Perplexity Comet explained: what the AI browser means for your website

How Comet reads your site

Perplexity's index collects pages ahead of time through its crawler, while Comet's assistant opens your page inside the user's browser when their request needs it in real time. The index decides whether you show up in an answer at all, while the live visit decides what the agent understands and does once it gets to your page.

During the live visit, Comet sees more than a human does. Brave's security team showed that Comet feeds part of the webpage straight to its language model and doesn't separate the user's instructions from the page's own text. Hidden spans and collapsed tabs go to the model, as does text behind spoiler tags. Because Comet is a full Chromium browser, it also runs your JavaScript the way Chrome does. That's a difference from simpler crawlers, which read only the raw HTML.

How much the agent sees depends on the visit. A logged-in user brings their cookies and session along, so the agent can open account pages that no crawler would ever reach. To check what it reads, turn off styling and read the plain text of a key page from top to bottom. That plain text is close to what the model gets.

What Perplexity Comet actually is

Comet is a consumer browser with Perplexity's answer engine inside it. It isn't a search engine or a crawler. Perplexity released it in July 2025 to Max subscribers paying $200 a month, then made it free worldwide on October 2, after a waitlist that the company said had reached "millions." According to TechCrunch, free users get the sidecar assistant, while paid plans add background tasks.

Comet's name gets used loosely for five separate Perplexity products, and each one touches your site differently. The table sets them side by side.

Pages Comet can act on

Comet can act on any page its user can open, and that includes pages behind a login. The agent works inside the user's own session, so your checkout and booking forms are open to it, and so are your account dashboards. If a person could click it, the agent can try.

Retailers have started to push back against this. Amazon sued Perplexity in November 2025 and claimed Comet was "disguising Comet as a Google Chrome browser" while it shopped for users. In March 2026, U.S. District Judge Maxine Chesney temporarily blocked Comet from Amazon's site and wrote that Amazon had shown "strong evidence" of access "without authorization."

Most sites will land somewhere in between, and how much the agent can do varies from page to page. A CAPTCHA or a multi-step email check can stop it partway through a task. Bot management at your firewall can also stop it. So go through each conversion page and ask whether you want an agent to finish that task. Then test whether it can.

How agentic browsers rank content

Agentic browsers don't hand the user a ranked list of ten links. They pick a small set of sources to build one answer, then pick the elements on a page that finish a task. The order of your page in a results list matters less than whether your page makes the final cut.

The first cut happens in retrieval. Perplexity searches its index for pages that match the user's request, so a page PerplexityBot never indexed can't compete. The second cut happens inside the model. It reads the pages it retrieved and keeps the passages that answer the question most directly. A paragraph that states a price or a spec in plain words is easier to lift than one that hints at it. The same applies to dates.

For a task such as "book a demo for Thursday," the agent looks for a button and a form it can identify with confidence. Clear labels win over clever design. If you check one thing here, read your main call-to-action buttons and ask whether the text alone says what each one does.

Structured data Comet prefers

Perplexity hasn't published a list of schema types that Comet favours. Treat structured data as a way to state facts so a machine can't misread them, and don't expect it to work as a ranking switch. The agent reads the rendered page in any case. So schema helps most when it confirms what your visible text already says.

Agents need facts about your products and company that schema.org markup can carry.

• Product with Offer, for price and availability
• Organization, for your legal name and logo
• Article, with datePublished and dateModified, so freshness can be read from the page
• SoftwareApplication, for plan names and pricing on SaaS pages
• BreadcrumbList, to show where a page sits on your site

Mismatches are the most common mistake. When the schema lists a price that doesn't match the price on the page, the model gets 2 conflicting facts and risks quoting the wrong one. Check your markup in a validator, then compare each value with the visible text on the same page.

How citations work in Comet

Comet cites the pages that its search step found and its model actually drew the answer from. You earn a citation when PerplexityBot can index your page and your page holds the clearest passage that answers the question. The citation sits next to the answer as a link, so the user can open your page and check the claim.

Perplexity also pays for some of this traffic now. Under Comet Plus, publishers in the program get 80% of subscription revenue from a starting pool of $42.5 million, Jessica Chan, Perplexity's head of publisher partnerships, told Axios. At the Comet Plus launch, Perplexity described requests to synthesize recent industry coverage as indexed traffic. It described Comet Assistant's calendar scans and article suggestions for the day's meetings by saying "that's agent traffic."

Most business sites aren't in that program, but the same split still applies to them. Indexed traffic depends on the crawler, while agent traffic depends on how readable the page is.

Optimising for agentic crawlers

The 3 changes that make your site readable to Comet start with crawler access. You'll also need to change the HTML your key facts sit in and any text you meant only for machines. None of these requires a new site. Each one is a configuration or content fix your team can test within a week.

Let Perplexity's crawlers in

Allow PerplexityBot in robots.txt. Perplexity's own crawler documentation recommends this "to ensure your site appears in search results" and adds that a web application firewall (WAF) needs an explicit rule for its published IP ranges. Don't count on robots.txt to keep Perplexity-User out, because that fetcher runs on a user's request. If you do want to block it, you'll need firewall rules. Cloudflare measured 3 to 6 million daily requests from an undeclared agent that looked like Chrome on macOS, alongside 20 to 25 million from Perplexity's declared ones. Perplexity disputes how Cloudflare attributed that traffic.

Put the answer in the HTML

Prices and plan limits belong in text, as do specs. An image or a PDF puts those facts at risk of being skipped or misread by the agent. Use real `<button>` and `<label>` elements so the agent can tell what a control does. And render the facts that matter on the server, so the indexing crawler sees the same page your users see.

Remove text meant only for machines

Hidden keyword blocks and white-on-white text get read as content, with a risk that the model follows them like instructions. That's the same channel attackers use for prompt injection. Review comments and reviews that users posted on your pages too, because Brave planted its test instructions in a Reddit spoiler.

B2B pages most at risk

B2B pages that hide their key fact behind a human step are the ones most exposed to agent visitors. An agent that can't find a number reports that the number isn't listed. Then it gives the user a competitor that does list it.

• Pricing pages that only say "contact sales"
• White papers and reports gated behind a lead form
• Product specs published only as PDF datasheets
• Comparison pages where the competitor names appear only in images
• Demo request forms with unlabelled fields or a CAPTCHA on the first step

If your buyers research with an assistant before they talk to sales, a gated page means your company is missing from the shortlist that agent puts together. Publish a price range, or a summary of the report, as text. Keep the form for the full version.

When this matters for you

This matters as soon as your buyers research with an assistant. Comet has been free since October 2025, so buyers already have access to it for that research. Start by checking whether PerplexityBot can reach your pricing and product pages and whether your firewall lets it through. Then open your own pricing page in Comet and ask the sidebar to summarise it.

Do the same for your main competitor's page. The difference between the two summaries shows you what Perplexity Comet tells your buyers today. Snoika audits websites and tracks brand visibility in Google and AI assistants, with human review on top. To see where you stand, check your AI visibility free.