For most of the web's history, a browser was a neutral window: it fetched a page, drew it on screen and left the reading to a person. AI browsers change that bargain. They sit an assistant next to (or on top of) every page, so a visitor can ask for a summary, compare your prices with a competitor's in another tab, or hand the whole task to an agent that clicks through your site on their behalf.
The category has also moved fast. OpenAI launched ChatGPT Atlas in October 2025 and has since retired it, folding its browser agent into the ChatGPT desktop app. Perplexity made its Comet browser free. Google built Gemini into Chrome and added an agentic "auto browse" feature. If you run a website, the practical question is not which browser wins. It is what happens when software, rather than a human, reads, summarises and operates your pages.
This guide explains how today's AI browsers work, what they do to visits and analytics, and the specific site changes worth making in 2026, without overreacting to a category that is still being reshaped every few months.
Key Takeaways
- AI browsers add three behaviours to normal browsing: reading (summarising the page you are on), remembering (optional memory of visited pages) and acting (agents that click, fill forms and add items to carts).
- The landscape is volatile: ChatGPT Atlas launched on 21 October 2025 and OpenAI scheduled it to stop working on 9 August 2026, moving its agent features into ChatGPT itself.
- Gemini in Chrome is the one with the largest built-in audience; its auto browse agent is a paid (AI Pro/Ultra) feature in the US.
- Many AI-browser visits look like ordinary Chrome sessions, so expect blurred analytics rather than a neat new referrer.
- The best preparation is unglamorous: clean HTML, accessible forms and buttons, clear pricing and policies, and fast, stable checkout flows.
What makes a browser an "AI browser"?
An AI browser is still a browser: it renders HTML, runs JavaScript and stores cookies. The difference is an assistant with access to the page content and, in agent mode, to the controls. That gives it three distinct jobs, and each one affects your site differently.
- Reading and summarising. A sidebar assistant reads the page the user is already on and answers questions about it. The user still visited you, but may read less of your copy.
- Remembering. Some browsers can, with permission, recall pages a user visited earlier and bring them back into later chats.
- Acting. An agent navigates, scrolls, fills forms and clicks buttons to finish a task such as booking, comparing or adding items to a basket, usually pausing for confirmation on sensitive steps.
If you want the bigger picture of agents beyond the browser, our explainer on agentic search and how AI agents browse and buy covers the wider shift.
The main AI browsers in 2026
Here is where the major products stand as of October 2026, based on each company's own announcements and help pages.
| Product | Status (Oct 2026) | Assistant on page | Agent that acts | Who can use the agent |
|---|---|---|---|---|
| ChatGPT Atlas (OpenAI) | Launched 21 Oct 2025 on macOS; deprecated, scheduled to stop working 9 Aug 2026 | "Ask ChatGPT" sidebar | Agent mode (moved into the ChatGPT desktop app and Chrome extension) | Originally a preview for Plus, Pro and Business |
| Comet (Perplexity) | Free for everyone since 6 Oct 2025 (desktop) | Built-in assistant that summarises pages and manages tabs | Background tasks such as comparing shopping options | All users; Comet Plus is a paid news add-on |
| Gemini in Chrome (Google) | US on Windows, macOS and Chromebook Plus; all US Android users since Aug 2026 | Summaries and Q&A about the page, linked Google apps | Auto browse (multi-step tasks) | Google AI Pro and AI Ultra subscribers in the US |
The Atlas story is a useful warning. OpenAI's help centre says it is "deprecating Atlas and moving browser-based agentic capabilities into ChatGPT and Codex". Building a strategy around a single AI browser is risky; building a site any capable agent can use is not.
How AI browsers read your pages
When a user asks a sidebar assistant to "summarise this" or "compare the plans on this page", the assistant works with whatever the browser has loaded. That has a few consequences:
- Content hidden behind tabs, accordions or images without text alternatives is harder for the assistant to use reliably.
- Prices, plan limits and policies that only appear in images or after interaction may be missed or misread.
- A page with one clear answer per section is far easier to summarise accurately than a long block of marketing copy. Our guide to writing passages AI can quote applies here too.
OpenAI was explicit that site owners can help its agent. In the Atlas launch post it wrote:
“Website owners can also add ARIA tags to improve how ChatGPT agent works for their websites in Atlas.”
— OpenAI, Introducing ChatGPT Atlas
In other words, the same accessibility work that helps screen-reader users helps software agents. Our accessibility and SEO guide is a good starting point.
How AI browsers act on your site
Google describes auto browse as "a powerful agentic experience that handles multi-step chores on your behalf", with examples such as comparing hotel and flight prices, filling out forms, getting service quotes, and finding products from a photo, adding them to a cart and applying discount codes. Google also says auto browse pauses and asks for confirmation before sensitive actions such as purchases or posting on social media.
For a business, that means your conversion paths may be walked by an agent. If the agent hits a CAPTCHA wall, a pop-up it cannot dismiss, or a button labelled only with an icon, the task fails and the user may simply try a competitor. Google has also said Chrome will support its Universal Commerce Protocol (UCP), an open standard for agentic commerce, which suggests retail checkouts will increasingly be designed for agents as well as people. See our guide on preparing your website for AI agents for the technical side.
What AI browsers do to traffic and analytics
This is the part most site owners care about, and where honesty matters most: there is no reliable public data yet on how much AI-browser activity changes visits, and the effect will differ by site type.
| Behaviour | Likely effect on visits | What you may see in analytics |
|---|---|---|
| Sidebar summary of your page | The visit still happens, but scroll depth and time on page may fall | Normal session, shorter engagement |
| Assistant answers from memory of earlier visits | Fewer return visits for simple facts | Lower repeat sessions; no referrer |
| Agent completes a task on your site | A real session, but navigated by software | Unusual paths, fast form fills, possibly flagged as bot-like |
| Agent compares you with competitors | Several quick page views across sites | Short sessions on pricing and product pages |
Because these browsers are built on standard browser engines, many sessions may not carry a distinctive referrer or user agent. Treat any vendor-specific detection as something to verify against the vendor's current documentation. For practical methods, read our guides on tracking ChatGPT and AI traffic in GA4, detecting AI agent visits and why AI traffic often shows up as direct.
A step-by-step plan to prepare your site
- Audit your key journeys. List the five tasks that make you money (request a quote, book, buy, sign up, call). Walk each one with keyboard only; if a person cannot finish it without a mouse, an agent will probably struggle too.
- Label every control. Use real
<button>and<a>elements, visible text oraria-labelon icon buttons, and<label>elements tied to form fields. - Put facts in text. Prices, plan limits, delivery times, opening hours and return policies should be in HTML text, not only in images or PDFs.
- Reduce friction walls. Review aggressive interstitials and cookie banners that block content, and check that bot protection does not hard-block legitimate browser sessions.
- Add accurate structured data. Product, Organization and LocalBusiness markup that matches visible content helps machines interpret pages; see our guide to structured data for AI search.
- Annotate and monitor. Note the dates of major AI-browser launches in your reporting, watch engagement on pricing and product pages, and test your top journeys in at least one AI browser each quarter.
Common mistakes
- Betting on one browser. Atlas lasted less than a year as a standalone product. Optimise for standards, not for a brand.
- Blocking everything that looks automated. Agent sessions run inside real browsers on behalf of real customers. Blanket blocking can turn away buyers.
- Hiding key facts in images or scripts. If the assistant cannot read your price, it may quote a competitor's instead.
- Confusing training opt-outs with browser access. OpenAI says pages that opt out of GPTBot are excluded from training even when a user opts in, but that is a training control, not a way to stop a user's browser from reading the page they opened.
- Reading too much into short sessions. A drop in time on page may reflect summaries, not worse content. Check conversions before rewriting.
Where this fits in your wider AI search strategy
AI browsers are one surface among many. Google's own guidance for AI features in Search says there are no extra technical requirements beyond normal SEO best practice, and that holds broadly here: crawlable, fast, accessible and factual pages work everywhere. Pair this work with a strong technical SEO foundation and the landing-page advice in our guide to AI referral landing pages.
Related Guides
- Apps in ChatGPT: Should Your Brand Build One?
- Ads in ChatGPT: What Advertisers Need to Know
- ChatGPT Memory and Personalization: Why Every User Sees Different Answers
Frequently Asked Questions
Is ChatGPT Atlas still available?
No. OpenAI announced it was deprecating Atlas and scheduled it to stop working on 9 August 2026. Its browser-based agent features moved into the ChatGPT desktop app, with a ChatGPT Chrome extension where available.
Do AI browsers count as real visits in my analytics?
Usually yes. A sidebar summary or an agent task runs inside a real browser session, so it generally appears as a normal visit, often without a distinctive referrer. Engagement patterns may look different.
Can I block AI browsers from reading my pages?
Not in a practical way. They load pages like any browser on behalf of a user. Robots.txt tokens such as GPTBot control crawling and training, not a person viewing your page in their browser.
Who can use Gemini in Chrome auto browse?
Google says auto browse is available to Google AI Pro and AI Ultra subscribers in the US, on desktop since January 2026 and on Android since August 2026. Gemini in Chrome itself is more widely available in the US.
What is the single most useful change to make?
Make your key journeys accessible: labelled buttons and form fields, text-based prices and policies, and no blocking pop-ups. That helps people with disabilities, regular visitors and AI agents at the same time.
Conclusion
AI browsers turn every page into something that can be summarised, remembered and operated by software. The products will keep changing, as Atlas showed, but the winning response stays the same: clear facts in text, accessible controls and frictionless journeys. If you want a second pair of eyes on how your site performs for both people and agents, our SEO services team can audit your key journeys, or you can request a free quote.
References
- OpenAI — Introducing ChatGPT Atlas
- OpenAI Help Center — Evolving Atlas into ChatGPT for browser-based agentic work
- Google — The new era of browsing: putting Gemini to work in Chrome
- Google — Gemini in Chrome on Android and auto browse
- TechRepublic — Perplexity's Comet AI browser is now free for everyone (Oct 2025)
- Google Search Central — AI features and your website
