Skip to content

Google I/O 2026: What Changes for Marketing, Brands, Commerce, and Ad Tech

Google I/O is a developer conference. Most of the keynote is chips, model benchmarks, and platform architecture that the average marketing team will never touch directly. But this year’s event contained a handful of announcements that change how products get discovered, how ads appear, how purchases happen, and how creative gets produced. Those are the announcements this article is about.

The short version: Google is rebuilding the path from curiosity to checkout around AI agents - systems that can monitor, compare, create, and act on behalf of users. For anyone working in performance marketing, e-commerce, brand strategy, or ad tech, this shifts the work from ranking pages and buying placements toward making brands legible to agents.

This article maps the landscape. In the coming days, we will publish individual deep dives into the most consequential topics - agentic commerce, AI Mode advertising, the creative production stack, and the developer harness - because each one deserves practitioner-level treatment.

Contents

The Model Layer Beneath Everything

Google skipped the expected “Gemini 4.0” headline and put the spotlight on Gemini 3.5 Flash: a faster, cheaper model that Google positioned close to frontier performance. It also introduced Gemini Omni, a model series for generating and editing video from multiple input types with stronger grounding in real-world physics.

For marketing teams, the API name matters less than the product consequence. Search, Ads, Shopping, YouTube, Gmail, and Workspace can absorb more AI when the underlying inference gets cheaper and faster. I/O showed exactly that pattern: model efficiency turning into more agentic product surfaces.

Gemini 3.5 Flash is already the default model behind AI Mode in Search globally. Google said AI Mode has passed one billion monthly users and that query volume has been more than doubling each quarter since launch. Those are Google-reported figures, but the trajectory matches the pressure search teams already feel: more complex questions, more conversational journeys, and fewer clean handoffs from query to click.

Read the model announcement as a cost-and-speed signal. Google is betting that practical agent tasks need fast, affordable intelligence more than maximum benchmark performance. That trade-off matters because it lets Google put agents into ordinary user flows rather than reserving them for premium demos.

Search Becomes a Multimodal Agent Surface

Google called this its biggest Search upgrade in over 25 years. The claim is aggressive, but the substance behind it is real. Three changes matter for marketing teams.

The Search Box Becomes Multimodal

The literal search box - the text field where billions of queries begin every day - now accepts images, PDFs, files, videos, and Chrome tabs as inputs alongside text. A user can photograph a dress, upload it, add a swatch of fabric from another tab, and ask Google to find a similar dress in the same colour for under £150. Search will return matches, then ask follow-up questions about size and occasion to narrow the results.

The rollout is live across countries and languages where AI Mode is available. For e-commerce brands, it means product photography quality, alt text, structured data, and product feed accuracy are no longer secondary optimisation targets - they are directly visible to the search interface itself. A poorly photographed product with thin metadata will not survive multimodal comparison queries where the user is literally showing Google what they want.

AI Mode Merges with AI Overviews

Google is unifying AI Overviews - the summary panels that appear atop traditional results - with AI Mode, the more immersive conversational experience. Previously, users had to choose between a traditional results page and an AI-forward experience. Now the two merge into a single flow, so a user can start with a standard query, receive an AI Overview, and seamlessly continue into a multi-turn conversation without switching modes.

For search marketers, this architectural change means the “blue link” results page is no longer the default destination for complex queries. Users who start in a traditional search may find themselves in a conversational journey before they ever see a ranked list. The content that earns visibility in this environment is content that AI systems can understand, summarise, and recommend with confidence.

Information Agents Monitor the Web on Your Behalf

Google introduced “information agents” in Search - personalised AI agents that users can set up to work in the background, 24/7, monitoring the web for whatever matters to them. A user could ask an information agent to track price drops on a specific product category, monitor a competitor’s blog for new content, watch for apartment listings matching specific criteria, or alert them when a sold-out item comes back in stock.

These agents look across blogs, news sites, social posts, real-time finance data, shopping feeds, and sports information. They are rolling out this summer for Google AI Pro and Ultra subscribers in the US.

For brands, these agents create a new layer between the customer and the message. The agent decides which price drop is significant, which restock deserves attention, and which competitor offer belongs in the comparison set. Clean feeds, clear pricing signals, accurate availability, and rich schema markup become visibility infrastructure.

The first release is gated behind paid Google AI subscriptions and starts in the US, so immediate reach is limited. The pattern carries the larger signal: users are beginning to delegate monitoring and filtering to AI systems. Brand urgency campaigns will increasingly meet an agent that has already compared the alternatives.

Agentic Commerce: Universal Cart, UCP, and AP2

This is the section that warrants its own deep-dive article (coming next in this series), but the overview is essential context.

Universal Cart

Google launched Universal Cart - an intelligent shopping cart that works across merchants and across Google services. You can add products while browsing Search, chatting with Gemini, watching YouTube, or reading Gmail. Once a product is in the cart, it runs in the background: finding deals, tracking price drops, surfacing price history, checking stock availability, and flagging compatibility issues. It runs on Gemini models, so it gets smarter as the models improve.

The cross-surface design is the critical detail. Universal Cart does not live inside a single store or a single Google product. It follows the user across the entire Google ecosystem, which means the shopping intent data it captures is richer than any single-channel signal. For advertisers, this creates a new surface where product visibility matters - inside search results, product listings, and the cart experience itself.

Universal Commerce Protocol (UCP)

UCP is the infrastructure layer underneath Universal Cart. It is an open standard co-developed with retail leaders (Etsy, Wayfair, Shopify, Target, Walmart have all been named) that standardises how AI surfaces interact with merchant checkout systems. Merchants remain the merchant of record - they keep customer data, relationships, and the post-purchase experience. UCP handles the discovery-to-checkout handoff.

Key expansion details from I/O: UCP-powered checkout is coming to Canada and Australia, with the UK planned later. UCP is also coming to YouTube in the US and expanding into hotel bookings and local food delivery.

The protocol is designed to be interoperable with Agent Payments Protocol (AP2), Agent2Agent (A2A), and Model Context Protocol (MCP) - which means it is built to work within Google’s ecosystem and with third-party AI agents and commerce systems.

Agent Payments Protocol (AP2)

AP2 lets AI agents make payments on behalf of users within defined limits. The user sets guardrails: specific brands and products they want, a spending cap, and criteria that must be met. When conditions align, the agent completes the purchase automatically.

AP2 creates a transparent, verifiable link between the user, the merchant, and the payment processor, with encryption protecting user data and tamper-proof digital records ensuring the agent always acts within the user’s defined boundaries.

For e-commerce and retail teams, the buying journey starts to compress. An AI agent can monitor, compare, wait for the right price, and complete a purchase within user-defined limits. That changes attribution models, weakens some retargeting loops, and forces a sharper question about brand differentiation when the buyer’s agent optimises for price, availability, delivery terms, and compatibility.

The infrastructure is ahead of everyday consumer behaviour. Google described AP2 as coming to its own products “in the coming months,” while routine autonomous purchasing remains early. The building blocks are moving faster than the operating practices around them. We will unpack the commerce layer in a dedicated deep dive.

Ads Move Into AI Mode

Ad inventory is moving into AI-generated experiences, and this is no longer speculative. Google is testing two ad formats inside AI Mode: Sponsored Stores (appearing inside product detail panels) and Direct Offers (embedding discounts directly into AI-generated responses). Both work through existing Shopping and Performance Max campaigns - there is no new campaign type to create.

The mechanics matter: if you are already running Shopping or PMax campaigns, you may already be appearing in AI Mode placements without realising it. What determines placement is asset quality, feed accuracy, and the contextual relevance of your products to the conversational thread. Google evaluates the entire multi-turn conversation, the full thread, which means intent signals are richer but also less predictable than keyword matching.

Early third-party analyses suggest AI Mode ads may generate higher engagement than traditional search placements while carrying higher CPCs. Treat those figures as directional until Google provides clearer reporting and independent benchmarks. The placement pattern already matters: advertisers may be entering AI Mode through existing Shopping and Performance Max campaigns before dedicated controls are fully mature.

Direct Offers are particularly interesting for e-commerce. They let brands share a tailored offer with a shopper who is close to purchase, without changing the offer for every user. Google has indicated these will expand beyond price into loyalty benefits and bundles.

For advertisers, AI Mode behaves like a layer across existing campaigns. Clean product feeds, strong creative assets, accurate inventory data, and contextually useful content become the entry requirements. Product feeds move from backend hygiene to front-line media infrastructure.

Reporting still lags the ambition. AI Mode placements can appear as top ads in Google Ads without a clean breakout, which makes isolation difficult. Google has signalled dedicated reporting, and media teams should start asking for it now.

The Creative Production Stack

Google’s creative tool announcements form a stack that, taken together, represents a meaningful shift in how campaign assets get produced.

Pomelli is a marketing agent built by Google Labs and DeepMind. Give it your website URL and it extracts your brand DNA - colours, fonts, tone, imagery - then generates social posts, ad creatives, and email banners that stay on-brand. Its Photoshoot feature, powered by Google’s Nano Banana image model, transforms a smartphone photo of a product into professional studio-quality and lifestyle imagery. Its Animate feature, powered by Veo 3.1, turns static marketing visuals into branded short-form video with a single click. A Catalog feature is now being tested, letting merchants point Pomelli at a product URL and have it ingest an entire inventory to generate batch creative. Pomelli is now available in over 170 countries.

Google Flow is an AI filmmaking tool integrating Veo, Imagen, and Gemini into a single creative environment. At I/O, Flow and Flow Music were expanded with Gemini Omni Flash, enabling conversational video creation and editing where users guide styles, characters, and scenes through dialogue rather than manual editing.

Stitch is Google’s AI design tool (comparable to Claude’s artifact design capabilities) that generates web and mobile UI designs from natural language prompts, using Nano Banana for image generation within the designs.

Nano Banana itself is Google’s image generation model that powers product photography across Pomelli and creative asset generation across the Ads ecosystem, including Google Ads Asset Studio.

SynthID and Content Credentials are the trust layer. SynthID has now watermarked over 10 billion AI-generated assets. Google is expanding SynthID detection into Search and Chrome so users can check whether content is AI-generated, and is adding C2PA Content Credentials support to verify whether content is an unaltered original from a camera or has been modified.

Creative teams should read the stack as a production multiplier. A product photo can become a studio-style image, a static asset can become short-form video, and a product catalogue can become a batch creative source. The work shifts from producing one polished asset toward managing adaptive asset systems across formats, intents, and surfaces.

The output quality will vary by category. Mid-market e-commerce gains the most where the alternative was limited or inconsistent product photography. Luxury, fashion, and premium categories still need a strong creative eye for lighting, material texture, composition, and brand nuance. These tools accelerate production; they do not remove taste from the process.

Antigravity 2.0 for Marketing Engineers

For ad-tech engineers and marketing technologists, Antigravity 2.0 is the most consequential developer announcement.

Google Antigravity is now an agent-first development platform with three surfaces: a standalone desktop application (Antigravity 2.0), a command-line interface (Antigravity CLI, which replaces the previous Gemini CLI), and an SDK for programmatic access to the agent harness. All three are optimised for Gemini 3.5 Flash and designed to orchestrate multiple agents running in parallel.

The keynote demo was deliberately dramatic: Antigravity 2.0 built the core framework of a working operating system in approximately 12 hours, spawning 93 parallel sub-agents and processing billions of tokens at a cost under $1,000. When the OS initially failed to run the game Doom because keyboard drivers were missing, they instructed Antigravity to generate the drivers in real time, and it succeeded.

Those figures are Google’s own claims from a controlled demo, not independently benchmarked. But the architecture they demonstrate - parallel sub-agents coordinated by a central harness, working autonomously on long-horizon tasks - is directly relevant to the kind of marketing agent systems we have been covering in this series. If you have read our earlier guides on agent anatomy and Claude Code architectural patterns, the Antigravity 2.0 architecture implements many of the same patterns: fork-join parallelism, dynamic sub-agents, scheduled background automation, and ecosystem integrations via MCP.

For engineering teams, Antigravity 2.0 is Google’s answer to the Claude Code / Codex pattern: a harness-first development environment where agents plan, split work, run in parallel, and produce artifacts. The Antigravity CLI replacing Gemini CLI nudges developers toward agent workflows, while the SDK opens the door to marketing-specific agents: campaign monitors, feed optimisers, creative QA agents, and analytics copilots. Google reported internal Antigravity usage at more than three trillion tokens per day, up from half a trillion in March.

Gemini Spark and Personal Agents

Gemini Spark is Google’s consumer-facing personal agent. Unlike Gemini’s existing chat interface, Spark takes actions on the user’s behalf - it runs on dedicated Google Cloud virtual machines, continues working with the laptop closed, integrates with Gmail, Docs, and other Workspace apps, and will support third-party tools via MCP over the summer.

The keynote demo showed Spark organising a block party by reading HOA rules from Drive, building a Sheets RSVP tracker, and chasing replies over Gmail - all autonomously. Spark is available to Google AI Ultra subscribers (now priced at $100/month for the standard tier).

Gemini Spark is the consumer-side counterpart to the commerce infrastructure. A personal agent that reads email, manages calendars, tracks shopping, and takes action changes the first reader of many marketing messages. A promotion may land in an agent’s context window before a human sees it. The agent weighs it against the user’s tasks, preferences, timing, and competing offers.

Marketing communications need to be machine-parseable, clearly valuable, and structurally clean enough for an AI agent to understand why they matter.

Practitioner Checklist

The announcements are dense, but the practical implications cluster around a few themes.

E-commerce teams: Audit your product feeds. Feed accuracy, structured data quality, pricing signals, inventory status, and product photography are no longer supporting details - they are the primary surfaces through which AI agents, Universal Cart, AI Mode ads, and information agents will discover and recommend (or ignore) your products. Investigate UCP adoption if you are a merchant of meaningful scale; the waitlist is open and the early movers will shape the standard.

Paid media teams: Check whether your Shopping and Performance Max campaigns are already appearing in AI Mode placements. You may be buying this inventory without knowing it. Push for asset quality improvements - feed accuracy and creative diversity now determine AI Mode visibility alongside bidding strategy. Prepare for attribution complexity: when a purchase happens through Universal Cart or AP2, the conversion path will not map cleanly to your current models.

Brand strategy teams: Start thinking about your brand’s visibility to AI agents as well as human audiences. Structured data, authoritative content, strong review signals, and clear product differentiation are the brand assets that agents can parse and trust. Your brand guidelines need a machine-readable layer: structured metadata that AI systems can extract and respect, alongside the visual standards your human designers already use.

Ad-tech teams: Read the UCP specification (it is open-source on GitHub). Understand how AP2, A2A, and MCP interact. Build for a world where ads appear inside conversations, carts span multiple merchants, and agents make purchase decisions within user-defined constraints. The auction model is shifting from keywords to intent moments - prepare your bidding infrastructure accordingly.

Creative teams: Learn Pomelli, Flow, and the Nano Banana ecosystem now, while adoption is low and the learning curve is gentle. Teams that master AI-assisted creative production will be able to generate far more asset variants for campaign testing than teams still relying on traditional production pipelines. The strategic skill is managing a library of adaptive assets across formats, surfaces, and intents.

Deep Dives Coming Next

This overview is deliberately wide. Each of the following topics will receive a dedicated deep-dive article over the coming days:

Agentic Commerce Deep Dive - Universal Cart, UCP, and AP2 unpacked for e-commerce operators and ad-tech engineers: protocol details, merchant integration, attribution, and open questions.

AI Mode Advertising - How ads work inside conversational search. Placement types, bidding mechanics, creative requirements, early performance data, and what the Sponsored Stores and Direct Offers formats mean for campaign strategy.

The Creative Production Stack - Pomelli, Nano Banana, Flow, Veo, and Stitch as a unified creative pipeline. Practical walkthroughs for marketing teams producing campaign assets at scale.

Antigravity 2.0 for Marketing Engineers - The agentic developer platform through the lens of marketing automation. How the harness pattern, parallel sub-agents, and MCP integrations connect to the agent architecture we have covered in previous guides.

Each deep dive will follow the same principle that has guided this series from the beginning: technical depth serves outcomes, and stays connected to commercial outcomes. Architecture enters the story when it explains a change in marketing, commerce, or business operations.

This article starts a Google I/O 2026 deep-dive sequence. Use the links below to separate the primary announcements from the surrounding AI News Hub context.

Primary Google sources

AI News Hub: translating frontier AI into actionable marketing playbooks.

Other Articles