Google used its annual I/O developer conference Wednesday to move generative artificial intelligence from a collection of experiments into the center of its product strategy, introducing its new PaLM 2 language model, expanding Bard to more users and previewing a version of Google Search that generates synthesized answers above conventional web results.

The company said PaLM 2 is already powering more than 25 Google products and features, including Bard and new Workspace tools. In its I/O announcement collection, Google presented generative AI as the connective tissue across Search, productivity software, cloud services, Android and consumer devices — a broad response to the rapid adoption of conversational systems such as OpenAI’s ChatGPT.

PaLM 2 becomes Google’s new foundation model

Google introduced PaLM 2 as a family of language models designed for stronger multilingual performance, reasoning and coding. The company’s PaLM 2 announcement says the model was trained with broader multilingual text and is intended to perform better across more than 100 languages, while specialized variants are being developed for domains including medicine and cybersecurity.

Rather than positioning PaLM 2 as a standalone chatbot, Google is embedding it into existing products. Bard has already been moved to PaLM 2, Google says, improving its ability to handle programming, mathematics and complex reasoning tasks. That distribution strategy draws on an advantage Google already possesses: billions of users interact daily with Search, Gmail, Docs, Android, Maps and YouTube.

Chief Executive Sundar Pichai framed the shift as one of the most consequential technological transitions the company has encountered. In his I/O keynote summary, Pichai said Google is at an “inflection point” in AI and described the company’s goal as making the technology useful across consumer products while developing it responsibly.

Search begins testing generated answers

The most strategically important announcement concerns Google Search, the company’s core business and the foundation of its advertising model. Google is opening a Search Generative Experience, or SGE, through Search Labs in the United States. The experiment can present an AI-generated overview for some queries, organize relevant information and allow users to ask conversational follow-up questions.

Google’s Search announcement describes the system as a way to answer more complicated questions that might otherwise require multiple searches. The generated overview is designed to sit alongside links to the web rather than replace the underlying information ecosystem, and Google says it is applying additional restrictions to queries where reliability or safety concerns are higher.

The company is deliberately labeling the experience experimental. Search is a higher-risk environment than a general-purpose chatbot because users often expect factual, current and authoritative answers. Large language models can generate fluent but incorrect statements, and Google must manage that weakness without damaging trust in a product used for billions of queries.

To control that rollout, Google introduced Search Labs, a testing environment where users can opt into experimental features and provide feedback. That structure gives the company room to change models, interfaces and sourcing before generative answers become a standard part of Search.

Bard loses its waitlist and becomes more connected

Google is also expanding Bard beyond the limited U.S. and U.K. launch that began in March. The company is ending the waitlist in more than 180 countries and territories, adding Japanese and Korean, and promising additional languages. It is also making Bard more visual and preparing connections with Google services and outside applications.

In the Bard update, Google said PaLM 2 has improved the system’s coding and reasoning capabilities and that future integrations will allow people to move work between Bard and products such as Gmail and Docs. The company also plans to incorporate images into prompts and responses, extending the assistant beyond plain text.

The expansion reflects competitive pressure but also a shift in how Google views conversational AI. Bard is no longer being treated primarily as a response to ChatGPT. It is becoming an interface layer for Google’s broader software ecosystem, capable of drafting, analyzing, coding and eventually invoking other services.

Generative AI moves into everyday office work

Google is simultaneously extending generative AI across Gmail, Docs, Slides, Sheets and Meet under the Duet AI for Workspace name. In its Workspace announcement, the company described tools for drafting and rewriting documents, generating images for presentations, organizing information in spreadsheets and assisting with meetings.

The productivity push places Google directly against Microsoft, which is adding OpenAI-powered Copilot features to Microsoft 365. The competitive question is not simply which company has the strongest base model. Enterprise adoption will also depend on reliability, security, data controls, integration with existing workflows and whether AI assistance produces enough measurable productivity to justify new costs.

Google’s distribution advantage is again significant. Workspace already sits inside the daily routines of businesses, schools and individuals. Embedding AI into those applications can make the technology accessible without requiring users to learn a separate product, but it also raises questions about how generated content is labeled, how organizational data are protected and how errors propagate into business documents.

Google emphasizes responsibility while accelerating deployment

The speed of the announcements creates a tension Google is trying to address explicitly. The company says it will move boldly on AI while applying safeguards, red-team testing, model evaluation and restrictions in sensitive contexts. A separate responsible-AI statement argues that long-term progress requires both rapid innovation and controls designed to reduce misuse, misinformation and unintended harms.

That balance will be tested most visibly in Search. A chatbot that invents an answer can frustrate a user; a search engine that confidently synthesizes an incorrect medical, financial or civic answer could create much broader consequences. Google is therefore trying to preserve links, attribution and user choice while changing the interface from a ranked list of pages toward a system that can compose an answer.

The company also faces an economic challenge. Search advertising generated the business model that funded much of Google’s AI research. Generative responses consume more computing resources than traditional search results and could change how users interact with sponsored links and publisher websites. The technical product may evolve faster than the commercial architecture around it.

By the end of I/O week, however, Google’s direction is unmistakable. PaLM 2 is becoming a common model layer, Bard is becoming a broader assistant, Workspace is receiving generative tools and Search — the company’s most important product — is beginning its most substantial interface experiment in years. The competition in generative AI has moved beyond model demonstrations. It is now a contest over who can integrate the technology into products people already use every day without undermining the trust those products depend on.