Agentic Search Is Here: What Happens When Google Completes the Task Instead of Sending the Click

The Search Box Just Became an Agent

At I/O last month Google did the thing it has been threatening to do for three years: it replaced the search box. Not redesigned it. Replaced it. The new entry point is built on Gemini, and for a growing set of queries it does not return ten blue links at all. It returns an agent that acts. This is agentic search, and if you run a website, a store, or a marketing team, it changes the question you should be asking. The old question was "how do I rank." The next question is "how do I get the agent to choose me, and then to complete the task with me."

I have watched a lot of search shifts since the General Motors and Quicken Loans days. Most were weather. This one is climate. Let me give you the plain read, then a framework you can actually run against.

What "agentic" actually means

Strip away the keynote language and three concrete things happened.

• Information agents. Google now lets you spin up agents that go monitor the web and report back, rolling out first to its paid Pro and Ultra subscribers this summer in the US. You ask once, the agent watches continuously.
• Agentic booking and commerce. Google expanded in-results booking to local services, home repair, beauty, and pet care, and launched a Universal Cart so a shopper can move from advice to checkout without leaving Google. The brand stays the merchant of record, but the journey happens inside the surface.
• Generative UI. For some queries Google now assembles a custom interface on the fly: a comparison tool, a tracker, a dashboard. The page is no longer a list. It is software built around the question.

Put those together and you get a search experience where discovery, comparison, decision, and transaction can all happen before anyone touches your site. That is not an AI Overview summarizing your blog post. That is the agent doing the job.

This is not the same story as GEO

I want to be precise, because it is easy to lump this in with what we already know. Two years ago the conversation was about citation: how do you get quoted in an AI answer. I wrote about that in detail in Generative Engine Optimization and about the traffic math in AI Overviews and the new click economy. Those still matter.

But citation is about being referenced. Agentic search is about being chosen and transacted with. An agent that books a plumber or fills a cart is not looking for the most quotable paragraph. It is looking for structured, trustworthy, machine-actionable signals: availability, price, terms, fulfillment, reviews, an offer it can execute against. The unit of optimization moves from the answer to the action.

If GEO was about feeding the model your words, agentic search is about feeding the agent your operations.

The AGENT Framework for Agentic Search Readiness

Here is the checklist I am using with teams right now. I call it AGENT, because the point is to be legible to one.

A: Actionable data, not just content. The agent needs to do something. That requires structured, current, machine-readable facts: inventory, price, hours, service areas, booking slots, return terms. Your product feed and your structured data are now front-line marketing assets, not back-office plumbing. If you want the deeper case for treating schema as infrastructure, I made it in Schema markup: the translation layer for machines.

G: Grounded identity. Agents disambiguate before they act. They need to know who you are, what you sell, where you operate, and that you are real. This is the entity work, not the keyword work. A clean, consistent entity footprint across your site, your knowledge panel, and third-party sources is what lets an agent confidently pick you. See entity-based SEO for the mechanics.

E: Executable offers. Agentic commerce runs on protocols. There is a real standards layer forming underneath all of this: Google's coalition built around a Universal Commerce Protocol, OpenAI and Stripe pushing in-chat checkout, and payment-authorization standards on top. You do not need to memorize the acronyms. You need to ask one question internally: can a trusted agent retrieve our current price and availability and complete a purchase or booking without a human re-typing it? If you are on a major commerce platform, much of this is being built for you. If you are not, this is a roadmap conversation for Q3.

N: Numbers you can defend. When the click disappears into the agent, your old dashboard lies to you. Sessions fall, but revenue may not. You need measurement that survives the zero-click reality. I will not relitigate the whole argument here; I made it in measuring SEO when the clicks fall. The short version: instrument assisted and agent-mediated conversions, watch branded demand, and stop reporting traffic as if it were the goal.

T: Trust the agent can verify. Agents lean hard on signals they can check: review sentiment, ratings, fulfillment reliability, return history, merchant-of-record status. Reputation is no longer soft. It is an input the machine reads and weights. The brands that win agentic placement will be the ones whose operational trust is visible and verifiable, not just asserted in marketing copy.

Run your business against those five. The gaps you find are your next two quarters of work.

What I would do this quarter

Concrete, in priority order:

• Audit your feeds and structured data first. Before you write another blog post, make sure the facts an agent needs are present, current, and correctly marked up. Stale price or availability is now a lost transaction, not a cosmetic error.
• Pick the queries where the agent will transact. For most businesses it is bookings, local services, and product purchases. Those are the surfaces moving first. Map which of your money queries are most exposed and prioritize there.
• Fix entity consistency. Reconcile your name, categories, locations, and descriptions everywhere they appear. Ambiguity is a reason for an agent to skip you.
• Rebuild the report. If your executive dashboard still leads with organic sessions, it is about to be wrong in a way that gets budgets cut for the wrong reasons. Get ahead of it.
• Protect the relationship. When Google or an assistant becomes the interface, your direct line to the customer thins out. First-party data is how you keep one. I walked through this in first-party data and the post-cookie playbook, and it matters more now, not less.

The honest read

I am not going to tell you the website is dead. I have heard that prediction every two years for a decade and it has never been true. The site is still where trust is built, where the brand lives, where the agent verifies you. What is changing is that the site is no longer guaranteed to be where the transaction happens. For a real and growing slice of demand, the agent will decide and act on a surface you do not own.

That is uncomfortable, and it is also clarifying. It rewards the boring disciplines I have argued for all along: clean data, real entity authority, verifiable trust, honest measurement. The teams that treated those as optional are about to find out they were load-bearing.

Numbers over noise. The agent is not impressed by your clever headline. It wants facts it can act on. Give it those, and you stay in the consideration set even when the click never comes.

Keep reading: The Search Console Generative AI Report Is Live: What It Measures, What It Hides, and How to Use It.

If your team is staring at falling sessions and trying to figure out whether agentic search is a threat or an opening, that is exactly the conversation I like having. The channel here is open by introduction. Bring your data, and we will read it together.