Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Google plans to overhaul its search engine to add features that will allow it to function similarly to an AI chatbot, as the company races to compete with rivals such as OpenAI.
From Tuesday, all US users will be able to activate “AI mode” in Google search and Chrome browser that will provide a conversational, question and answer experience akin to OpenAI’s ChatGPT, rather than a traditional list of links.
Sundar Pichai, chief executive of Google’s parent Alphabet, described the move as a “total reimagining of search” during its annual I/O developer conference in Mountain View this week.
The release represents the latest effort to inject generative AI into Google’s dominant search engine, even as investors worry the new products will cannibalise its cash cow.
The new “AI Mode” follows the introduction of “AI Overviews” at I/O last year. The brief, AI-generated answers to common topics have become popular with users, but have reduced click rates on adverts that the company relies on for much of its revenue.
Pichai has been criticised for the slow pace of AI releases after Google was previously the industry leader that developed the research behind the generative AI boom. That has allowed start-ups like OpenAI and Anthropic to grow into multibillion-dollar companies that are starting to challenge Google’s monopoly over online queries.
In response, Google has been forced to accelerate its launch timetable and make swift improvements to its large language model Gemini. It claims the newest 2.5 version surpasses OpenAI’s ChatGPT and Anthropic’s Claude on many benchmark tests, especially coding.
Pichai said Google still enjoys a vast distribution advantage over rivals with its 8.5bn daily enquiries. “Search is bringing AI to more people than any other product in the world,” he said.
Executives were tight-lipped about how they plan to integrate advertising into the new AI search, browser and app-based agentic offerings. Google made $50bn in search advertising revenue in the first quarter, more than half the total $90bn for Alphabet overall.
As it pivots to AI, Google is moving further from its ad-supported free model. Google’s standard “AI pro” subscription plan costs $25 a month.
Those who want early access to the most advanced features will have to pay $250 a month for a new “ultra” package, more expensive than OpenAI’s top $200 a month tier.
Google also joined the rush of companies offering AI agents, with its prototype called Project Mariner to be rolled out to subscribers over the summer. Users can allow the agent to take control of their browsers and other software to execute tasks such as booking trips, shopping and compiling complex research reports.
Google users can also grant Gemini access to their historical data to provide more “personal context” to actions and responses. However, experts have raised concerns over privacy as AI companies start building long memories into their models.
“Our recent updates to Gemini are critical steps towards unlocking our vision for a universal AI assistant,” said Sir Demis Hassabis, head of its AI research arm DeepMind. “It’s intelligent and understands the context you are in, and can plan and take actions on your behalf across any device. This is our ultimate goal.”
Many of the “agentic” features announced on Tuesday are already on offer from smaller rivals, with Anthropic and OpenAI releasing their own versions of computer use agents in October and January, respectively. However, those products remain glitchy and often struggle with basic tasks. Google could gain an edge if it is able to offer a more reliable experience using its vast resources and trove of user data.
Like Microsoft, which is hosting its own developers’ conference in Seattle this week, Google said it would adopt Anthropic’s Model Context Protocol (MCP), a framework which allows AI agents to communicate with each other in a standardised format, essential for navigating between multiple apps and websites.
Google also previewed an upgraded version of Project Astra, a “multimodal” agent that can respond and act on real-time voice queries and commands, using a mobile phone camera or smart glasses to look at the world. A video demonstration showed a man asking for advice on how to fix a bicycle, ordering parts from a local store as he worked.
Google ultimately plans to integrate Astra directly into search, but will also allow access to devices running Apple iOS software, not just those using its Android mobile operating system.
