Google Search AI and Gemini app updates

The Antigravity-powered mini apps launch first for Google AI Pro and Ultra subscribers in the US.

The framing is clear. Search isn’t just an answer engine any more. It is becoming a build-an-answer engine.

Gemini app gets a 24/7 personal AI agent in Gemini Spark

According to Josh Woodward, VP of Google Labs, Gemini app and AI Studio, the Gemini app gets the year’s most visible redesign, alongside a roster of new agentic features. 

The new design language is called Neural Expressive. It introduces fluid animations, vibrant colour, new typography and haptic feedback across the whole Gemini experience, with Gemini Live conversational mode now integrated directly into the main app rather than living as a separate mode. 

Users can switch between typing and talking without losing context, thanks to a re-engineered mic that supports free-flowing thinking aloud rather than cutting users off mid-thought. Regional dialect support is coming soon. Neural Expressive is rolling out globally today across web, Android and iOS.

The bigger structural change is the move toward proactive, agentic features that work on the user’s behalf rather than waiting to be prompted.

Daily Brief is the new morning digest agent. Once a user opts in, Gemini runs in the background across connected apps, pulling updates from Gmail, tracking Calendar events, and compiling relevant follow-up details into a skimmable morning summary that suggests next steps. Users can train it by giving it a thumbs-up or thumbs-down over time. Built on Google Labs’ CC experiment, Daily Brief rolls out today to AI Plus, Pro, and Ultra subscribers, starting in the US.

Gemini Spark is the headline announcement of the keynote and arguably the most significant structural shift Google has made to Gemini since launch. Spark is a 24/7 personal AI agent that runs on Gemini 3.5 with the Antigravity harness and sits across the Workspace tools users rely on daily, including Gmail, Docs, and Slides. Because Spark is cloud-based, it keeps running when a user closes their laptop or locks their phone, allowing it to continue complex tasks for hours at a time.

Three use cases give a flavour of what Spark is actually meant for:

Spark also gets a growing list of MCP (Model Context Protocol) connections, with Canva, OpenTable and Instacart launching today, and a wider partner roster integrating now. In the coming weeks, Spark will be able to use those MCP connections to take action on behalf of users, plus add abilities to be texted and emailed directly, to create custom sub-agents, and to operate a local browser. Importantly, Google has built in a clear consent layer that requires Spark to ask first before any high-stakes action, such as spending money or sending an email.

Spark rolls out to trusted testers this week, with a beta for US Google AI Ultra subscribers landing next week. The Gemini app for macOS, available for download today, will also gain Spark and new voice features over the summer.

Gemini Omni: edit cinematic video by talking to it

Play

The third major announcement is Gemini Omni, a new generative media model line introduced by Koray Kavukcuoglu, CTO of Google DeepMind and Chief AI Architect at Google. Omni is Google’s next step after the Nano Banana image model that landed last year, and the framing is straightforward. Omni can create anything from any input, starting with video.

The first model in the family, Gemini Omni Flash, lets users combine text, images, video and audio as inputs and generate high-quality video outputs grounded in Gemini’s real-world knowledge. The model rolls out today to AI Plus, Pro, and Ultra subscribers globally through the Gemini app and Google Flow, and is available for free through YouTube Shorts and the YouTube Create App starting this week. Developer and enterprise access via API will follow in the coming weeks.

What sets Omni apart from the demos Google has shown is that editing happens through conversation. A user uploads footage, then types or speaks an instruction. Characters stay consistent across edits, the physics hold up, and the scene remembers what came before. You can see some examples of what prompts can be used and their results in the video below.

Play

Three claims from Google’s launch demos stand out for verification once retail testing begins:

Better physics understanding. Omni claims an improved, intuitive grasp of gravity, kinetic energy, and fluid dynamics, which should mean less of the uncanny, floaty motion that plagued earlier AI-generated videos.

Knowledge-grounded creativity. Omni draws on Gemini’s broader knowledge for context, which Google argues lets it produce more accurate explainers (e.g., claymation protein folding) rather than just visually plausible nonsense.

Reference any input. Omni can use any combination of images, text, video, and audio as references, with full audio input support, including voice, rolling out in stages.

Crucially for content workflows, Omni includes a built-in AI avatar feature that lets creators put a digital version of themselves in the video, with their own voice. Every Omni-generated video carries Google’s SynthID digital watermark, and users can verify any video’s origin through the Gemini app, Gemini in Chrome, or Google Search.

SynthID, C2PA and the case for verifiable provenance

The fourth pillar of the I/O announcements is content transparency, and the timing isn’t accidental. As Google ships a more capable video generation model, it is also rolling out verification infrastructure so users can check whether anything they see online was AI-generated.

Laurie Richardson, VP of Trust and Safety, and Pushmeet Kohli, Chief Scientist of Google Cloud and VP of Google DeepMind, shared a few key updates.

SynthID, introduced three years ago, has now watermarked over 100 billion images and videos, as well as 60,000 years of audio content. SynthID verification has been used 50 million times globally in the Gemini app since launch, and is now expanding to Search starting today and to Chrome over the coming weeks. Users can ask “Is this made with AI?” through Search Lens, AI Mode, Circle to Search, or Gemini in Chrome, and get a verification result.

C2PA Content Credentials, the industry standard for tracking how media was created and modified, are also being pushed deeper into Google’s product stack. Pixel 10 was the first smartphone to provide Content Credentials for camera images, and that capability now expands to video on Pixel 8, 9 and 10 in the coming weeks. C2PA verification arrives in the Gemini app today, with Search and Chrome integration following in the coming months.

The cross-industry partnerships are arguably more meaningful than the Google-only features. OpenAI, Kakao, and ElevenLabs are all integrating SynthID into their own AI-generated content. NVIDIA is partnering with Google to watermark AI-generated video from its Cosmos world foundation models. Meta, a fellow C2PA Steering Committee member, will start labelling camera-captured media with Content Credentials on Instagram, meaning authentic photos shot natively on Pixel phones will soon carry that authentication badge when shared there.

For enterprises, Google Cloud is also launching a new AI Content Detection API on the Gemini Enterprise Agent Platform, enabling businesses to spot AI-generated content from both Google and other popular models. Use cases include sorting feeds, preventing insurance fraud, fact-checking and labelling synthetic media. The API launches with a group of trusted partners and will iterate from there.

The honest read on this is that Google has shipped its side of the provenance problem reasonably well. Whether the broader open web actually adopts these standards depends on how quickly competing AI providers and major platforms outside Meta and Instagram fall into line.

Personal Intelligence goes global

A quieter but consequential rollout is Personal Intelligence in AI Mode, expanding to nearly 200 countries and territories across 98 languages, with no subscription required. Users can securely connect Gmail and Google Photos, with Google Calendar support coming soon, and Search uses that personal context to make answers more relevant. Connection is opt-in, and apps can be disconnected at any time.

That global rollout matters because much of what Google announced today, agents, Spark, and Omni access for subscribers, lands first in the US or for paid subscribers. Personal Intelligence is the rare I/O feature that arrives meaningfully wide on day one.

What this means

The I/O 2026 keynote is unusually coherent for a Google product event. The four announcements all point toward the same destination: an AI that takes action rather than just answering questions. Search becomes an agent platform. The Gemini app becomes a 24/7 personal worker. Omni turns text prompts into video, and SynthID and C2PA provide the verification rails that the new generative tools genuinely need.

Whether this lands with users as Google hopes depends on three things. First, whether the agentic features actually work reliably across the long tail of real-world tasks, since 95% reliability looks great in demos and breaks trust in production. Second, whether the partnerships behind Spark (Canva, OpenTable, Instacart and the rest) get genuinely useful rather than locked into narrow flows. And third, whether the content transparency story scales beyond Google and its current partners, since SynthID and C2PA are only meaningful when adversaries adopt them as well.

For users, the most immediately visible change is the reimagined Search box rolling out today. The longer-term shift, though, is toward AI agents that work on a person’s behalf. As Google frames it, this is the agentic era. We will know in roughly six months whether the framing was prescient or premature.