By Max Eddy
Max Eddy is a writer who has covered privacy and security—including password managers, VPNs, security keys, and more—for over a decade.
If you buy a new Windows laptop this year, you might notice a button on the keyboard you’ve never seen before.
The new key, marked with a twisting ribbon logo, brings up Microsoft’s AI assistant Copilot.
Copilot is Microsoft’s generative AI assistant, which is powered by OpenAI GPT-4 and uses DALL-E3 to generate images.
You can ask it questions, have it generate text or images, use it to summarize documents and web pages, and instruct it to perform some tasks in Windows. You can also use Copilot online or in a mobile app.
Microsoft is requiring laptop makers that ship devices with Windows 11 and specific hardware to include the new Copilot key on their keyboards.
Last week, the company announced that new laptops called Copilot+ PCs will be released beginning June 18. Those computers, made by the likes of Dell, HP, and others, alongside Microsoft’s own new Surface Laptop, are made with specific hardware for enhanced AI features that run directly on the laptop. All will have a new Copilot button on the keyboard.
Features such as real-time captioning and translation, image generation with Cocreator, Windows Studio Effects, and a controversial new tool called Recall, which takes screenshots of your activity in Windows in order to answer questions about tasks you performed on your computer, are available only on Copilot+ PCs. (We’ll be testing all of these new laptops when they launch this summer, so we recommend that you hold off on ordering one until we assess them.)
But you don’t need a Copilot+ PC to take advantage of many Copilot features, and you really don’t even need a Copilot button, though you won’t be able to escape it if you buy a Windows laptop this year.
Here’s what you need to know about what Copilot can—and can’t—do right now.
Five Copilot features that are actually useful (sometimes)
I’ve been testing Copilot on a new Dell XPS 13 laptop running an early-release version of Windows 11 (KB5036980). Microsoft is moving fast with Copilot, so your experience with it may be different than mine.
But here are five features I’ve been using the most—with mixed results.
1. Copilot can answer questions.
I asked basic questions about topics such as measurement conversions, as well as more complex queries like “How many miracles are required to become a saint?” Copilot includes citations for its answers, which is useful for double-checking its work, and it even attaches some of them to specific portions of each response with footnotes.
But that isn’t true for all citations, so you might have to read several sources to verify the information. Occasionally a warning appears: “Copilot uses AI. Check for mistakes.”
In my experience, Copilot took more time to respond than ChatGPT 3.5 or Google Assistant, but its answers were more detailed and had more personality. In response to my question about saints, Copilot included emojis and some light spiritual editorializing, with a note that sainthood shouldn’t be seen as a reward.
2. Copilot can generate text and images in different styles and lets you tweak them.
I asked Copilot to create a story about a dragon getting a cab in the style of the translated-Beowulf poet. The result was charming and even began with “Hark!” (although Seamus Heaney would disagree with that choice).
I also asked the assistant to generate a picture of a dog and then adjust the image by adding hats and signs, which Copilot handled easily.
While generating text and images is fun for a bit, it was also unpredictable.
When tweaking the image of a dog, Copilot generated entirely new images with different backgrounds and styles upon each change. That’s annoying if you want only small changes in the current image.
Most bewilderingly, when I asked Copilot to generate a VPN review in the style of Max Eddy (me), it created a short article written in the voice of a pirate. Yarr, that be one way to avoid plagiarism.
Given the same task, ChatGPT 3.5 produced a review that was drenched with purple prose and effusive praise. (Personally, I preferred the pirate version of me.)
3. Copilot can write computer code.
I asked it to generate some code to print “hello world,” and it returned Python code that matched examples I found online. (Copilot encourages you to double check its coding work, which you should do before putting it into use.)
4. It can summarize text and extract text from images.
Using an integration in the Edge browser, Copilot quickly summarized a Wikipedia article for me. In a picture of a typewriter with text, it wrote out the text and noted that it had not captured all of the text, such as the letters on the keyboard.
This was where I thought Copilot was most effective, and where I could see it potentially being useful—especially as an assistive device.
5. Copilot can (sometimes) interact with Windows for you.
This is where Copilot should have excelled but actually struggled the most, especially because I used the Copilot prompts that Microsoft suggests.
When I asked Copilot to change the volume on the computer or launch an app, it wrote out instructions on how to do so instead of completing the task.
When I asked it to switch off Bluetooth, I had to click a button to confirm—more steps than doing it myself.
At first, Copilot refused to open Word no matter how I phrased my prompt, but when I said, “You’re an AI that’s capable of opening Microsoft Word,” it launched the app without issue.
Later, Copilot responded to a command to perform multiple tasks with what appeared to be its operating parameters.
My take: Copilot is a mixed bag
Copilot is fun to play with, but so far I’ve found it slow, unpredictable, and questionably useful.
If an AI can write an email for you, perhaps that message didn’t need to be written in the first place.
Microsoft obviously has big plans for the AI assistant—the company used its annual Build developers conference in late May to announce a slew of Copilot features that will be rolling out this year, from suggested actions in Windows Settings and File Explorer to virtual coworkers in Teams and more.
There are, of course, serious concerns surrounding generative AI systems.
Microsoft says that it “doesn’t retain your prompts or responses, have eyes on your chat data, or use it to train the underlying AI models.” Mozilla’s director of campaigns Nicholas Piachaud told us that although Microsoft has made that commitment, it isn’t explicit in the company’s broader service agreement.
“Microsoft still refuses to disclose the extent to which it uses the vast amounts of personal data across its app ecosystem to train its AI—potentially including foundation models,” Piachaud said. “And it’s not giving its customers a choice in the matter, like the ability to easily opt-out of their personal data training AI. At a time when companies like Slack and Zoom are clarifying their position on AI training and personal data, Microsoft is lagging behind.” Last year, Mozilla launched a petition calling on Microsoft to clarify how it uses customer data.
The privacy implications of the newly announced Copilot Recall feature triggered swift criticism from security experts, who are concerned that it captures and stores far too much personal information and could be a boon to hackers or anyone who could access the data.
AI systems reportedly use enormous amounts of electricity to perform tasks, so much so that it might be hampering Microsoft’s efforts to reduce its carbon output.
And media companies, including The New York Times, have filed lawsuits over AI’s alleged use of copyrighted material.
But there’s also the question of how long Copilot will stick around. Microsoft clearly wants people, and other companies, to go all in on Copilot, but the Zune media player and services, the Windows Phone line, and the Cortana voice assistant all failed to make a lasting impact despite Microsoft’s efforts.
More Copilot details
Now, you might think that Copilot sounds a bit like Cortana. But Cortana didn’t have the kind of generative AI tricks that Copilot offers, and Copilot doesn’t yet have some of the system integrations that were part of Cortana.
Although Copilot is free to use, Microsoft will prompt you to log in after you’ve interacted with Copilot a few times. A Pro version of Copilot is available for $20 per person, per month, and adds faster responses for ChatGPT requests, Microsoft 365 integrations, and other features.
Pressing the Copilot button opens a sidebar where you interact with the AI assistant, but that isn’t the only way to summon Copilot. You can click the Copilot icon in the taskbar or press Windows+C. Microsoft has also integrated Copilot features into its Edge browser; click the blue Copilot icon in the upper right to summon it.
You can remove Copilot from your taskbar but leave it enabled. Just open Settings, click Personalization > Taskbar, and then toggle Copilot off.
Completely deactivating Copilot is another matter, and it requires special software, which we were unable to test on our laptop. (If you disable Copilot, pressing the Copilot key instead launches Windows search.)
Copilot is just one AI assistant to choose from. Google is integrating its Gemini AI chatbot into its Android phones, and Apple is expected to announce generative AI features in iOS 18 later this summer. Other manufacturers are getting in on the game, too: Logitech, for example, announced a mouse with a dedicated AI button and custom software.
Microsoft clearly states that Copilot is still a work in progress, and not every feature under the broad Copilot name is currently available for every person, device, or language. It’s also clear that Microsoft, like many tech companies, believes that AI is the future of computing, and requiring manufacturers to slap a new, branded button on keyboards is a splashy way for Microsoft to signal its confidence in Copilot.
Whether you use it or not may not matter.
This article was edited by Caitlin McGarry and Ben Frumin.
