In November 2025, Amazon sued Perplexity over a browser. Not a chatbot, not an API integration, an actual web browser called Comet that Perplexity built to let its AI agent shop on your behalf, including inside your logged-in Amazon account. Nine months and one Ninth Circuit ruling later, that lawsuit is still reshaping how agentic commerce gets built, and Comet is the browser everyone in ecommerce is watching to understand where this is headed.
If you sell physical products online, Comet matters for a different reason than ChatGPT Shopping or Google's AI Mode. It doesn't live inside a search results page or a chat window you have to visit. It sits on top of every website you already have, silently reading your product pages, your pricing, your reviews, and your checkout flow, then acting on what it finds. That's a genuinely different surface, and most brands haven't thought about it at all.
What Comet actually is
Comet is a Chromium-based web browser, not a plugin or a chatbot wrapper. Perplexity launched it on July 9, 2025, initially gated to subscribers paying $200 a month for Perplexity Max. That pricing didn't last. By October 2, 2025, Perplexity made Comet free worldwide on macOS, Windows, Android and iOS, with no account required to download it and no subscription needed for the core browsing experience.
The browser ships with an AI assistant docked in the sidebar. Unlike a separate chat tab, this assistant can see whichever page you're currently looking at, hold context across your whole session, and act inside the page itself: filling forms, clicking buttons, switching tabs, and reading content it wasn't explicitly shown. Perplexity calls this "agentic" browsing, and the shopping use case is where the company has invested the most visible effort.
What makes Comet different from a shopping chatbot:
- It doesn't require a dedicated shopping mode. You don't open a "shopping" tab. You just ask the assistant a question while you're on any page, or in the sidebar with no page open at all, and it decides whether to search, compare, or act.
- It reads pages you're already on. If you're looking at a product page, the assistant already has that context loaded and can compare it against competitors without you re-explaining what you're looking at.
- It can operate across tabs. Comet can open several retailer tabs at once, pull details from each, and return a single synthesized comparison, something a standard chat interface can't easily do.
The shopping features brands should actually know about
Cross-retailer product research
Comet's most consistent strength, according to hands-on reviews from eMarketer and others, is agentic shopping research: taking a loose, natural-language brief like "find me waterproof hiking boots under $150 that run true to size" and surfacing options by reading review sites, retailer listings, and editorial roundups, then summarizing the tradeoffs in plain language. This is closer to how a knowledgeable friend shops for you than how a search engine returns a page of blue links.
For brands, this means the "search result" your customer sees might never be a link at all. It might be a paragraph describing your product's fit, materials, and price versus three competitors, generated from whatever text and structured data Comet's assistant could scrape off your site and everyone else's.
Price comparison and promo code discovery
Comet can pull pricing across multiple retailers in a single request and, per multiple reviews including AutoDS's shopping guide, actively hunts for promo codes and discounts while a user is on a checkout page. If your brand runs frequent discount codes through a specific channel, an agent that scans coupon databases and forums in real time could apply them (or find better ones from a competitor) before your customer manually would.
Agentic checkout, with real limitations
Comet's assistant can attempt to complete an entire purchase end to end: researching, adding to cart, filling shipping and payment details, and submitting the order, all without step-by-step prompting. In practice, this is the least reliable part of the stack. eMarketer's review found the assistant successfully navigated a boot search through to the checkout page, then repeatedly hit errors trying to actually complete the transaction. A separate test asking Comet to book airport parking produced hallucinated dates and a false claim that the booking had succeeded.
I wouldn't read this as "agentic checkout doesn't work." I'd read it as "agentic checkout works often enough that Amazon sued over it, but not reliably enough that you should assume every AI-initiated purchase completes cleanly." Brands should expect a rising share of attempted purchases coming through agents like Comet, alongside a real abandonment rate at the final step that has nothing to do with your own checkout UX.
The Amazon lawsuit, and why it's the most important shopping story here
This is the part every ecommerce brand should actually understand, because it sets the legal boundary for what any AI browser can do on your site.
Amazon first flagged Comet's behavior in November 2024, then deployed technical blocks against it in August 2025, which Perplexity reportedly worked around within 24 hours. After a cease-and-desist in late October 2025 that Perplexity publicly refused (in a blog post titled "Bullying is not innovation"), Amazon filed suit in the Northern District of California on November 4, 2025. The claim: Comet's agent was logging into customers' Prime accounts and shopping on Amazon.com in ways that violated Amazon's terms of service, without Amazon's authorization, even though the actual account holder had authorized the agent to act for them.
In March 2026, a district judge sided with Amazon and granted a preliminary injunction blocking Comet from accessing password-protected Amazon pages. Perplexity appealed, and on August 4, 2026, the Ninth Circuit vacated that injunction, ruling that Perplexity's AI assistant is a tool acting on a user's direct instruction, not an independent party "accessing" Amazon's systems itself. The panel sent the case back to the district court for further proceedings under that framework, so this isn't fully settled, but the appellate ruling is a meaningful win for every company building agentic shopping tools, not just Perplexity.
Why this matters for your brand specifically: if you sell on a platform with strict terms of service around automated access (Amazon, big marketplaces, some DTC platforms with anti-bot protections), the legal ground under agentic browsers is shifting in real time. A platform blocking Comet today might not be able to sustain that block legally tomorrow. Don't build a shopping strategy around assuming any single marketplace can permanently wall off agent traffic.
The pivot from "Buy with Pro" to free agentic shopping
Comet's checkout capability grew out of Perplexity's earlier "Buy with Pro" feature, which limited in-chat checkout to Perplexity Pro subscribers. Ahead of Black Friday 2026, Perplexity revamped that into a free agentic shopping product available to all US users, with a notable structural change: PayPal merchants now act as the merchant of record, handling the actual purchase, customer service, and returns directly, rather than Perplexity sitting in the middle of the transaction. PayPal's buyer protection policies apply across the board. This puts Perplexity in more direct competition with OpenAI's Instant Checkout and Amazon's and Google's own shopping agents, and signals that Perplexity sees free, frictionless checkout (not a paywalled feature) as the strategic bet.
How Comet is different from chat-based shopping surfaces
A browser-native agent and a chat-based shopping answer solve the same problem in different ways, and the optimization implications aren't identical. Chat surfaces like ChatGPT lean on formal integrations such as the Agentic Commerce Protocol, a standard for handing off payment between the assistant, the merchant, and a payment processor. Comet, by contrast, mostly transacts by directly operating a normal checkout page.
| Comet (browser agent) | Chat-based shopping (ChatGPT, Perplexity search) | |
|---|---|---|
| Where it operates | Inside any live webpage, across tabs | Inside a chat window or search results page |
| How it discovers products | Live page scraping plus web search, in real time | Primarily structured product feeds and indexed content |
| Checkout method | Attempts to complete forms directly on merchant sites | Uses dedicated protocols like the Agentic Commerce Protocol or merchant partnerships |
| Reliability of purchase completion | Inconsistent, fails often at final checkout step | More consistent where a formal integration exists (e.g. Etsy, Shopify via ACP) |
| What brands optimize for | Clean HTML, accurate on-page pricing/specs, fast page loads, working checkout forms | Structured product feeds, schema markup, merchant-of-record integrations |
The practical takeaway: a feed-based integration (like submitting your catalog to OpenAI's commerce API) gets you reliably surfaced in one specific chat surface. Being "Comet-ready" is closer to classic technical SEO and UX hygiene, because Comet's agent is reading your actual live site, not a feed you submitted somewhere. If your product pages have accurate, current pricing in plain text (not buried in a JavaScript-rendered price widget that only loads after a click), clear stock status, and a checkout flow that doesn't require unusual steps like a CAPTCHA-heavy login, you're more likely to survive an agent's attempt to transact on your site.
What brands should actually do about this
Fix the basics an agent needs to read cleanly. Comet's assistant works by scraping and interpreting live pages. If your pricing lives inside an image, a delayed AJAX call, or a cookie-gated popup, an agent may misread it or give up. Server-rendered pricing, product specs in plain HTML, and schema.org Product markup all reduce the odds of a misfire.
Audit your checkout flow for agent-friendliness, not just human UX. Multi-step forms, unusual field labels, and aggressive bot-detection (which will often flag an agentic browser as suspicious traffic) are the most common failure points reviewers have documented. You don't need to build a special "agent checkout." You need a checkout that doesn't assume a human is the only thing filling it out.
Don't treat marketplace terms of service as a permanent shield or a permanent risk. The Amazon ruling shows this area is legally unsettled. If you sell primarily through a marketplace that's currently blocking agentic browsers, that block may not hold. If you sell direct-to-consumer, don't assume you're immune either, since your own terms of service and bot-detection tooling could end up in a similar dispute down the line.
Watch the PayPal merchant-of-record shift closely if you run payments through PayPal already. Perplexity's move to make PayPal merchants the merchant of record for its free agentic shopping product is a lower-friction entry point than negotiating a direct feed integration with OpenAI or Google. If you already process through PayPal, ask your account rep whether you're eligible.
Track Comet-driven traffic separately in your analytics. Referral data from an agentic browser often looks unusual (direct traffic with no referrer, or a generic Perplexity user-agent string) rather than a clean UTM-tagged click. If you haven't already set up AI-driven traffic tracking, you're probably undercounting how much of your traffic already comes from agents like this one.
Why this is a different problem from "ranking in AI answers"
Most AEO advice right now is about getting cited or recommended inside a generated text answer. Comet is a different animal because the "answer" isn't text at all, it's an action taken on your actual website. You're not optimizing to be mentioned. You're optimizing to survive an automated visit that reads your page and tries to transact on it without your usual analytics, cookie consent flow, or checkout safeguards behaving the way they do for a human.
I think this is the part most brands are underestimating. It's easy to focus on getting a product card to show up in a ChatGPT answer, since that's visible and measurable. It's much harder to notice that an agentic browser tried to buy your product, failed at checkout because of a form quirk nobody ever tested against a bot, and quietly moved on to a competitor instead. That failure mode doesn't show up in a bounce-rate report. It shows up as a sale you never knew you almost had.
Where this goes next
Comet is still evolving fast, and the legal fight with Amazon isn't over, just remanded. Expect three things to keep shifting: which marketplaces try to block agentic browsers next, how reliably checkout actually completes as Perplexity refines the assistant, and whether other browser makers (Google's Chrome, Microsoft's Edge with Copilot) push harder into the same agentic-shopping lane, since Comet's biggest long-term risk isn't Amazon's lawsuit, it's competition from browsers people already have installed.
If you're running an ecommerce brand and haven't audited your site for how it behaves when an automated agent (not a human, not a known crawler) tries to browse and buy from it, that's a reasonable next project for this quarter. It's a small, concrete piece of work with a clear test: open your own checkout in Comet and try to buy your own product.
If you want help figuring out where your brand actually stands across ChatGPT, Perplexity, Google AI Mode and now agentic browsers like Comet, AI Peekaboo tracks exactly this kind of visibility. Email filipe@aipeekaboo.com or book a 30-minute call and we'll walk through your current setup.
