ChatGPT Web Browser, Perplexity Comet, and Google Mariner: The Future of Search is Agentic

On July 9th 2025, the future of search fundamentally changed. Perplexity released their comet web browser, and Reuters published an article stating OpenAI will release a ChatGPT web browser in the coming weeks.

OpenAI's upcoming ChatGPT web browser, Perplexity's newly launched Comet browser, and Google's enhanced search capabilities aren't just competing products - they represent a key turning point in how we interact with the internet, and company search dominance.

For OpenAI and Perplexity, this is a core step in building products less dependent on Google infrastructure - escaping Google's 'operating system' to build platforms that can think, act, and execute tasks autonomously on behalf of users. It is also, importantly, an opportunity to collect user search data.

The timing of these releases couldn't be better: Google faces unprecedented antitrust pressure that limits its ability to defend its browser ecosystem.

Strategic Implications: The Battle for Platform Control

How Google's Antitrust Makes This a Perfect Opportunity

The race to build agentic web browsers is happening at a uniquely opportune moment. Google faces unprecedented legal pressure with multiple antitrust rulings stacking up:

• August 5, 2024: Federal Judge Amit Mehta ruled Google holds an illegal monopoly in search markets

• April 17, 2025: Second ruling found Google violated antitrust law in digital advertising technology markets

• Ongoing: Department of Justice seeks to force Chrome browser divestiture by August 2025

This creates an unprecedented window where competitors can challenge Google's dominance. Google can't aggressively defend its browser ecosystem without risking further antitrust violations. The company that once paid Apple over $20 billion annually to remain the default search engine now faces potential restrictions on these very deals that built its moat. (Apple has been actively looking for an AI partner amid Google partnership fall out)

Why are ChatGPT and Perplexity Creating their own Web Browsers?

ChatGPT already offers an app to make it the default search engine in Chrome. There are also a couple community chrome extensions that augment user experience with ChatGPT. However, for both business and technical reasons, there is strategic importance in platform control.

1. Data Collection & User Context

Besides market share, Google's competitive moat isn't just its search algorithm, it's the comprehensive user context from Gmail, Drive, GBP Reviews, Calendar, and most importantly user search history.

If you watched Google I/O 2025, you know these data sources are the cornerstone of Google Personal Context - something Google said they would be rolling out this summer, and have already enabled opt in for. Personal Context is just what it sounds like - personalized search results - whether it be through Google AIO, Gemini, AI Mode, or whatever they end up calling Mariner.

ChatGPT recently released memory - which some thought would make it the stickiest product in history - but the context of AI conversations is nothing compared to the vast stores of user data Google has. This is one of many reasons ChatGPT and Perplexity have created their own web browsers: to collect more user data. As Perplexity CEO Aravind Srinivas stated on the TBPN podcast:

“That’s kind of one of the other reasons we wanted to build a browser, is we want to get data even outside the app to better understand you,”

2. Agentic Capabilities Require Platform Control

Browser extensions simply can't implement the deep agentic features these companies envision. Autonomous task execution requires system-level access to authentication flows, payment processing, cross-site interactions, and tab management. When an AI agent needs to book a flight across multiple websites, manage authentication across platforms, and complete payment workflows, it needs control over the entire browsing environment.

3. Escaping Google's Data and Experience Control

Despite using Chromium as their foundation, these browsers control the crucial user experience layer. They can set different default search engines, implement native ad blockers, and embed AI assistants directly into the browsing interface. Platform control matters more than the underlying engine—just as Android's value comes from Google's services layer, not the Linux kernel beneath.

Importantly, Google cannot "pull the rug out" from under competitors using Chromium. The open-source project operates under permissive BSD-style licensing that allows anyone to use, modify, and distribute the code. Companies like Microsoft (Edge), Opera, Vivaldi, and Brave all build successful browsers on Chromium, and these licenses cannot be retroactively changed. While Google employees do most of Chromium's development, the project's open nature means competitors are free to fork and develop independently if needed. Further, if Chrome is divested, this would become even less likely.

Technical Capabilities: What Should we Expect from Agentic Web Browsers?

OpenAI's Operator & ChatGPT Browser Features

Computer-Using Agent (CUA) Model: OpenAI's CUA, ChatGPT Operator, combines GPT-4o's vision with reinforcement learning to interact with graphical interfaces through screenshots and mouse/keyboard actions, without requiring custom API integrations.

Cloud-Based Execution: Operator runs on remote servers rather than locally, enabling simultaneous task handling and centralized safety controls.

Performance Benchmarks: CUA achieves 38.1% success on OSWorld task automation (vs. Anthropic's 22.0% and human 72.4%). Google's Mariner scored 83.5% on WebVoyager browser tasks. These measure real-world task completion—booking reservations, filling forms—rather than search quality.

Current Limitations: Operator requires human takeover for logins, payments, and CAPTCHAs. Testing reveals hallucination problems with incorrect dates and booking errors, limiting reliability for critical tasks.

Perplexity Comet's Agentic Web Browser Features

Deep Integration: Comet embeds AI through homepage input, address bar, and sidebar, allowing conversations from any context. The AI can see and interpret current tab content for webpage summarization and contextual understanding.

Autonomous Task Management: Comet manages tabs autonomously and accesses browser history. With permission, it operates across Gmail and Google Calendar, shifting from assistance to active task execution.

Social Media Integration: When logged into platforms like X, users can prompt Comet to post replies via headless browsing mode, demonstrating machine-speed interaction with web applications.

Privacy Trade-offs: Full functionality requires extensive Google Account permissions—screen viewing, email sending, contact access, and calendar management. This enables powerful automation but demands significant data access. Comet Privacy Notice

Google Mariner Agentic Features

Technical Capabilities: Project Mariner uses Gemini 2.0 to take screenshots, process them in the cloud, and send navigation commands back to Chrome. Running on cloud-based virtual machines, it handles up to 10 simultaneous tasks without blocking users.

Mariner achieved 83.5% success on WebVoyager browser automation benchmarks. However, it cannot currently complete purchases, accept cookies, or agree to terms of service, requiring human oversight for sensitive actions.

Integration Strategy: Google integrates Mariner capabilities into the Gemini app via "Agent Mode" and Google Search's AI Mode, enhancing existing products rather than creating new platforms..

Browser Automation Protocols and Architecture

Current Implementation Approaches: Agentic browsers use proprietary automation systems rather than standardized protocols. OpenAI's Operator uses screenshot analysis with cloud-based control, while Perplexity Comet integrates directly with browser APIs and DOM manipulation.

For foreground operations, systems use computer vision and direct interface control. Background automation remains limited, with most current solutions requiring user supervision for complex workflows.

Authentication and Cross-Platform Challenges: Implementations face authentication barriers across sessions and website compatibility issues, as many sites block automated interactions. Speed bottlenecks remain significant, with agents requiring several seconds to analyze and act.

Agentic Browser Comparison: Key Differences

Key Takeaway: While all three leverage AI for web automation, they represent fundamentally different strategic approaches—OpenAI focuses on platform control, Perplexity targets search market disruption, and Google enhances its existing ecosystem dominance.

Conclusion: Appear Across Agentic Search

The emergence of the ChatGPT web browser, Perplexity Comet, and enhanced Google capabilities signals a fundamental shift in how we think about web interaction. Rather than one company "winning" it appears we're heading toward a fragmented agentic search landscape.

As AI agents gain autonomy in performing tasks on users' behalf, the traditional concept of "search" dissolves entirely. Users won't search for flights—agents will book them. Users won't research products—agents will compare and purchase them. The browser becomes an execution environment rather than an information discovery tool.

As search shifts to AI agents, brands must optimize for agent discovery rather than traditional SEO. This means understanding both the AI systems and the personal context of users you want to reach.