AEO & GEO

What Is Agentic Commerce? A Guide for Retailers

By Jon Earnshaw|01 Nov 2026|10 min read

AI shopping is when people use AI to find, compare and buy products. Agentic commerce takes it further: AI agents do the work on a person's behalf, from finding products and checking the details to, increasingly, completing the purchase. Instead of a human searching, clicking and buying on a website, a customer's personal agent talks to a brand's business agent, and the decision can happen inside a conversation.

Transactions are moving away from websites and into conversations across AI surfaces. That's no longer a prediction. The infrastructure exists, and in the US people can already buy inside Google's AI Mode and Gemini.

At Pi we've been predicting the emergence of conversational search for many years. Now AI agents are working their way into those conversations, and every online business needs to prepare.

In this guide I'll cover what agentic commerce and AI shopping agents are, what they mean for SEO (especially ecommerce SEO), how to prepare your product data for Google's AI Mode, and how to see whether AI recommends your products and your store. At the end, I round up where the main platforms stand today.

What is Agentic Commerce?

For as long as we've had search engines, humans have searched, clicked blue links, made decisions and bought things. Agentic commerce changes who does that work. Autonomous AI agents handle the interactions, and a consumer's own personal agent (Google's Gemini Spark is an early example) can communicate directly with a brand's business agent to have the conversation and complete the purchase.

Watch the short explainer below:

As I put it in that short, business agents are going to be the brand's trusted voice, the voice of reason. And when personal agents start talking to them, that's where the transaction will happen: inside that walled garden, potentially without even a visit to a website.

What does agentic commerce look like for a retailer?

Imagine a shopper asks an AI assistant to find waterproof walking boots under ÂŁ150 for a weekend hike. The assistant narrows the options, compares prices and stock, and suggests where to buy. If the platform supports checkout, the shopper may also be able to authorize a purchase.

Your product could make the shortlist while another retailer gets the recommendation. That gives retail teams two questions to measure. Are our products part of the conversation? And when the shopper is ready to buy, is our store the one AI recommends?

What is an AI shopping agent?

An AI shopping agent is an AI tool that shops on someone's behalf. You tell it what you need, and it searches for products, compares options, answers follow-up questions and, where the platform allows it, completes the purchase.

AI shopping agents and agentic commerce are related, but they're not the same thing. The AI shopping agent is the tool that does the browsing and buying. Agentic commerce is the wider shift: the protocols, platforms, payment rails and business agents that let those tools find products and complete purchases.

You can already see AI shopping agents at work, mostly in the US:

  • Google's agentic checkout can buy a tracked product on the merchant's own site when it hits a target price, once the shopper confirms.
  • Amazon's Buy for Me can complete a purchase from another retailer's site for eligible products.
  • Perplexity's Instant Buy lets US users buy without leaving Perplexity, with the retailer as merchant of record.
  • Google's Gemini Spark is a personal agent that keeps working in the cloud after you close your laptop. It isn't available in the UK yet.

An AI shopping agent is one kind of AI agent, which is AI that can take steps toward a goal instead of just answering a question. For the wider picture, read What is agentic AI?


From pages to protocols to agents

The big shift is architectural. In our February webinar I called it the big collapse. Google had the discovery, retailers owned the evaluation and the payment providers owned the checkout, but now it’s all collapsed into one conversational surface.

Yesterday’s stack:

  • Website (CMS + checkout)
  • Analytics
  • Organic search (plus ads) driving sessions to pages

The stack that’s now being built:

  • Open protocols for agentic commerce (ACP, UCP) that define the “language” between agents, merchants and payment providers
  • AI surfaces (Google AI Mode, Gemini, ChatGPT etc.) that host the conversations
  • Business agents that represent your brand on these surfaces
  • Personal agents that represent the customer, orchestrating what to buy, from whom, and under what constraints

Websites remain important, but they’re no longer the only place where buying decisions happen. The two kinds of agent in that stack, business agents and personal agents, are where brands need to focus first.

Ecommerce agents: business agents and personal agents

An ecommerce agent is an AI agent that takes on shopping tasks. Pi’s glossary describes an AI agent as a digital companion with built-in skills, like ChatGPT, Perplexity or Claude, that you can give tasks to and hand authority and responsibility. In commerce there are two kinds, and they arrive at different speeds.

Business agents: your new digital storefront. Business agents act as branded AI staff on surfaces such as Google Search, Gemini and ChatGPT. They draw on your Merchant Center data, website content and commerce protocols to answer questions, guide purchases and drive checkout around the clock. They can advise on products and compare options, guide configuration, bundles and financing, and support post-purchase needs like returns, warranty extensions and reorders.

Put simply, as I said in my February webinar, a business agent is nothing more than a conversational experience inside Google Search and Gemini that lets any customer chat with your brand.

Some are already out in the wild. In our July webinar I showed Macy’s assistant, Ask Macy’s. It was so good that people didn’t just like it, they spent more: according to Google Cloud, revenue per visit was roughly 4.75 times higher among customers who used Ask Macy’s during beta testing. For a brand, the model is simple. You claim your agent in Merchant Center, tell it who it is, and link it to the right information through an MCP layer. Watch me walk through it at 15:04.

Personal agents represent the buyer: storing preferences, payment details and recurring tasks, and operating across AI platforms and devices while talking to any compatible business agent through open protocols. They are taking longer, and for a good reason. As I said in the same webinar, delegating search is easy, but delegating a decision takes time. People are happy to let something narrow down options for them. Handing over their money, their taste and their style is a different matter.

When personal agents do arrive, they won’t shop one brand at a time. As I put it in February, your personal agent isn’t having a conversation with one business agent, it’s having it with several. If your business agent can’t answer a question about, say, compatible accessories, the personal agent will switch to the next one.

Strategic implications for businesses

  1. Compressed buying journey: discovery, evaluation, and checkout can happen in one conversation.
  2. Shifted control: you do not control the surface where decisions are made, but you remain the merchant of record.
  3. Execution matters: stock accuracy, fulfillment, and support quality become key signals, because agents optimize for reliability and trust.

The agentic commerce protocols: ACP, UCP and MCP

What it isWho is behind itWhere it shows up today
ACP (Agentic Commerce Protocol)Open standard for how AI agents and businesses exchange product feeds, carts, checkout and delegated paymentsMaintained by OpenAI and Stripe (beta)Product discovery in ChatGPT; checkout through merchants’ own sites and ChatGPT apps
UCP (Universal Commerce Protocol)Open standard for the full shopping journey across agents, businesses and payment providersGoogle, co-developed with Shopify, Etsy, Wayfair, Target and WalmartBuy button and Universal Cart in AI Mode and Gemini (US first)
MCP (Model Context Protocol)A way for an AI agent to connect securely to a data source or toolOpen standard used across the industryBehind business agents, and in tools like Pi’s own MCP server

Underpinning both commerce protocols is MCP: a standardized way to expose structured data, like product details and return policies, to AI agents securely, instead of relying on fragile scraping. Our glossary calls it a USB for AI. Google’s Universal Commerce Protocol is designed to work with MCP, and that’s where a business agent can pull detailed product information or check whether someone is already a customer. For more on how MCP works, read What is an MCP?

What agentic commerce means for SEO (especially ecommerce)

SEO was never just about keywords. It was about showing up, being understood, and getting chosen when someone wants to get something done. Agentic commerce doesn’t replace SEO; it changes what SEO is for.

In AI-driven shopping, agents handle multi-step tasks and clarify ambiguous intent. Customers no longer search for “best running shoes for flat feet.” Instead, agents guide them through conversations, narrowing options, checking stock, and even completing purchases. This is also why questions such as “what prompts should I be tracking for LLMs?” are becoming increasingly difficult to answer, as visibility is shaped by entire conversations rather than individual prompts.

For SEO, the goal shifts from ranking pages to making your brand:

  • Eligible for the agent’s candidate set
  • Understood with accurate, structured data
  • Trusted with strong reviews, clear policies, and reliable operations
  • Chosen through compelling offers and effective agent performance

New ranking factors include:

  1. Protocol and feed eligibility: implement ACP or UCP, keep product feeds complete, consistent, and up to date.
  2. Structured knowledge and policies: make shipping, returns, warranties, sizing, and ingredients machine-readable.
  3. Trust and performance signals: reviews, customer service metrics, stock accuracy, and fulfillment speed.
  4. Agent performance: how often your Business Agent converts conversations to orders, handles questions, and resolves issues.

SEO’s role expands from page optimization to commerce graph optimization across AI surfaces, including web search, AI assistants like ChatGPT and Gemini, and marketplaces such as Amazon and Walmart. In my own planning notes I put it this way: “SEO will move upstream from getting content ranked to making business truth legible to machines.”

Practically, SEOs become schema and protocol leads, conversation strategists and measurement translators, shaping how agents interact with your brand and reporting on AI-driven visibility and conversions. You can already measure part of this, and I explain how in the AI Shopping Tracker section below.

The businesses that get this right will move beyond the old “SEO versus PPC” debate and instead focus on eligibility and performance across all AI surfaces.

SEO shift to agentic commerce ranking factors

Google Merchant Center conversational attributes: prepare your products for AI Mode

Whatever the checkout timeline in your market, the product data AI reads is something you can improve now. Google has given retailers a direct way to do it.

Conversational attributes are optional fields in the Google Merchant Center product data specification. Google says they help AI systems and conversational agents understand the specific details of your products, so shoppers can find information about them across AI surfaces like AI Mode in Search, as well as in traditional search (Google Merchant Center Help).

Why they matter for AI shopping

I covered this in our 3 November webinar. If your products are also sold by other retailers, you need a way to stand out. And when personal agents talk to business agents, your brand gets reduced to attributes. These attributes help you supply the detail that wins those conversations.

Shopping with AI isn’t one question and one answer. People describe their problem and then go back and forth over several turns. Say a shopper looking for a backpack mentions that the zip on their last one leaked after a couple of hours and asks which bag offers better protection. That’s the kind of detail you can put into these attributes, and variant options help match the right variant to that shopper.

My advice in the webinar was simple: “I would be picking my best products right now, today, before the end of this week, and start to complete those out.” As I said, “You don’t have to do a lot, you just have to follow these attributes.”

Watch this part of the webinar from 40:29:

The six conversational attributes

These are the attributes Google lists, with Google’s own example values:

AttributeWhat it’s forExample
Question and answer [question_and_answer]Questions and answers, or FAQs, about a product“Does it have a headphone jack?”:”This version doesn’t have a headphone jack.”
Document link [document_link]Links to related PDF documents, such as a manual or assembly instructions. Separate more than one with commas.https://example.com/manual.pdf
Related product [related_product]Related products, such as accessories or required parts, given as relationship type, identifier type and identifierrequired_part:id:AZ7B, accessory:gtin:811571013579
Item group title [item_group_title]A title for a product that comes in several variants, used with item group ID [item_group_id]Organic Cotton Men’s T-Shirt
Variant option [variant_option]All the properties that identify a variant, given as name and value, used with item group title and item group IDShoe width:narrow,size:8
Popularity rank [popularity_rank]How popular the product is, as a percentage of your total inventory. The higher the value, the better it performs compared with your other products.95.5

How to submit them

Google gives you three options:

  • A supplemental data source. This is the option Google recommends, and it keeps the new fields separate from your main feed.
  • Your primary data source, alongside your existing attributes.
  • The Merchant API.

Google says adding these attributes won’t affect the approval status of your existing products. If you already cover a detail in the description, product highlight or product detail attributes, you don’t need to repeat it. The help page also links to a TSV template you can work from.

How to prepare your product data for AI Mode and AI shopping agents

  1. Start with your best products. Pick a short list, such as your top sellers or the products where competitors get recommended instead of you.
  2. Use real customer questions. Look at support tickets, live chat, reviews and on-site search. Write short, factual answers for question and answer.
  3. Fill gaps, don’t repeat. Check what’s already in your description, product highlight and product detail attributes, and use the new fields for what’s missing.
  4. Link the documents shoppers need. Add manuals, size guides or assembly instructions with document link.
  5. Connect accessories and parts. Use related product to point to accessories and required parts by ID or GTIN.
  6. Sort out your variants. Give each product group an item group title, and use variant option to say what makes each variant different, such as size, color or fit.
  7. Add popularity rank if you can work it out reliably from your own sales data.
  8. Submit and check. Upload through a supplemental data source, check Merchant Center for any issues, then roll out to more products.
  9. Measure the result. Track whether your products appear and whether your store is recommended in AI shopping answers, before and after.


What ecommerce businesses need to do now

In this short, I outline the three things to strengthen first: your data, your content and your trust signals. The five steps below show how to put that into practice.

1. Get on the rails

Get on the rails

  • Implement ACP (for ChatGPT) and UCP (for Google AI Mode/Gemini) early.
  • Make sure your Merchant Center data is complete, accurate, and policy-compliant, and add conversational attributes to your top products (see above).
  • Consolidate product and content sources into a single, canonical catalog and a single source of truth for policies, FAQs, sizing, and instructions.

2. Make your business agent-ready

Make business agent-ready

  • Audit data quality: attributes, titles, SKUs, GTINs, and legal flags must be consistent and complete.
  • Audit policy clarity: AI agents must be able to answer questions about returns, warranties, safety, and other policies without ambiguity.
  • Audit operational reliability: stock feeds, fulfillment estimates, and delivery times must be accurate. Agents bypass messy or unreliable vendors.

3. Build a first-generation Business Agent

Build business agent

  • Define the agent’s tasks, pre-purchase guidance, discount rules, and hand-off points to humans.
  • Prototype internally using your existing knowledge base and product data.
  • Capture gaps and errors to improve your content and data.
  • Decide the agent’s brand voice and style for recommendations, upsells, and handling complaints.

4. Re-tool SEO for agentic commerce

Re-tool SEO

  • Participate in ACP/UCP implementation and own taxonomy and schema.
  • Collect and analyze conversation data from AI surfaces as a new source of insights.
  • Track visibility, agent-assisted conversions, and gaps where competitors are recommended, for example with Pi’s AI Shopping Tracker.
  • Ensure website content is structured for easy extraction by agents.

5. Scenario-plan for the future

Scenario planning

  • Consider what happens if a significant portion of sales shifts to agentic surfaces.
  • Assess risks like defaulting to cheapest options and opportunities like machine-visible sustainability or superior policies.
  • Use two or three scenarios to stress-test your roadmap. Agentic commerce changes who controls discovery, decision, and transactions.

How to prepare for AI shopping: a checklist

Here is everything above in one list. You don’t need to do it all at once. Start with your best products and build from there.

Product data

  • Keep your Merchant Center feed complete and accurate: titles, GTINs, prices, stock and policies.
  • Add conversational attributes to your top products, starting with real customer questions.
  • Group variants with item group title and variant option.
  • Keep one source of truth for product details, policies, FAQs and sizing.

Policies and trust

  • Make shipping, returns and warranty terms clear and machine-readable, for example with MerchantReturnPolicy structured data.
  • Back up claims like “sustainable” with structured data, not just page copy.
  • Keep stock levels, delivery estimates and fulfillment accurate.
  • Make it easy for customers to leave reviews, and show them.

Platforms and protocols

  • Check which AI shopping platforms matter in your markets, and whether checkout is live there yet.
  • Get on the rails that fit: UCP for Google AI Mode and Gemini, ACP for ChatGPT.
  • Check that your payment provider supports agent payments.
  • Consider an MCP server so agents can reach accurate product and policy data.
  • If you’re eligible, set up Business Agent in Merchant Center (US only for now).

Measurement

The brands that do this first won’t just survive the shift from websites to agents. They’ll be the ones the agents keep picking when it matters, at the moment of decision.

Track how AI shopping agents see your products

Start by measuring where you show up today. Pi’s AI Shopping Tracker shows which products appear in AI shopping results, which websites are cited, and which sellers are recommended. Look at those signals together to see where you’re already visible and where shoppers are being directed to competitors.

Preparing your product data is half the job. The other half is knowing whether it works. When shoppers ask AI what to buy and where to buy it, do your products appear, and are you the recommended seller?

A brand mention, a citation to your website, a product listing and a seller recommendation each play a different role in a shopping conversation. Your product might appear while another retailer is recommended as the seller. Or your website might be cited without your product making the shortlist.

Pi’s AI Shopping Tracker shows you the difference. With it you can see:

  • Which of your products appear in AI shopping results, and where competitors take your place
  • Where your website is linked, and which sellers AI recommends buying from
  • Which retailers compete for the recommendation, and how your prices compare with those shown in AI results
  • The shopping questions that bring products into the conversation, by product, brand, category and query
  • How all of this changes over time

It works for retailers and brands. Retailers can see whether they’re recommended as the seller for products they stock. Brands can see whether their products appear and which retailers AI recommends for them.

My colleague Peter Longton explained why the seller recommendation matters in his brightonSEO talk this October: if a purchase journey happens without human involvement, the recommended seller is likely to be the one the agent makes the deal with. And to see how your brand performs across whole conversations, not single answers, Pi’s AI Authority Score brings it into one benchmark against your competitors.

Once you can see where you stand, our Agentic AI Services can help you strengthen the content, product information and trust signals that support AI-assisted decisions.


Demo the AI Shopping Tracker

Where agentic commerce stands in late 2026

Here’s what has actually launched, and where.

OpenAI and the Agentic Commerce Protocol (ACP). OpenAI and Stripe launched Instant Checkout in ChatGPT, powered by ACP, in September 2025. In March 2026 OpenAI changed course. It said the first version of Instant Checkout didn’t offer the flexibility it wanted to provide, so merchants now use their own checkout while ChatGPT focuses on product discovery. ACP lives on as the layer merchants use to share product feeds and promotions with ChatGPT. As of October 2026, OpenAI’s merchant page says shopping is live for ChatGPT users in the US and purchases are completed on merchants’ own websites or apps.

Google and the Universal Commerce Protocol (UCP). Google launched UCP on 11 January 2026 at the National Retail Federation show, co-developed with Shopify, Etsy, Wayfair, Target and Walmart. It covers the whole journey, from discovery and buying to post-purchase support. It powers a buy button on eligible product listings in AI Mode and Gemini, and the retailer stays the merchant of record. In September 2026 Google said UCP already enables direct checkout for hundreds of thousands of brands and retailers, and added a UCP integration hub in Merchant Center with cart transfer to the merchant’s site and checkout testing. It is rolling out in the US first, with Australia and Canada to follow early in 2027.

Google’s agentic checkout and Universal Cart. Google’s agentic checkout (rolled out in the US in November 2025) can buy a tracked product on the merchant’s own site, with the shopper’s confirmation, when it hits a target price. On 15 October 2026 Google launched Universal Cart in the US: one cart across Search and the Gemini app that tracks price drops and restocks and checks out with Google Pay or hands the cart over to the retailer’s site. Google named Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair and many Shopify merchants among the first stores.

Google’s Business Agent. Google launched Business Agent on Search on 12 January 2026 with retailers including Lowe’s, Michael’s, Poshmark and Reebok. It’s the branded agent I describe above: eligible US retailers switch it on and customize it in Merchant Center (Shopify stores can activate it through the Google & YouTube app). In September 2026 Google opened a beta that puts Business Agent inside YouTube ads in the US, and made AI performance insights in Merchant Center generally available in the US, Canada, Australia, New Zealand and India, so retailers can see their share of voice in AI Mode and AI Overviews.

Amazon, Perplexity and Shopify. Google isn’t the only one. In May 2026 Amazon combined Rufus and Alexa+ into Alexa for Shopping for US customers; its Buy for Me feature can complete a purchase from another retailer’s site on the shopper’s behalf for eligible products. Perplexity and PayPal launched Instant Buy for US users in November 2025, with the retailer as merchant of record. And Shopify’s agentic storefronts make eligible stores’ products available in ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta, with checkout inside the AI channel on some of them.

Personal agents. Google’s Gemini Spark, announced on 19 May 2026, is a personal agent that keeps working in the cloud after you close your laptop. As of late October 2026, Google’s help pages still exclude the UK, the European Economic Area, Switzerland and Nigeria.

What this means if you sell in the UK. Most of the checkout features above are US first. Google’s Merchant Center help says UCP-powered checkout applies to products eligible in the United States, Canada and Australia, and that it’s only available for select merchants for now. Google’s own roadmap names Australia and Canada next, for early 2027, and gives no UK date. Business Agent is US only, and AI performance insights in Merchant Center isn’t yet offered in the UK. Payments are moving faster: in March 2026 Visa launched its Agentic Ready program in Europe, including the UK, with issuers such as Barclays, HSBC UK, Nationwide, Revolut and Santander testing agent-initiated payments, and Mastercard has offered Agent Pay since April 2025. So for UK teams the immediate job is visibility and data readiness, not plugging in a buy button.

Plan for several platforms, not one. When I went through this with Sam Hailstone in our March webinar, he called it “a protocol land grab.” He also warned against preparing for agentic commerce as if it were a single shift, because it isn’t one. Each platform has its own rails, so for most teams it’s a question of prioritization and, crucially, resourcing. Payments are part of it. If your payment provider isn’t on the right rails, Sam said, agents can recommend you but can’t complete the transaction. Some platform details from that session have since changed (OpenAI’s checkout, for one). The point stands.

AI agents enabling commerce tasks across multiple platforms

The future of ecommerce: from websites to agents

As transactions move from websites to conversations, and from conversations to background negotiations between agents, the businesses that thrive will be the ones that build the most robust, transparent and standardized data infrastructure. The website stops being the only destination and becomes the publication layer for the data that feeds the agents.

The timeline has moved more slowly than the January headlines suggested. Checkout inside AI is live in the US (in Google, Perplexity and Amazon) but limited to eligible merchants and products, OpenAI has stepped back from native checkout to focus on discovery, and personal agents like Gemini Spark aren’t yet available in the UK. But the rails are in place, and the brands preparing their data, policies and content now are the ones agents will be able to evaluate when it counts.

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.


At Pi, we provide expert strategic guidance on agentic commerce and how to get the most from the Pi platform. Our team works with brands on agentic search optimization, helping businesses prepare their data, protocols and AI visibility so agents can discover, understand and recommend their products.

Related

View all

Never miss a post

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