How Agentic Browsers Are Changing SEO

The launch of agentic browsers and agentic browsing extensions is transforming the web from a manual navigation experience into a machine-operable layer where AI agents both read and act on content autonomously.

For SEO, this represents a shift from visibility to operability. Traditional ranking factors remain important, but they are becoming the baseline for creating content and websites that AI agents can understand, navigate and act on.

This article explores this shift and its impact on user behaviour at both the discovery and the conversion stages of the funnel. You will find out how AI agents ingest content and interact with the web, how to detect agentic traffic on your website, what protocols have emerged to help make websites agent-friendly, and how to structure content for AI ingestion.

What Are Agentic Browsers, and Why Do They Matter?

In a traditional web browsing scenario, you typically use a search bar to land on a web page and then interact with it.

In an agentic browsing experience, the user can “send” an AI assistant to visit the page on their behalf and perform the action a human would typically take, from “reading” the page’s content and summarising it for the user to performing more complex actions, like filling out forms and making purchases.

But that’s not all. Users can also personalise their AI assistants by allowing them access to their browsing history, and in the case of Gemini, also mailboxes, photos and YouTube. As a result, users get more relevant, tailored responses and also don’t have to switch between tabs that much.

So, the key things that set agentic browsers apart from traditional browsers are:

• Autonomy: You can literally delegate some tasks to AI assistants, like checking the menu at 10 different restaurants and giving you a recommendation based on your dietary preferences.
• Multi-tasking: You might be reading a blog on party decoration ideas in the main window, while your AI assistant is coming up with a menu for 10 people and adding groceries to your shopping cart.
• Personalisation: Your agentic browser can know so much more about you than a traditional browser or an AI tool like ChatGPT would, which means it can help you in so many new ways. For example, your browser can summarise those 5 videos about running shoes that you watched on YouTube last week, select the best shoes for your training purposes and find stores nearby that sell them.

At the same time, you can still use an agentic browser as a regular one and also choose what answer engine to use in it – a traditional one, like Google, or conversational. For example, in Comet, when entering a query, you can choose whether to search via Google or Perplexity, allowing a hybrid approach depending on your needs. You can even set a different search engine as your default, ensuring traditional SERPs remain accessible if desired.

Here is a quick overview of the agentic browsers and AI-embedded browser experiences currently available.

Perplexity Comet

Perplexity was the first to release its own agentic browser, Comet, back in July 2025. Comet can help automate research workflows, browse autonomously, and continuously refine results as new information emerges. Designed as an AI-first browser, it replaces traditional search-driven interactions with proactive AI assistants that guide users toward outcomes rather than links.

Instead of juggling tools or switching contexts, Comet supports a wide range of tasks, from in-depth research and meeting preparation to coding and ecommerce, directly within the browsing experience. It remains the most prominent standalone agentic browser and the clearest example of what a fully AI-native browser looks like in practice.

ChatGPT: From Atlas to Embedded Browser

ChatGPT Atlas launched in October 2025 as OpenAI’s standalone agentic browser, available for macOS only. Built on a Chromium-based framework with an integrated AI chatbot, it could browse the web, summarise YouTube videos, assist with job hunting, and complete shopping workflows. On August 9, 2026, Atlas was discontinued.

What replaced it reflects a strategic shift in how OpenAI thinks about agentic browsing. Rather than asking users to adopt a new browser, the ChatGPT Work launch on July 9, 2026, brought the same browsing capability into tools users already have.

The update introduced a built-in browser in the ChatGPT desktop app, with the same agentic capabilities as browsing pages and completing multi-step workflows, delivered through a different mechanism.

OpenAI also updated its Chrome extension to bring ChatGPT into Chrome’s sidebar, giving existing Chrome users access to its own agentic features without switching products. OpenAI cited Atlas learnings as the foundation for this new approach.

AI in the Browser You Already Use: Extensions and Embedded AI

Early discussions about agentic browsing suggested that it required adopting new software. The 2026 picture is quite different. AI is moving aggressively into the browser people already use through native integrations and extensions that require no switching at all.

Gemini in Chrome

Chrome evolving into an agentic browser was an expected catch-up move from Google. In January 2026, Google started rolling out Gemini-powered auto-browse to users in the US. Built on the Gemini 3 model, Chrome’s agentic features let you complete workflows, pull context from multiple web apps, and multitask across the web without switching between tabs.

Google extended this further in April 2026 with the launch of AI Mode in Chrome, a Search-integrated AI sidebar embedded directly in Chrome, with side-by-side page browsing and cross-tab search capability.

→ Your Guide to Visibility in Google AI Mode: How Agentic Browsers Are Changing SEO

Chrome Extensions from OpenAI, Perplexity and Anthropic

ChatGPT, Claude and Perplexity all offer browser extensions for Chrome, giving users access to AI-powered browsing without leaving their existing browser. These extensions are compatible with other Chromium-based browsers, such as Brave, Arc and Microsoft Edge. Additionally, AI add-ons and extensions are also available for other browsers, like Safari and Firefox.

Outside the consumer browser space, Anthropic has integrated a built-in browser window into Claude Code, its AI-powered development tool, allowing the AI to read, click and type on external websites directly within the development environment.

Extensions and native browser AI face no switching-cost barrier. They reach the existing browsers’ user base immediately. The addressable audience for AI-assisted browsing is now far larger than that of the dedicated agentic browsers. The SEO implication of all this is important: websites need to be optimised not only for discoverability and UX, but also to be operable by AI agents.

How This Shift Is Reshaping Search Behaviour

Traditional browsers have shaped how people search by acting as passive gateways to the web. You type a query, scan results, open multiple tabs, and manually piece together information. 

For decades, search behaviour has been built around navigation rather than outcomes. We’re now starting to see signs of the shift in that dynamic. 

By embedding AI directly into the browsing experience, browsers introduce memory, context awareness, and the ability to take action. Instead of moving between pages, users increasingly delegate tasks: researching, summarising, filling forms, or completing workflows without leaving the browser environment.

Personalised Discovery Loops

One of the most powerful features of agentic browsers is their ability to remember context and browsing intent. With this information, Atlas, Comet, and Gemini-powered Chrome can create what we think of as “personalised search conversations.”

Unlike traditional browsers, AI agents can track your actions, preferences, and prior searches to deliver more relevant answers over time. 

For example, if you’re researching the best laptops for video editing, the agent remembers your previous queries about performance, price range, and reviews. Over multiple sessions, it can proactively summarise insights, suggest comparisons, or even surface new products that match your interests, and you don’t have to repeat yourself twice.

However, this persistent memory layer introduces a new complexity. Because agentic browsers retain contextual data, they can also store sensitive information such as your credentials and personal details. 

Malicious actors can exploit this through techniques like memory injection or data poisoning by embedding hidden prompts or manipulated data into webpages that influence what AI remembers and later recalls. 

What happens then is the altered memory that may shape future responses, leak private information, or even enable reidentification of data that was originally anonymised. 

Potential for a Zero-Click User Journey

AI tools like ChatGPT, Perplexity, and Google’s AI overviews and AI Mode have already eroded organic click-through rates. By satisfying queries directly in the SERP or the synthesised response, they’ve commoditised information, often removing the need for a user to ever visit your site.

With the rise of agentic browsers, the traffic that does reach your website is increasingly non-human. Theoretically, the entire user journey, from discovery to conversion, can now happen without a human ever landing on a page.

This has forced a 180-degree turn from industry gatekeepers. Last year, Cloudflare introduced a one-click “block all AI bots” feature as a default defensive measure. Now, they’ve pivoted to launching Markdown for Agents. This tool allows websites to serve a machine-readable version of their content specifically for AI crawlers, effectively acknowledging that blocking bots is no longer a viable strategy for growth.

“As a business, to continue to stay ahead, now is the time to consider not just human visitors, or traditional wisdom for SEO-optimization, but start to treat agents as first-class citizens.”

Cloudflare

From Passive Research to Autonomous Action

Agentic behaviour is currently still concentrated at the top of the funnel. HUMAN’s 2026 State of AI Traffic report found that 77% of agentic AI activity occurs on product and search pages (more on this in the Adoption Statistics section below). While bots are excellent at synthesising options, they are not yet prolific buyers – only a small fraction of these interactions currently reach checkout or account-level execution.

This gap in transactional autonomy is a result of three main friction points:

• Infrastructure: Many checkout and account flows are designed for human users rather than AI agents and APIs.
• Browser Functionality: Agentic browsers such as Comet and agentic browsing add-ons like AI Mode in Chrome or ChatGPT’s in-built browser are still developing the capabilities needed to reliably hand off and complete complex tasks.
• User Trust: Users remain reluctant to delegate decisions and authority, particularly for purchases and recurring subscriptions.

That does not mean agentic commerce is limited to research. As websites become more agent-friendly, developments like Cloudflare’s machine-readable layers and Google’s WebMCP (Web Model Context Protocol) help transition from AI-assisted research to AI-assisted transactions, although it may not happen entirely inside the AI interface itself.

The OpenAI Instant Checkout story illustrates why that friction proved harder to overcome than anticipated. Launched in September 2025, Instant Checkout allowed users to complete purchases inside ChatGPT. By March 2026, after roughly five months, OpenAI pulled the feature.

The reasons were structural:

• User behaviour skewed heavily toward discovery and comparison rather than completing purchases inside the chat.
• Real-time product data synchronisation across millions of retailers did not scale (only around 12 of Shopify’s millions of merchants went live).
• Fraud prevention added further friction for users who did attempt to complete transactions.

OpenAI has since moved towards routing transactions through third-party retailer apps, including Instacart, Target and Booking.com. The Agentic Commerce Protocol (ACP), co-developed with Stripe, continues to support app-based purchases, but with a narrower scope.

This is better understood as a change in where the transaction happens, rather than a retreat from agentic commerce. AI can still influence the purchase journey by helping research products and compare options, sending qualified users to retailers.

For SEO, that distinction matters. A brand does not necessarily need an AI system to complete the checkout itself to benefit from agentic traffic. Being surfaced in ChatGPT’s shopping results can still generate qualified visits to a retailer’s website, making on-site conversion architecture, product data and Merchant Centre feed quality increasingly important.

The Broader Agentic Commerce Picture

The shift away from native AI checkout does not mean that agentic commerce is slowing down. Instead, the industry is catching up by building infrastructure that allows AI agents to discover products, access merchant data and hand users off to retailers for the final transaction.

Here’s how we know:

• Shopify launched Agentic Storefronts in March 2026, enabling millions of merchants to surface products inside ChatGPT, Google AI Mode, Gemini, Microsoft Copilot, and Meta. The experience is managed centrally through Shopify Admin and is active by default for eligible stores.
• Google’s Universal Commerce Protocol (UCP), co-developed with Shopify, is also being built to support agent-led commerce, including native checkout for Copilot and, for select US merchants, Google AI Mode and Gemini.
• Adobe reported 693% year-on-year growth in AI-referred retail traffic during the 2025 holiday season. Shopify reports that AI-driven traffic to merchant stores increased 8x year-on-year in Q1 2026, while AI-attributed orders increased nearly 13x.

For now, AI is likely to play a bigger role in product discovery and referrals than in completing purchases, with checkout remaining on merchant websites and apps. But the infrastructure for end-to-end agentic commerce is rapidly developing.

For ecommerce sites, the preparation is the same either way: maintain clean, structured data, make pages accessible to AI agents, and provide product feeds that AI can parse and trust.

→ eCommerce GEO: A Guide to AI Search Optimisation for Online Stores: How Agentic Browsers Are Changing SEO

Expert Insight: Agentic Browsing Is The Future Standard

While the technical shift toward agentic browsing is clear, the implications for brand communications and SEO strategy are even more profound. As browsers evolve into agents, traditional habits like scrolling and manual page analysis are being replaced by intent-driven automation.

In this clip, SUSO’s VP of Partner Growth, Cara Corbett, and PR expert Andrew Bruce Smith discuss why agentic browsers are set to become the new standard, and how this shift is compressing the funnel – from brand sentiment analysis to autonomous task completion.

Adoption Statistics of Agentic AI

HUMAN analysed more than one quadrillion digital interactions in 2025, revealing how quickly agentic traffic is moving from experimentation to measurable activity. Traffic from AI agents and agentic browsers grew 7,851% year over year, making it the fastest-growing category of AI-driven traffic.

What is particularly notable is where that traffic is going:

• 77% of agentic AI activity occurred on product and search pages.
• 8.82% occurred on account pages.
• 4.95% occurred on authentication flows.
• 2.31% reached checkout pages.

The figures show that agents are already moving beyond simply reading web content, but product discovery and research remain their primary use cases.

The commercial impact is also becoming clearer:

• 46.6% of agentic traffic went to retail and ecommerce sites.
• 28.5% went to streaming and media.
• 19.2% went to travel and hospitality.
• For agentic browsers specifically, ecommerce accounted for 55.8% of traffic.

This suggests that agentic browsing is moving quickly from a niche experiment towards a meaningful source of digital traffic. For brands, the opportunity is no longer limited to being discovered by AI. Websites increasingly need to support the agents that navigate pages, access accounts and, eventually, complete transactions on users’ behalf.

How AI Agents Read the Web

Brands and marketers have already started switching their focus from getting ranked in traditional search engine results pages to getting cited in AI-generated responses. Agentic browsing is accelerating this shift as AI agents now move beyond discovering and summarising content to interacting with websites and completing tasks on users’ behalf.

This means that the goal of AI search optimisation should now be to attract and convert agentic traffic, alongside human users.

Before we dive into how to do exactly that, it is important to first look at how today’s agentic and AI-embedded browsers interpret and interact with the web.

Content Parsing and Source Ranking Logic

How an AI agent reads a page depends on what it’s there to do. Agents summarising content for citations, like ChatGPT and Perplexity, work mostly from raw text and structure. Agentic browsers acting on a page, like Gemini in Chrome or Comet, need more than that to click a button or fill in a form.

Per Google’s own developer guidance for building agent-friendly websites, agents draw on up to three sources to understand what’s on a page:

• Screenshots: a vision model reads the rendered page. Useful for judging layout, size and position, but slower and more resource-intensive to process.
• Raw HTML: the DOM itself, showing how elements are nested and related to one another.
• The accessibility tree: a stripped-down, browser-native structure originally built for screen readers. It describes each element’s role, name and state, and ignores styling entirely.

Most capable agents don’t rely on just one of these. They typically read the DOM and accessibility tree for structure, then cross-check that against a screenshot to understand layout and grouping. This is also why semantic HTML, using <button>, <nav> and <main> rather than generic <div> elements, is becoming as relevant to agent readiness as it has long been to accessibility.

When an AI agent lands on a page, here’s what it typically does:

1. It breaks down the text into chunks and interprets the structure

AI agents process text in structured blocks, reading key headings and paragraphs in order of relevance. This “chunking” process is guided by structural signals that tell the agent what to ignore and what to extract:

• Heading hierarchies: H1, H2, H3
• Semantic groupings: logical paragraph clusters, bulleted and numbered lists, and tables
• Structured data: JSON-LD markup (Organisation, NewsArticle, FAQ Schema, and others).

However, standard HTML is often cluttered with “noise” – navigation menus, tracking scripts, and UI elements that may confuse AI crawlers. This is why Cloudflare’s Markdown for Agents is a significant shift in technical SEO. By automatically converting a webpage into a clean Markdown version, Cloudflare allows agents to bypass the “bloat” of traditional web design and ingest the core information directly.

2. It evaluates relevance and authority signals

Once the text is parsed, the agent begins filtering what information is worth using. This is where source quality and topical trust come into play.

AI agents tend to prioritise content that shows strong domain and clear authorship. This is why major media organisations are cited more than 27% of the time. Rather than pulling random paragraphs, the system looks for credible, high-confidence sources that align closely with the user’s intent.

3. It extracts, cross-checks, and synthesises information

After identifying a trusted and relevant section, the agent doesn’t simply copy and paste the content word for word. Instead, it synthesises a consolidated answer or summary. Some agents leave a link to the source as well.

That’s why precise language, clean metadata, and focused explanations perform well in AI environments.

This still describes how agents make sense of a page as it stands today. WebMCP, covered later in this article, points to a different model: instead of an agent inferring what a page does from its structure, a site declares it directly. 

Risks and Vulnerabilities

Agents can evaluate content, but they aren’t immune to manipulation. The level of autonomy they receive is not without risks. Because AI agents act on your behalf with the same access privileges as a human user, they can be exploited by malicious actors hidden in web content, documents, and emails.

• Malicious workflows: Attackers can trick agents into granting access to sensitive accounts, like business emails or cloud storage.
• Prompt injection attacks: Malicious instructions embedded in trusted apps can lead the AI to take unintended actions, such as embedding harmful links in calendar invites.
• Disguised malware downloads: AI agents cannot fully inspect files, so a seemingly harmless file required for a workflow could contain malware or ransomware.

Workflow-based attacks are another real risk of using agentic browsers. AI assistants excel at getting things done, but they don’t have the common-sense instincts to avoid malicious links or unsafe downloads.

Agentic browsers are 85% more vulnerable to phishing attacks than traditional browsers. Manual browsing allows time to stop and think, while AI assistants take instant actions since they automatically process web content and may act with elevated permissions. 

Optimising for Agentic Search: The New Rules of AI-Native SEO

It’s tempting to say agentic AI will make classic SEO obsolete, but the reality is more nuanced. Early investigations into ChatGPT Atlas’s behaviour reveal that it still relies on Google’s search ecosystem, leading to the conclusion that good old SEO fundamentals still matter. 

When pages were opened through Atlas, analysts observed that Googlebot sometimes crawled those pages immediately afterwards, often identifying those visits as coming from Google’s crawler rather than a unique Atlas user agent. This suggests Atlas relied heavily on Google’s infrastructure and indexing behaviour to surface web content.

This overlap means that traditional signals like crawlability, structure, and authority still play a role in making your content discoverable, even within AI-driven browsers. But at the same time, AI agents are changing how discoverability works.

This hybrid environment, where classic SEO still influences what AI systems see, but AI readability and citation potential determine what they answer with, has given rise to a new concept often called AI-Native SEO.

AI-Native SEO doesn’t abandon traditional search optimisation but also optimises for how AI systems read, understand, and summarise content.

1. Structure for Readability, Not Just Indexing

AI systems reward clarity. Use a clear semantic hierarchy (H1–H3) and include bullet summaries, TL;DR boxes, and section highlights to make content scannable.

Keep paragraphs short and context-rich as AI models favour well-structured text and are more likely to lift it for summaries or answers.

2. Prioritise E-E-A-T and Entity Signals

Factual credibility remains crucial. Include prominent author bios, brand mentions, and outbound citations to reinforce trustworthiness.

Leverage schema markup for Organisation, Author, and Article, and integrate verified Linked Data to help AI systems understand the relationships between entities, authors, and sources.

3. Optimise for AI Quotations

Write with the expectation that AI agents may lift your content directly into answers. Use concise, quote-ready phrasing, short definitions, and structured explanations that can be naturally extracted.

Answer key questions clearly and provide bite-sized insights that are easy for AI assistants to summarise for users.

4. Improve Crawlability and Page Context Density

Even AI-native strategies rely on solid technical foundations. That’s why ensuring fast load times, clean code, and proper metadata is also essential for both AI comprehension and user experience.

Use internal linking and site structure to group semantically related concepts together so AI models can capture the full topic coverage and provide accurate, contextual summaries.

This matters just as much for agents that act on a page, not only ones summarising it. As covered earlier, agents reading the DOM and accessibility tree rely on proper semantic HTML, such as <button>, <nav> and <main>, rather than generic <div> elements with click handlers, to identify what a page contains and what it lets a user do. A page that scores well on accessibility is, by extension, easier for an agent to interpret and act on reliably.

If your website runs on Cloudflare, consider enabling Markdown for Agents to make it easier for AI bots, crawlers, and agents to consume your content.

5. Track AI Visibility

Traditional SERP metrics aren’t enough anymore. Start monitoring AI visibility to understand how often your brand or content is cited or referenced in LLM-generated responses.

Build internal dashboards to track visibility across ChatGPT, Perplexity, Claude, Gemini, and other AI agents, and use this data to refine your AI-native SEO strategy over time.

Additionally, start monitoring AI referral traffic in Google Analytics. Your GA4 dashboards should also show you if AI traffic is driving conversions and revenue, provided the key events are set up correctly.

Besides tracking the general AI citations number and visibility share, it is important to monitor the sentiment and accuracy of these mentions and citations. With AI actively summarising and quoting your content, there’s a growing need for AI content verification, watermarking, and bias-aware optimisation.

Monitoring how your content is interpreted ensures your brand maintains visibility safely and ethically. By proactively auditing AI citations and ensuring your content is responsibly structured, brands can protect their reputation while thriving in agentic ecosystems.

6. Prepare for the Agent-Ready Web: WebMCP, ACP, and UCP

As AI agents move from reading web pages to taking action, the way websites expose information and functionality is becoming increasingly important. Three protocols are particularly relevant here: WebMCP, ACP, and UCP. They solve different parts of the problem, so it is worth looking at what each one actually does.

WebMCP (Web Model Context Protocol) is Google’s approach to helping AI agents interact with websites through structured interfaces rather than relying entirely on traditional page elements or DOM scraping. In early preview within Google’s ecosystem, it supports two ways for websites to expose actions:

• Declarative APIs: Define common actions directly in HTML, such as sign-ups, bookings, and support requests.
• Imperative APIs: Use JavaScript for more complex or dynamic interactions, such as configuring products or applying filters.

Unlike the commerce-focused protocols below, WebMCP is not limited to online shopping. Any website that needs an AI agent to perform an action could potentially benefit from it.

ACP (Agentic Commerce Protocol), co-developed by OpenAI and Stripe, focuses specifically on commerce within AI-powered experiences. It supports product discovery, cart creation and delegated payments.

Following the pullback of OpenAI’s Instant Checkout, however, its role has become more focused. Larger retail partnerships can support more direct purchasing experiences, while smaller merchants may still rely on AI-driven discovery that sends users back to their own websites.

UCP (Universal Commerce Protocol) takes a broader approach to agentic commerce. Backed by Google, Shopify, Visa, Mastercard, and Stripe, it is designed to connect merchants, AI agents and commerce platforms around tasks such as identity, orders and payments.

Because it works across platforms that already host large product catalogues, merchants can gain agentic commerce capabilities without building every integration themselves.

For most websites, this does not mean rushing to implement three new protocols. The more immediate priority is making sure the information agents rely on is complete, accurate and easy to interpret.

Ecommerce businesses should continue to keep product data, pricing, availability, shipping information, and structured data up to date. Platforms such as Shopify and Google Merchant Centre can handle much of the technical infrastructure behind agentic commerce, but they cannot fix inaccurate or incomplete information.

The principle is simple: an agent needs to be able to understand what you sell, trust the information it finds, and act on it. Clean data is therefore becoming table stakes in the competitive landscape, not just a technical SEO task. An agent comparing products against a price limit or delivery requirement may favour the competitor whose information is clearer and easier to verify.

7. Optimise for Agentic Commerce Discovery

AI agents are becoming an increasingly important part of product discovery. ChatGPT’s shopping results, Perplexity’s product carousels, and Google’s AI Mode can all surface products from structured, indexed sources.

A December 2025 Semrush survey found that 43% of US shoppers had already discovered a new brand through AI. With Instant Checkout no longer the focus, the referral model is increasingly straightforward: discover through AI, then complete the purchase on the retailer’s site.

So what makes a product more likely to appear in AI-generated shopping results? The fundamentals are familiar, but the emphasis is shifting towards information that agents can easily understand and verify:

• Accurate product feeds: Keep pricing, availability and shipping information up to date in Merchant Centre and other relevant feeds.
• Complete Product schema: Include key details such as price, availability and reviews.
• Conversational product copy: Address the kinds of questions shoppers actually ask, rather than relying solely on keyword matching. This matters when shoppers provide specific constraints upfront: 52% state requirements such as budget, features or compatibility when searching.
• Rich product information: Include materials, compatibility, use cases, specifications and useful Q&A wherever relevant.
• High-quality images: Provide clear, relevant product images and enough visual information to help both shoppers and AI systems understand the product.

For Shopify merchants, much of the underlying infrastructure is already being connected to AI shopping experiences. As mentioned earlier, Shopify Agentic Storefronts allow eligible merchants to surface their catalogues across channels including ChatGPT, Google AI Mode, Gemini, Copilot and Meta.

That makes product data quality even more important. Complete product information, populated metafields and strong review coverage can influence how much useful information AI systems have to work with across these channels.

→ Webinar: Building Discoverable Ecommerce Stores for AI-Powered Search: How Agentic Browsers Are Changing SEO

Monitoring Agent Activity

To monitor agentic traffic on your website, start with three approaches: log file analysis, CDN bot insights, and GA4 referral tracking. Each captures a different slice of the picture, and when used together, they give you the most complete view of how AI agents interact with your site.

-> Should You Block AI Bots from Your Website?: How Agentic Browsers Are Changing SEO

How AI Agents Appear in Analytics Tools

Monitoring Agent activity is undoubtedly important, but it’s also challenging because these agents don’t behave exactly like traditional bots. In Google Analytics 4, AI agents can appear as:

• Referral traffic from sources such as chat.openai.com or perplexity.ai 
• Direct traffic when no referrer is available 
• Unassigned when the platform cannot classify the visit 

In some cases, activity from AI systems like OpenAI’s agent has been reported as Bing organic or paid traffic because the agent may perform a search through Microsoft Bing before visiting your website. 

These nuances make monitoring the traffic source harder to interpret, especially since AI agents can mimic human browsing patterns. As a result, analytics data may feature high engagement rates, unusually low bounce rates, or misleading session duration metrics. 

Limitations of Tracking AI Agents in GA4

Tracking this activity in analytics platforms is also limited by privacy and consent requirements. AI agents cannot legally provide cookie consent, and even if they technically click “Accept,” it doesn’t count as valid consent for analytics tools.

The sessions that come from AI agents are often only captured if:

• Tracking scripts run before consent is obtained
• Your website uses server-side or cookieless tracking setups that do not rely on consent-based cookies

Because of these limitations, many AI agent visits either appear as direct traffic, are partially tracked, or are filtered out entirely by analytics systems.

-> How to Track AI Traffic in GA4: How Agentic Browsers Are Changing SEO

Using Log Files to Detect AI Agents

To gain more accurate insights, analysing server log files is often the most reliable method.

Log files capture every request made to a website in its rawest form, including data that analytics platforms may filter out or never record. Access logs typically contain the following:

• Visitor’s IP address
• Timestamp
• Requested URL
• HTTP status code returned by the server
• Referral URL
• User agent string that identifies the browser or bot

With this information, it may be possible to detect AI agents, analyse how they navigate a site, and understand how frequently they access specific pages. 

Additional Ways to Monitor AI Agent Traffic

Beyond analytics tools and log files, there are a few other ways to monitor AI agent activity on your website. 

• CDN with bot insights

Using a Content Delivery Network (CDN) with bot insights can help detect and label traffic from AI agents, monitor how automated systems interact with your website, and block or rate-limit certain crawlers if necessary. 

• Server-side analytics without cookies 

Server-side analytics without cookies can also provide anonymised insights into AI agent visits while remaining compliant with regulations like GDPR, although this approach typically offers less detailed behavioural data.

• Analysing user behaviour 

Instead of only looking at traffic sources, you can identify AI agents by how they behave on your site. For example, sessions with an engagement time of 0 seconds, where a visitor immediately clicks a link or moves to another page, often indicate automated activity rather than a real user.

You can also find cases where users appear inactive and simply navigate from one page to another without meaningful interaction, which is another sign of bot traffic. 

Creating segments in analytics platforms based on these behavioural signals or unusual traffic sources can help identify and monitor potential AI agent visits more effectively. Additionally, some website backends show “wordpress.com” as a traffic source, which can indicate visitors coming from ChatGPT-related browsing sessions.

Monitoring AI traffic is currently one of the bigger analytics challenges, but in the near future, it will become a critical issue for businesses. Traffic generated by AI will continue to grow, which makes it essential to distinguish between “reading bots” and modern “shopping bots” that perform real tasks on behalf of customers.

We need to prepare for entirely new attribution and remarketing rules. A situation where an AI agent browses products on Monday, and the final conversion appears a few days later from a normal source, will become standard. Without precise monitoring of these agents’ activity in server logs, our data on user behaviour and the real effectiveness of sales will simply become distorted.

Marcin Walkowiak, SEO Expert at SUSO

Building for the Browsers That Think and Act

Search has evolved into conversation. The future of agentic browsing lies in AI capabilities distributed across the tools people already use.

Comet remains one of the clearest examples of a dedicated agentic browser. But the bigger shift in 2026 is that agentic browsing is no longer limited to specialist products. Chrome’s embedded AI, ChatGPT’s built-in browser experience, and AI extensions are bringing agentic capabilities to users without requiring them to switch browsers.

For SEO practitioners, this means the window for preparation is getting shorter. Optimising content for AI extraction is no longer enough. Websites also need to support AI-first user journeys, where agents can understand content, navigate pages and take action on behalf of users.

The shift is ultimately from visibility to operability. The brands that prepare for both will be better positioned as AI becomes an increasingly active participant in how people discover, evaluate and buy online.

FAQs

• What is an agentic browser?

  An agentic browser uses AI to browse the web and complete tasks on a user’s behalf, such as browsing pages, comparing options, filling forms, and completing tasks without requiring manual input at each step. Rather than just displaying results, it functions as a delegate that can execute goals autonomously.

• What are the security risks of agentic browsers?

  Because AI agents act with the same access privileges as the person delegating to them, they can be exploited by malicious content hidden in web pages, documents and emails. Key risks include prompt injection, where hidden instructions trick an agent into taking unintended actions, malicious workflows that trick agents into granting access to sensitive accounts, and disguised malware in files an agent can’t fully inspect before downloading. Agentic browsers have been found to be significantly more vulnerable to phishing than traditional browsers, partly because agents act instantly rather than pausing to assess a link or request.

• How do I detect agentic traffic on my website?

  The most reliable method is log file analysis, which captures raw request data, including user agent strings that identify AI crawlers. You can also use Google Analytics and referral traffic data to identify traffic from AI platforms, while behavioural patterns can provide additional context. Because AI agents do not always identify themselves consistently, no single signal is completely reliable.

• How is agentic search optimisation different from traditional SEO?

  Traditional SEO focuses on ranking in search engine results pages so human users click through to your site. Agentic search optimisation focuses on making content easy for AI systems to find, understand, cite and act on. The technical SEO fundamentals still matter, but agentic search adds another layer: whether an AI agent can reliably interpret and interact with your website.

• What is the accessibility tree, and why does it matter for AI agents?

  The accessibility tree is a browser-native structure, originally built for screen readers, that describes each element on a page by its role, name and state, and ignores styling entirely. AI agents use it alongside raw HTML and screenshots to work out what’s on a page and what actions are available. Sites built with semantic HTML, using <button>, <nav> and <main> rather than generic <div> elements, produce a cleaner accessibility tree and are easier for agents to interpret reliably.

• What is agentic commerce?

  Agentic commerce is the use of AI agents to discover, compare and purchase products or services on a user’s behalf. It can range from AI-powered product recommendations to delegated tasks such as adding products to a cart, placing orders and completing payments.

• Can AI agents make purchases on my behalf?

  Some can, though full in-chat checkout has proven harder to scale than expected. OpenAI’s Instant Checkout, launched in September 2025, was discontinued by March 2026 due to fraud risk and the difficulty of syncing real-time product data across millions of retailers. Most agentic commerce today works differently: an AI agent helps with discovery and comparison, then hands the user off to the retailer’s own site to complete the purchase.

• What is WebMCP?

  WebMCP (Web Model Context Protocol) is a proposed web standard, developed by Google and Microsoft through the W3C, for enabling AI agents to interact with web pages structurally, rather than through fragile DOM scraping. It allows sites to expose defined actions, such as bookings, purchases, and support requests, that agents can trigger reliably. It has moved from early Chrome preview into an origin trial, though adoption on live websites is still minimal, and it represents the next evolution of agent-ready web infrastructure rather than something to implement today.

• What are ACP and UCP?

  ACP (Agentic Commerce Protocol) and UCP (Universal Commerce Protocol) are two competing standards for agentic commerce, alongside WebMCP. ACP, co-developed by OpenAI and Stripe, focuses on product discovery, cart creation and delegated payments within AI-powered experiences. UCP, backed by Google, Shopify, Visa, Mastercard and Stripe, takes a broader approach, connecting merchants, AI agents and commerce platforms around identity, orders and payments across multiple channels. Unlike WebMCP, which supports any type of agent action, both protocols are purpose-built for commerce.