Perplexity Comet and the AI-Powered Future of Web Browsing (Premium)

Earlier this month, Perplexity announced its AI-powered web browser, Comet. At the time, it was limited to subscribers who paid the company $200 per month for Perplexity Max, but it said at the time that it would expand availability to other users, and it set up a waitlist for those interested in its take on the future of web browsers.

As I’m sure you know, I am quite interested in the future of web browsers and how AI will impact this more important of apps. And so I signed up for the waitlist. And waited. Until last night, when I was unexpectedly told I could download Comet and check it out.

A quick level-set first. The web browsers we use today–mostly Chrome, sometimes Safari if you’re an Apple user, and then a seemingly endless list of secondary (Microsoft Edge, maybe Mozilla Firefox) and tertiary (Brave, Opera, Vivaldi) options–all offer some form of AI integration today. But the implementations vary functionally and from a user experience perspective. And there are bolder pushes into a possible future of AI/browser integration from fringe players like The Browser Company and its new Dia browser. Which is limited to the Mac for now and still an early preview.

Comet falls into that latter bucket, of course. Like Dia, it’s still in early preview and limited, in this case to Windows. I’ve only just started using it, and I have next to no experience with Perplexity AI. But here’s what sticks out in my early testing. Very early.

?️ Yes, it’s Chromium

As soon as I got past the introductory screens and saw the browser itself, it was immediately clear that Comet is based on Chromium, the open source de facto standard browser used by Chrome, Edge, Brave, Opera, Vivaldi, and many other third-party browsers.

Some will bemoan this, but that’s  wrongheaded. I’ve made the case many times that a web browser engine is where we need standardization, not choices, and that browser makers should collaborate on that piece of the puzzle and innovate on top, adding their own unique differentiating features.

The success of Chromium being used that way should speak for itself. But Comet is yet another proof point: Rather than reinvent the wheel, Perplexity can simply do the thing it wants and needs to do and build an AI-powered product with whatever features and functionality. The existence of Chromium doesn’t just make this possible, it frees Perplexity from wasting time on basic browser functionality, not to mention compatibility and performance. It’s just there, and it just works.

Using Chromium isn’t just the right decision, it’s obvious.

? First run

The Comet onboarding process is quick and easy. After a short animation, you pick a username (really, a profile that’s specific to this install), an avatar (from a set of bowling ball-like circles), and a theme (light, dark, or system).

And then you’re provided with a useful New tab page that, as it turns out, is not the normal New tab page, listing many of the ways in which you can use Comet with your email, calendar, the web, web search, and more. This is an excellent starting point, but I wish there was an obvious way to get it back. (It’s at comet://newtab if you’re using that browser, as opposed to the default New tab page, which is a Perplexity URL.)

Once the browser is up and running, you can optionally create or sign in to a Perplexity account. I created an account, curious what limitations I might run into, given that I don’t pay for Perplexity Pro ($20 per month) or Perplexity Max.

? Basic browser features

Comet is, in many ways, a standard web browser with tabs, an address bar and toolbar, a single menu, and a multi-page settings interface. So that’s all familiar, and the basic browser features all look and work as expected. This is similar to how Dia looks and works. But The Browser Company goes further in hiding its Chromium roots with a settings interface of its own (on top of the traditional and more complete settings) and other changes. This makes sense: Dia is based on years of work with its predecessor, Arc, and Comet is brand new.

To my eyes, Comet most closely resembles Brave, with a minimalistic look that’s bland and color-less by default. In addition to the Back, Forward, and Refresh buttons to the left of the address bar, Comet has a Voice dictation (microphone) button, which I think subtly indicates one of many ways in which web browsing may change thanks to AI and its natural language interactions. The address bar is mostly stock, with View site information (hidden until mouse-over), Bookmark this tab, and Copy URL buttons.

The bigger UX changes, still subtle, are to the right. There, Perplexity places buttons for Summarize the current webpage, Launch voice mode, and Assistant, which I assume are core features and differentiators. After that, there are buttons for the menu and your profile(s).

I haven’t done much in Comet settings yet, but I did perform a few basic housekeeping tasks before diving in.

I used the EFF’s Cover Your Tracks website to see whether Comet provided any built-in protections against ads, trackers, and fingerprinting, and to my surprise it aced this test similarly to Brave. Nice.

I also installed my core extensions, like Proton Pass and Instapaper via the Chrome Web Store (of course). Doing so adds an Extensions button to the Comet toolbar, and you can pin extensions normally, so I did so for those I use most often.

And throwing caution to the wind, I also made Comet my default web browser. What the heck. And applied a light color theme because seriously.

? Core AI features

I noted the Enable voice dictation button to the left of the address bar and the Summarize the current webpage, Launch voice mode, and Assistant toolbar buttons to its right above. Their inclusion (and subtle prominence) in the UI suggested to me that maybe they would be a good place to start.

? Summarize the current webpage

This one is easy enough to test as I write reams of text every day and publish them to this website. So I navigated to thurrott.com, signed in through Proton Pass, and found a reasonably long Premium article. Then I clicked the Summarize the current webpage button. As expected, a sidebar opened, and Perplexity began filling in the text of its summary, which took perhaps several seconds.

The summary itself is quite good, with an introduction, a bulleted list of key points, and an accurate summary. Looking at the sidebar and those three AI buttons, I can see that Summarize the current webpage is a function (or mode) of the Perplexity AI assistant that’s built in to the browser. Basically, it’s grounding the AI in the current webpage only, as you would expect.

There are buttons for editing the query and copying it to the clipboard at the top of the summary, and then several buttons at the bottom for sharing a link to the summary, saving the query as a shortcut (which is interesting), marking the summary as helpful or unhelpful, copying it the clipboard (again), viewing the sources (not necessary here), and reporting a problem.

Also, when you engage with this feature, you are in fact just using the browser’s assistant, and so the Assistant toolbar button gets an “X” to indicate it’s on, and you can toggle off.

?️ Enable voice dictation and Launch voice mode

I was curious why there are two voice-related buttons in the Comet toolbar, Enable voice dictation over by the Back and Forward buttons and Launch voice mode over on the right of the address bar. But it’s fair to say that AI may one day render the “browsing” aspect of web browsing obsolete to some degree. And that voice interaction could be a key aspect of that. For example, it’s not difficult to imagine a future in which I’m writing as I do now, hit a point when I need to look up some date, name, or whatever specific fact, and instead of switching over to a browser window and using Google Search, I speak my question out loud, whatever AI perks up and listens and then answers that question audibly, and then I just keep writing, having obtained the information I needed.

To see if that future is happening today–“when will then be now?”–I clicked the Enable voice dictation button. Doing so selects the entire URL of the current webpage, but nothing else happened, and there were no visual cues. So I just spoke: “Summarize this webpage.” And that text appeared in the address bar, overwriting the URL (and explaining why it was selected in the first place). This works like typing in the address bar, in other words, so you get a drop-down with options like “Ask Perplexity,” “Ask Google,” and other search-type choices.

What it doesn’t do is kick off Perplexity: You have to tap Enter to select the default (“Ask Perplexity”) choice or the arrow keys and Enter (or the mouse) to choose one of the other options. It’s also not persistent, meaning that Voice dictation ends once you take action on your query.

Because I asked for a summary, which is a standalone feature I wrote about above, I was curious how or if it would differ from the sidebar experience I had just seen. Go figure, it’s the same and different, in that the summary appears in what I assume is the default Perplexity AI website experience. It also had trouble with a Premium article that had worked fine in the sidebar, which is interesting.

Clicking the Launch voice mode button opens the sidebar, so I guess this, too, is a Perplexity assistant feature. I have to guess because all it did was churn for a bit and then display an error message in that assistant. I’ll have to get back to this one.

?‍? Assistant

The Perplexity assistant looks and appears to work much like other AI chatbots. It appears in the sidebar and features familiar controls like “Ask anything…” and buttons for adding a screenshot and using your voice. What it doesn’t have is an upload button, though. That’s curious because the Comet New tab page features a similar interface, but with additional controls for choosing between Search, Research, and Labs modes; choosing a model; and attaching a file (free users get up to three uploads per day).

When you open the assistant while viewing a webpage, the title of that page and the website’s favicon appear above the “Ask anything” box.

And as you type a query, ostensibly about this page, that favicon appears next to the auto-complete suggestions it generates.

What you get here will vary by the query, of course, but I asked it a very general, almost obtuse question and got a summary that was, as before, useful and accurate.

☄️ But wait, there’s more

Adding a few buttons to a web browser does not a revolution make. But I didn’t really do anything dramatic yet, and I suspect that everything I tried to this point would work similarly in Edge with the Copilot sidebar, and with other browser-based or standalone AIs. But Comet is an AI web browser, and there is much more going on here than those basic features.

Using that Try Comet Assistant page at comet://newtab as a guide, I tried the following.

? Email

I navigated to Gmail, opened the Assistant, and typed “Summarize my recent email.” Using that page as its only source, Perplexity gave me a nicely organized summary of the unarchived (still in Inbox) emails from the past several days. It’s split into groups like Finance and Statements, Personal and Family, Work and Business, and so on, and the individual mail summaries go beyond what I see in the more limited Gmail views.

At the end of this list, the assistant added a note hinting at potential follow-ups: “If you’d like a summary focused on a particular topic (finance, travel, newsletters, etc.), or want more detail about specific senders or threads, let me know!” That seems useful, so I asked it which emails require me to follow-up in some way. Here, again, I got a nice–and useful–summary that was exactly what I was looking for. This is getting interesting.

? Calendar

Next up was Google Calendar. Here, I asked it to summarize my schedule for the coming week … but with a twist. Curious how it would respond, I asked this question from the New tab page. That is, I never navigated to Google Calendar. Here’s what happened.

Interesting.

Once again, I threw caution to the wind, connected Perplexity with my Google account, and gave it full permissions to all the email and calendar data it asked for. Then I asked it to summarize my schedule for the coming week again, surprised it didn’t just do it. The coming week is light, calendar-wise, so it didn’t have much to work with, and there was a weird HTML formatting error, but I guess it worked as expected otherwise.

? Browser functions

You can use the assistant to control the web browser, doing things like closing or grouping tabs. Typing those queries doesn’t make a lot of sense, since you can do most of that more easily with the mouse. But speaking those commands could work like the example I use above in the “Enable voice dictation and Launch voice mode” section. So I tried that.

First, I opened a new tab. Then, I typed Alt + V to enable voice dictation and said, “close this tab.” That didn’t work, as voice dictation is just a way to type with your voice. So I started over with voice mode, typing Alt + Shift + V and using the same voice prompt. Which also didn’t work because I can’t get voice mode to work, as noted previously.

Tiring of this, I tried to close the tab via a text prompt, but it told me it could not do that for “security reasons.” So I tried, “group this and the previous tab,” which did work.

As did “ungroup these tabs.” Not super-interesting. But I do think that app control will be a thing, and not just with the web browser. App Actions, the Settings agent, and other emerging features in Windows 11 point to a future in which apps expose individual features so that they can be controlled and automated with AI. What I see now in Comet isn’t particularly compelling, yet, but it’s also not completely useless. And it does fit into that future that Windows 11 is heading towards as well.

? Search

You can, of course, kick of an Internet search of any kind, normally using your voice or with a text prompt. But given my voice failures, I stuck with typing this time. This one is perhaps the most useful in that it’s a common need. But it’s also well understood and something we already do today, whether we come at it from Google Search or an AI chatbot. Moving on.

?️ Control a website

This is where things got really interesting.

Perplexity provides an example in which its browser assistant is used to generate walking directions in Google Maps. So I tried that: I opened Google Maps, and then I opened the assistant and typed “Starting from my current location, generate a scenic walk of Roma Norte.” (I’m in Mexico City currently.)

This worked, but not quite as expected, as it didn’t interact with Google Maps in any way. Instead, I got a reasonable list of local places, each with a nice summary. It was at least based solely on Google Maps, which is good. But the places were not listed in order by distance from my current location. So it’s not really a walk, just a list of places.

I started over with “Starting at my current location, providing a walking route to Parque Mexico in Condesa.” What happened next was … interesting. And it took quite a while to complete, longer than expected. Mostly because it was struggling with Google Maps, which I had never opened in this browser yet.

First the Google Maps display was surrounded by a blue aura, and a “Stop Comet Assistant” overlay with a Stop button appeared at its bottom. Huh.

Then, the assistant went into a reasoning mode in which it babbles to keep you distracted while it does its thing. Location detection was the big holdup, so I had to enable that in Google Maps. But I was fascinated to see that Perplexity was trying to–and then succeeded at–control Google Maps as if it were using the mouse and keyboard itself.

After it finally got the coordinates to my location–my precise location is difficult for Google Maps to see on a PC because it’s connected to Wi-Fi and not a cellular network–it filled them into the current field, but it didn’t overwrite the “Your location” text, it just added the  coordinates to that text. It saw this mistake, told me it would clear the field and start over, and then did so. And then it just worked. Except that it then noticed that it had reversed the starting point and destination. So it fixed that too.

And then it gave me three possible routes. The walk I requested is straightforward: You could turn right after leaving the apartment and just walk straight down Coahuila (the road) and hit that park without trying. But whatever. It’s correct.

And that is very interesting. This was the first time I’ve used an AI to control a web app (other than Search, I guess), and as with the other experiments above, I think it points to a future in which we don’t so much browse or open individual websites and perform tasks as we will direct AI to take action with the features they expose. Put more simply, this is the future of web browsing. It’s still early, yes. There are rough spots, missing features, and more work to be done. But you can see it now. You can see that it will happen.

? Baby steps

I will keep using Comet as my default browser and see what happens over time. Hopefully, I will stumble onto unique use cases that aren’t immediately obvious or handed to me by Perplexity’s hints and tips. And it’s clear that my usage of the browser and its built-in assistant will build some body of background information that each can work with. I just have to use it more.

And so I will. More soon.