vision knew how to play the game and what the goal was. When I asked how it knew about this little-known game, Copilot said it could read what was on my screen and, “Yeah, I’ve got a knack for games.”
Recommended by Our Editors
(Credit: Microsoft/Crazy Games/PCMag)
What Can’t Copilot Vision See or Do?
I asked Copilot Vision whether it would be watching if I went on a pornography website and got a thoughtful answer saying, “for safety and privacy, I don’t store or share personal info.” Microsoft’s documentation states that Copilot Vision doesn’t use input for AI training. In other words, it doesn’t see sensitive or legally protected information, including bank account credentials or passwords.
When I tested the prerelease version of Copilot Vision in Edge, it stopped functioning as soon as I navigated to a bank website or signed into a OneDrive page. With this released version, it kept on working. When I asked if it shouldn’t stop viewing these pages, it gave me the same “I don’t store or share” response. I prefer the more cautious approach of the preview version.
Copilot Vision can’t open a new web page, which makes it less helpful than it could be. The tool can at least now detect your cursor position, which it couldn’t do in my tests of the preview version.
Copilot Vision technically can’t see video, but it can provide feedback based on still frames from a video. It can’t hear and interpret web page audio either.
Finally, Copilot Vision can’t provide a written transcript of your interactions with it; it would be nice to be able to see its answers. The regular Copilot, whether in the Edge sidebar or a separate app, does this.
Is Copilot Vision Useful?
Copilot Vision is good at providing detailed spoken descriptions of what’s on a web page, alongside rich background and context. It speaks sort of like a friend who has no opinions of their own—something you might appreciate! I also like how Microsoft now integrates Copilot Vision into the existing sidebar-based Copilot. And the new eyeglasses icon makes it clear that the AI is watching your screen.
But there are some gaps that limit its usefulness. For example, I wish it could open web pages for you and turn itself off on your voice command. And I would have preferred it if Microsoft kept the preview version’s behavior of not operating at all when you log in to a website, or at least made this an option. I hope the forthcoming version of Copilot Vision for Windows doesn’t have these same drawbacks.
For more on Copilot, check out our comparison between Copilot and Copilot+ and cool things you can do with a Copilot+ PC.
About Michael Muchmore
Lead Software Analyst
