If you’ve been watching how Google has been putting AI in the middle of search for the past few years, Google I/O this week confirmed what you’ve already heard: the union between search and AI is closer than ever. The three products present each point in a similar way, and although none of them represent a complete shift to AI mode (as some have feared), together they raise questions about measurement, site design, and customer journeys that SEO teams should be working on now.
Search agents
Knowledge agents are AI-powered agents that work behind the user’s Google account, around the clock, without having to ask for anything. The user sets a brief once (say, he wants a flat in a certain area, within a certain price range, with certain features) and the agent takes it from there.
It continuously monitors the web: news sites, blogs, social posts, real-time data on prices and availability. If something is a match, it sends them a consolidated review of what they found and what it recommends through the Google app. The agent observes, reasons, and informs. The user waits, and may not visit your site at all.
That means that a single short stop replaces many queries that previously would have produced separate, trackable signals over weeks or months. So search volume, as a measure of demand, will start to distort what is actually happening.
This makes impressions the most important metric to watch. If your content is always part of what an agent pulls from when compiling their updates on a topic, you’re reaching users in a meaningful way even if they don’t click. Clicks are still happening (at a lower rate, or in theory a more relevant rate), but for content that agents say, impressions are increasingly a signal that your content is being received.
This raises questions that don’t yet have clear answers: how do you measure any of this? Will there be a view in the Google search console that shows which URLs are cited in reviews compiled by the agent? If the notifications come through the Google app, will those impressions and clicks be tracked at all? Being defined in an agent-driven search environment is still something the industry needs to do.
This same concept (few active searches, common usage, residual measurement behind the product) applies directly to what Google is doing locally and commercially.
Agent booking
Booking agent takes the same basic concept (an AI agent working for you) and applies it to transactions. Instead of gathering information and sending you a summary, the agent makes the reservation. Finds a restaurant, event or service that matches your preferences, checks availability, and makes a reservation. For brands in any service category where a booking is a conversion, this raises an immediate question: can an agent actually complete a booking on your site? For many, the honest answer is no, and the reason comes down to how the sites are built.
Most sites are designed for patient human visitors. Most of the time, a customer can navigate a complicated process to make a purchase. Agents can’t. The agent reads your site the way a physical machine would: it looks at the raw code at the bottom of the page so if your booking form only appears after clicking a button and a piece of JavaScript is running, the agent might not see it. If important actions on your site rely on JavaScript to exist at all, they are invisible to the agent trying to complete the task.
Agent-friendly sites store important information and actions in raw HTML: prices, availability, booking steps, contact information. In addition to this, structured data, the machine-readable markup that tells Google what your page is about, needs to be accurate and complete, not a box-ticking exercise.
Beyond this, there is a broad set of guidelines in place that define what full agent readiness looks like. Web MCP (Model Context Protocol) means that your site can communicate its capabilities directly to the agent, rather than the agent having to navigate as the customer would. AP2 and APC are payment processes that allow agents to complete transactions on behalf of the customer, without needing to be present at checkout. A2A (Agent-to-Agent) allows agents across different services to provide one another. UCP connects all of these into one system.
Together, these agreements make it possible for the customer’s intent to be fulfilled from end to end, from acquisition to payment, without them touching a single page. Most sites don’t use these yet, but the rate at which they are being built, and the sites that are getting the foundation right now will be better placed when they do.
Cloudflare’s IsItAgentReady.com is a useful first place to diagnose where a site is currently failing against these criteria.
We are moving from suggestive AI to active AI. Nowhere is this more apparent than Google’s built-in shopping experience.
Universal Carriage
Universal Cart is Google’s new shopping cart. The idea is simple: instead of having a separate cart for each merchant website, you have one persistent cart that lives in your Google account and works everywhere. You can add a product while searching on Google, add another while watching a review on YouTube, add a third while reading email in Gmail, or while chatting with Gemini, and it’s all in the same place. Google then works in the background to check prices at all retailers, top deals, and let you know when the time to buy seems right.
For e-commerce products, Universal Cart changes what your competitors are doing. It’s no longer just about which page is ranking for which query but whether your product data is complete and readable enough for Google agents to find, compare, and display.
Essentially, that means availability, pricing, deals, and specific product attributes all need to be consistent and up-to-date across your site, in your curated data, and in Google Merchant Center.
Universal Cart is designed expressly for finding deals. That’s the core value proposition: Google finds you the best price without you having to go look. Sites that regularly show promotions, price drops, and offers will benefit from that, because deal signals become acquisition signals. The agent gets them because getting deals is a job. Premium or premium brands that compete on size rather than price are faced with a difficult question: how do they communicate the value that makes their product the right choice in the system, that is, at its core, optimized to present the cheapest option?
That question connects to the broader change the Global Cart represents in the way people shop. Like search information agents, the Universal Cart moves the user to a passive state. Tell Google what they want; now they are waiting to be informed when the right time comes. Also, the user searches less. They set a goal once, and then the system checks it.
When search volume is used by SEO teams to predict, and a reasonable portion of demand is now expressed as static agent commands instead of active queries, the search volume data is increasingly less than what actually exists. Do forecasting models need to shift to product-level signals, or some measure of topic-level demand, instead of keyword volumes? The forecasting and measurement framework will need to evolve.
Another dimension is personalization. Google’s human intelligence means that agents don’t just compare products and queries, they match them to individual consumers, based on context, preferences, and past behavior. That puts a premium on the specificity of your product data: not just whether you sell trainers, but which trainers, for specific uses, at what price point.
We’re still exploring how to effectively communicate that specification, but structured data and business development in your products are obvious places to start.
Bottom line
For SEO teams, the results from these three declarations reside in three places.
Rating: since most of the customer journey is through agents rather than direct search, impressions become a more important signal than clicks, and how the attribute works in that area is something that needs to evolve.
Site layout: agent readiness means accurate and complete structured data, key actions accessible in raw HTML rather than JavaScript, and understanding the protocol infrastructure that defines what an agent fully looks like.
Product data: Universal Cart doesn’t find pages, it finds products, which means the competitive landscape changes from page rankings to data quality. The consistency and clarity of your product information across your site, your structured data, and the Google Seller Center will determine whether an agent can find, compare, and display what you’re selling.
