Technical

How to Optimise for AI Browsers: ChatGPT Atlas, Perplexity Comet, and Dia

An AI browser is a full web browser with a generative model embedded in the session itself, one that reads, summarises, and acts on the page you are currently viewing rather than answering from a separate chat window. ChatGPT Atlas, Perplexity Comet, and Dia are already changing which pages get read, quoted, and clicked. Here is how optimising for an AI browser differs from optimising for a chat-based answer engine.

Neil Walsh·July 2026·8 min read

An AI browser is a standard web browser with a generative AI model wired directly into the browsing session, reading the live page a user is looking at, summarising it on request, filling in forms, and completing multi-step tasks such as booking, comparing, or checking out, instead of routing every question through a separate chat window. OpenAI's ChatGPT Atlas, Perplexity's Comet, and Dia (built by the team behind Arc and acquired by Atlassian in 2026) are the three that matter right now, and each one changes the citation and click-through mechanics that AEO has spent the last two years learning to optimise for.

The distinction matters because an AI browser does not just retrieve and summarise a page the way ChatGPT or Perplexity's chat interface does. It renders the page, sometimes clicks through it, and can act on what it finds, filling in a checkout form, extracting a price, or navigating to a linked page without the user ever seeing the raw HTML. A page that reads well to a retrieval crawler can still fail inside an AI browser if it cannot be rendered, interacted with, or trusted enough to act on.

Why AI browsers are a distinct optimisation surface

Chat-based answer engines such as standard ChatGPT, Claude, and Google AI Overviews work from an index: a crawler fetched your page at some point in the past, chunked it, and stored a retrievable version. An AI browser often skips the index entirely. It navigates to your live URL in real time, in a real rendering engine, and reasons over what actually loads in that moment. That means the JavaScript-rendering failures that quietly hide content from GPTBot and ClaudeBot matter less here, because the browser executes your scripts just like a human visitor's browser would, but a slow page, a cookie banner that blocks the main content, or a login wall that a crawler would simply skip past can now stop an agentic session in its tracks.

It also means the unit of success is not always a citation. Inside a chat answer, being cited means your name and link appear in the response. Inside an AI browser, success can be the model choosing to click through to your page, filling in your form, completing a purchase on your site, or quoting your content back to the user without ever surfacing a link at all. Measuring AI browser visibility the same way you measure ChatGPT citation share will understate the channel, because a large share of the value shows up as a completed action, not a footnote.

The three AI browsers you need to know

ChatGPT Atlas

OpenAI shipped Atlas in October 2025 as a full Chromium-based browser with ChatGPT built into the sidebar and an agent mode that can carry out multi-step tasks such as research, shopping, and job hunting inside the live page. It launched on macOS first, with Windows and mobile following later, and it reads whatever is currently rendered in the tab rather than pulling from a separate index, which means standard on-page structure and load speed carry more weight than they do for a pure chat retrieval flow.

Perplexity Comet

Comet moved from a paid, invite-only product to a free download on every major platform in 2026 and climbed as high as third place on the iOS App Store shortly after. It leans on the same citation-dense retrieval style Perplexity's chat product is known for, but applies it while browsing live, surfacing sourced summaries and comparison tables as the user moves between tabs rather than waiting for a single chat query.

Dia

Dia, built by the team behind the Arc browser, was acquired by Atlassian in 2026 for a reported $610 million, signalling that AI browsers are being treated as a durable enterprise surface rather than a consumer novelty. Dia's model reads open tabs as shared context, so a page that makes sense only alongside a specific referring page or search result can lose meaning once it is read in isolation inside a browser session.

Traffic volume from AI browsers is still small, generally well under 1% of total organic sessions for most sites in 2026, but the sessions that do arrive convert at a disproportionately high rate. Treat it as an early, high-intent channel worth instrumenting now, not a mass-traffic replacement for organic search.

How AI browsers decide what to read, quote, and act on

Because an AI browser renders your live page rather than a crawled snapshot, the signals it weighs most heavily are the ones that affect a real browsing session:

  • Render speed and stability: a page that is slow to hydrate, or that shifts layout after load, can cause an agent to misread or abandon the page before extracting the intended content.
  • Clean, unblocked main content: cookie banners, paywalls, and interstitials that a human would dismiss in one click can stall or derail an automated session that has no instinct to click 'accept'.
  • Machine-actionable forms and buttons: labelled inputs, semantic HTML, and predictable checkout or contact flows let an agent complete a task instead of giving up and reporting failure to the user.
  • Self-contained passages: the same declarative, answer-first writing that wins chat citations also gives a browsing agent something quotable the moment it lands on the page, without needing to scroll or infer context from a different tab.
  • Schema and structured data: Product, FAQPage, and Organization markup still give an agent a fast, unambiguous read of price, availability, and identity, which matters even more when the agent is deciding whether to act, not just whether to quote.

An optimisation checklist for AI browsers

  1. Test your key pages inside Atlas, Comet, and Dia directly, not just with a standard crawler check, since rendering failures and blocked interstitials only show up when something actually browses the live page.
  2. Remove or defer any cookie banner, modal, or login wall that sits in front of your primary content, or provide a crawlable, agent-friendly path around it.
  3. Keep critical content in the initial render rather than behind a client-side fetch that fires after user interaction, since an agent session may not trigger the interaction a human would.
  4. Add or verify Product, Organization, and FAQPage schema on transactional pages, since AI browsers lean on structured data heavily when deciding whether to complete an action.
  5. Write the primary answer or price point as a single self-contained sentence near the top of the page, so an agent skimming the live render does not need surrounding tabs for context.
  6. Monitor referrer strings and user-agent tokens associated with Atlas, Comet, and Dia in your analytics now, even at low volume, so you have a baseline before the channel grows.

If your site already passes a standard AEO audit, schema in place, crawlers allowed, declarative passages, the incremental work for AI browsers is mostly about removing friction in the live rendering path: banners, modals, and slow scripts that a retrieval crawler never had to deal with in the first place.

What doesn't change from standard AEO

AI browsers are a new surface, not a new discipline. The same declarative sentence structure that wins a ChatGPT citation is what a browsing agent quotes back to the user, the same Schema.org markup that improves chat citation odds is what lets an agent act confidently on price and availability, and the same robots.txt access checks still apply to the crawlers these companies also run for indexing, even though the live browsing agent itself typically respects a normal browser user-agent rather than a bot token. Building for AI browsers is an extension of AEO fundamentals into a rendering environment that behaves more like a real visitor and less like a crawler, not a separate discipline to learn from scratch.

Run a free CiteRank audit to check whether your key pages have the crawler access, schema, and passage clarity that both chat-based answer engines and agentic AI browsers are selecting for.

Frequently asked questions

What is an AI browser?

An AI browser is a standard web browser, such as ChatGPT Atlas, Perplexity Comet, or Dia, with a generative AI model built into the session itself. It reads the live page a user is viewing, can summarise or answer questions about it on request, and in agent mode can fill in forms, click through pages, and complete multi-step tasks such as comparisons, bookings, or checkout.

How is an AI browser different from ChatGPT or Perplexity chat?

Chat interfaces answer from a previously crawled and indexed version of the web. An AI browser instead renders your live page in real time and reasons over whatever actually loads in that moment, including JavaScript, cookie banners, and interactive elements a standard retrieval crawler would never encounter.

Does AI browser traffic show up in Google Analytics yet?

Partially. Sessions from Atlas, Comet, and Dia typically arrive with a normal browser user-agent rather than a bot token, so they blend into standard organic or direct traffic unless you specifically look for the referrer patterns and known IP ranges associated with each product.

Should I block or allow AI browser agents in robots.txt?

Live browsing agents generally behave like a real visitor's browser and are not the same as the training or retrieval crawlers, such as GPTBot or PerplexityBot, that robots.txt rules typically target. Blocking those separate crawler tokens can still remove your content from the index an AI browser's chat mode draws on, so check your existing AI crawler access rather than trying to block the browser itself.

Why would a cookie banner hurt me in an AI browser but not in a chat answer engine?

A retrieval crawler fetches raw HTML and never has to interact with a banner, so it usually reads straight through to the content underneath. An AI browsing agent renders the live page and may stall, misread the page, or fail to complete a task if a banner or modal sits in front of the content and nothing in the session dismisses it automatically.

Is Dia the same as the Arc browser?

Dia was built by the team behind Arc, and it carries over Arc's tab-centric design, but it is a separate AI-first browser product. Atlassian acquired Dia's parent company in 2026 for a reported $610 million, positioning it as an enterprise-facing agentic browsing product rather than a direct continuation of Arc.

How much traffic do AI browsers actually drive in 2026?

Still relatively little in absolute terms, generally well under 1% of total organic sessions for most sites, but early data suggests visitors arriving through agentic AI surfaces convert at meaningfully higher rates than typical organic traffic, which makes the channel worth instrumenting even while the volume stays small.

What is the single highest-leverage fix for AI browser readiness?

Removing anything that blocks or delays your primary content from rendering immediately, cookie banners, slow client-side hydration, and login walls chief among them, since an AI browsing agent has no guaranteed instinct to dismiss an interstitial the way a human visitor would.

Free tool

See your AEO score in seconds

Paste your URL and get a full audit across all 9 AEO signals - schema, crawlers, E-E-A-T, and more.

Audit my site - it's free

Related reading

Technical

Why Schema.org markup is the single biggest lever for AI citation

May 2026
Technical

Is your robots.txt accidentally blocking ChatGPT and Claude?

May 2026