AEO & GEO

What is agentic AI? Meaning, examples and what it means for search

By Jon Earnshaw|03 Nov 2026|10 min read

Agentic AI is AI that can pursue a goal on its own. Instead of answering one question and waiting for the next, it plans the steps, uses tools and takes actions, checking in with a person when it needs approval. You give it an outcome ("find me a hotel under £200 near the center and book it") rather than a prompt, and it works out how to get there.

That shift, from answering to acting, is why every search team needs to understand it. When AI starts doing the searching, comparing and choosing for people, the way brands get found changes.

What does "agentic" mean?

"Agentic" comes from agency: the capacity to act. IBM describes it as these models' "capacity to act independently and purposefully."

So agentic AI is AI with a degree of independence. It isn't just generating words; it's deciding what to do next. Google Cloud's definition makes the same point: traditional AI mainly responds to commands or analyzes data, while agentic AI can set goals, plan and carry out tasks with little human input.

How agentic AI works

Most agentic AI systems follow the same loop:

  1. Goal. A person (or another system) sets an objective.
  2. Plan. The AI breaks the goal into smaller tasks.
  3. Act. It uses tools to carry out each task: searching the web, reading documents, querying a database, filling in a form or calling another service.
  4. Check. It looks at the results, adjusts the plan and repeats until the goal is met or it needs a human decision.

The tools are what make it agentic. On its own, a language model only produces text. Connect it to search, a calendar, a payment method or company data, and it can do things.

The Model Context Protocol (MCP), which I like to call the USB for AI, is one of the main ways agents connect securely to data and tools. It's why we built our own MCP server.

Most real systems keep a person in the loop for anything high stakes. Google says its Gemini Spark agent asks you first before it spends money or sends emails.

Agentic AI vs generative AI

Generative AI creates content. Agentic AI uses that ability, along with tools and planning, to get something done.

Generative AIAgentic AI
What it doesCreates text, images, code, audio or video from a promptWorks toward a goal over several steps
How it behavesResponds, then waits for the next instructionPlans, acts, checks and continues
Uses tools?Not necessarilyYes: search, apps, data, payments, other agents
OutputA piece of contentA completed task or decision (which may include content)
ExampleAsk a chatbot to write a product descriptionAsk an agent to find three suppliers, compare prices and draft an order

Put simply, generative AI mostly reacts, while agentic AI works toward a goal (IBM explains the difference well). An agent might write something along the way, but its job is to finish the task.

Is ChatGPT agentic AI? Mainly it's generative and conversational, but it has agent features that can browse, use apps and carry out multi-step tasks.

Agentic AI vs AI agents

What is an AI agent?

An AI agent is the individual system that does the work. In our glossary we describe it as a digital companion with built-in skills, like ChatGPT, Perplexity or Claude, that you can give tasks to and hand authority and responsibility. The simplest way I can put it: you delegate the task, shut the lid and walk away. To see what this means for brands, read our guide on how AI agents are changing how people buy.

Agentic AI is the broader capability. It often involves several agents working together, each handling part of the task.

In this short glossary video, we explain what an AI agent is.

Shopping is one of the first places you can see AI agents at work. Read our guide to AI shopping and agentic commerce to see how to prepare your products.

Examples of agentic AI you can see today

These have all launched, though availability varies by country.

  • Personal assistants that keep working. Google announced Gemini Spark in May 2026 as a personal AI agent that works around the clock. Because it runs in the cloud, it keeps working when you close your laptop. As of July 2026 it was available to paying subscribers in the US and many other markets, but not in the UK or the EEA.
  • Shopping that completes the purchase. Google's agentic checkout (US, from November 2025) can buy a tracked product on the merchant's site when it reaches your target price, after you confirm. In October 2026 it added Universal Cart in the US, which tracks prices across retailers and checks out with Google Pay. This is agentic commerce, the commercial side of agentic AI.
  • Booking inside search. In the UK, Google's AI Mode can help you find and book restaurant tables.
  • Brand assistants. Retailers are already testing business agents that answer product questions and guide purchases. Macy's has an assistant called Ask Macy's, and it tested a version of it in AI Mode. I tried it live in our July 2026 webinar with Lumar (from 15:32).

Gemini Spark changed how I think about timing. You give it a goal, shut the lid and it carries on in the cloud, then reports back. That's where search turns agentic: the search doesn't stop when you walk away.

We thought personal agents were a long way off. They're arriving much faster.

In this short video, I explain why Gemini Spark could change online shopping behavior.

Business agents and personal agents

In shopping and search, I find it helps to split agents into two kinds, because they're arriving at different speeds.

Business agents represent a brand. They let any customer chat with your brand inside Google Search or Gemini, using the answers and information you give them. Personal agents represent the person. They search, compare and eventually act on someone's behalf, and trust is what slows them down.

That difference matters for brands. A personal agent can talk to many brands at once, and if your business agent can't answer the shopper's question, it will simply move on to the next one. I cover both in more detail in what agentic commerce means for brands, customers and SEO teams, and you can see how products and sellers show up in AI shopping answers with our AI Shopping Tracker.

Watch me explain business agents and personal agents at brightonSEO (from 05:43).

Search is already agentic in places

For 25 years, search meant typing keywords, getting blue links and doing the hard work yourself. That's changing in stages:

  1. Search finds. Classic results pages rank pages for a query.
  2. Generative search synthesizes. AI Overviews pull information from several sources into one answer.
  3. Conversational search explores. AI Mode and ChatGPT let people refine a problem over several turns.
  4. Agentic systems evaluate and act. Agents compare options and, increasingly, take the next step for the user.

The first three are mainstream. The fourth is emerging and mostly US first, but our glossary already lists agentic as one of Google's four key tracks, alongside multimodal, visualization and personalization.

In this short glossary video, we explain multimodal search and Google's four key tracks, including agentic.

Google AI Mode already behaves a little like an agent. It takes a question, breaks it into parts with a query fan-out and pulls together an answer that becomes the start of a conversation.

I see ChatGPT and AI Mode as two competitors that both want to win the agentic race and build a walled garden. For more on how the two compare, watch our AI Overview vs AI Mode webinar.

What agentic AI means for search and brand visibility

When an agent does the research, the person may never visit your website. The agent reads your product data, reviews, policies and what other sites say about you, then decides.

As I wrote in From clicks to conversations, brands are no longer experienced. They're interpreted. So the question every team now has to answer is how their brand survives being interpreted by machines.

It's tempting to treat the zero-click side of agents as a traffic problem. I don't think it is. It's an identity problem.

When a machine chooses a brand on someone's behalf, the brand doesn't get to be the brand. The person never sees your images, your video or your story. All the machine has to go on is a set of attributes.

Three things change.

1. Visibility becomes a ladder, not a position.

A brand used to aim for page one. In an agentic journey it needs to be ranked, then mentioned, cited, recommended, evaluated and finally selected.

You can drop out at any step, and being eligible decides whether you take part in the conversation at all. Classic search still matters at the bottom of that ladder.

2. Conversations matter more than single queries.

Agents work through several turns before they decide. It's not really a prompt; it's a conversation.

Tracking one prompt tells you little; you need to know whether your brand stays present as the conversation narrows. (More in What prompts should I be tracking for LLMs?)

3. Your brand gets compressed into attributes.

Agents compare facts: price, specifications, availability, delivery, returns, reviews. Beautiful, wordy, flowery language gets compressed by AI and ignored by agents.

They want something concrete. What survives is specific, structured and consistent information, backed up by what the wider web says about you.

That wider web carries a lot of weight. In March 2026 I hosted a webinar with Sam Hailstone, former group head of SEO at Yoox Net-A-Porter Group, we studied US women's wear prompts in ChatGPT and AI Mode, using Pi data. Both platforms mostly cited editorial and informational sites rather than retailers.

As Sam put it, a brand's representation is "being mediated by what other people say about you rather than what you say about yourself." Your reputation increasingly becomes what the ecosystem agrees you are.

There's also a timing point. People are adopting agents for research faster than for buying.

Delegating search is easy. Delegating a decision takes time.

That gap gives brands a window to get their data and content in order before agents start making more of the decisions. Watch me explain the gap (from 19:54).

What search teams can do now

You don't need to build an agent to prepare for agentic AI. If I had to narrow it down, I'd strengthen three things: your data, your content and your trust signals. Start there.

In this short video, I share three things you can do today to prepare for agentic search.

  1. Measure where you appear in AI conversations today. Check whether your brand is mentioned, cited and recommended across ChatGPT, AI Mode, Gemini and Perplexity, and whether it stays present over several turns. Our AI Search Visibility tool is built for this.
  2. Tidy your structured data. Product feeds, Merchant Center attributes, schema, prices, stock and policies need to be complete and consistent. My first tip in the short above: take your top 50 products and get them into the best shape you can.
  3. Test your content the way an agent would. Build a GPT, link it only to your website, put the guardrails up, then push it hard and test it as if it were an agent. If it can't answer basic buying questions, neither can a real agent. The question to ask yourself is simple: does your website know enough to hold a conversation?
  4. Strengthen trust signals. Show who wrote your content and why they're qualified.
  5. Keep classic SEO strong. Rankings and SERP features are still the foundation that AI answers draw on.

If you want help putting this into practice, our team works with enterprise brands on agentic search optimization.

Want to see where this is heading? Join me and Peter Longton on December 2, 2026, at 3pm GMT for "Google's agentic Search future: will your business make the shortlist?" We'll look at how personal and business agents are developing and how to prepare your business to be discovered, trusted and chosen.

Related

View all

Never miss a post

Join our mailing list and have our SEO news delivered straight to your inbox.