# Agentic commerce: what it means and how stores get ready

> Source: https://rankxai.com/blog/agentic-commerce-guide · Last updated: 2026-11-13

Agentic commerce is shopping where an AI agent does part of the buying for a person: it finds products, compares them and can start or complete checkout once the shopper approves. For online stores, it means accurate product data, crawlable product and policy pages, and checking whether ChatGPT and Google AI Mode recommend you.

## What is agentic commerce?

Agentic commerce is buying and selling in which an AI agent acts for the shopper: it reads a request, searches for products, compares options and carries the purchase towards checkout, with the person approving the spend. The short definition is in our [glossary entry on agentic commerce](/glossary/agentic-commerce); this article covers what it asks of an online store.

### What makes shopping agentic

Agentic describes software that takes a sequence of steps towards a goal. An AI assistant that lists five kettles is AI-assisted search. An agent that checks stock, applies the shopper's budget and prepares the order is agentic shopping.

Many live features sit between the two. The AI assistant shortlists products and explains the choice, and the shopper clicks through to buy on the merchant's website.

### Who is involved

Agentic AI shopping usually involves four parties:

- **The shopper**, who sets the goal and approves the spend.
- **The AI assistant or search surface**, such as ChatGPT or Google AI Mode, which acts as the agent.
- **The merchant**, whose product data, policies and checkout the agent uses.
- **The payment provider**, which authorises the payment.

An agent acting on delegated authority has to show what it is allowed to buy, and for whom. That is why identity and authorisation come up alongside product discovery in the commerce protocols covered below.

To see where your products stand before changing anything, the section on telling whether AI agents recommend your products shows a manual method you can run today. It also shows how the [AI Shopping module](/features/ai-shopping), opt-in and switched on per website, checks the same prompts on a schedule.

## How agentic commerce works, from a shopper's request to checkout

Agentic commerce works in five stages: the shopper states a need, the AI agent gathers candidate products, compares them against the request, presents a shortlist for approval and hands off to checkout, either inside the assistant or on the merchant's own website. Each stage reads a different part of your store's data, so each is a place to win or lose the sale.

### 1. The request

A shopper writes a prompt such as "a quiet burr grinder for espresso under $200, delivered this week". The agent turns it into constraints: product type, use, price ceiling and delivery date. It may run several searches behind the scenes to cover each one.

### 2. Discovery

The agent gathers candidates from product feeds that merchants submit, from structured data on product pages and from the text of the pages themselves. Reviews, comparison articles and forum threads can shape which products it trusts. A product missing from every source cannot be shortlisted.

### 3. Comparison

The agent filters on hard facts first: price, availability, delivery time, returns and specifications. Then it weighs softer signals, such as how well the description matches the use the shopper described. Gaps or contradictions between your page and your feed give the agent less to work with.

### 4. Approval

The shopper reviews a shortlist, asks follow-up questions and confirms a choice. In fully agentic flows, this is also where the shopper authorises payment, and the payment provider confirms the agent is allowed to spend on their behalf.

### 5. Checkout

Checkout either happens inside the assistant through a commerce protocol or finishes on the merchant's own website or app. Three open standards matter here:

- **Agentic Commerce Protocol (ACP)**: [Stripe describes it](https://stripe.com/blog/developing-an-open-standard-for-agentic-commerce) as an open standard, developed with OpenAI, that lets AI agents and businesses complete purchases together. In Stripe's description, the business stays the merchant of record. In ChatGPT today, [OpenAI's merchants page](https://chatgpt.com/merchants/) says ACP "supports product discovery and more accurate, personalized recommendations through shared product data", that "purchases are completed on merchant-owned websites or apps" and that "Shopping is currently live in the U.S."
- **Universal Commerce Protocol (UCP)**: [Shopify's UCP page](https://www.shopify.com/ucp) describes it as "Co-developed with Google, open to everyone", and [Google describes it](https://developers.google.com/merchant/ucp) as an open standard for agentic commerce.
- **Agent Payments Protocol (AP2)**: [Google's announcement](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol) describes it as an open protocol for agent payment authorisation, covering how to prove "that a user gave an agent the specific authority to make a particular purchase."

A protocol only matters to you if the AI assistants your buyers use support it and your ecommerce platform or payment provider has implemented it. Ask your provider before you build anything.

## What do shoppers use AI agents for?

Shoppers use AI agents for purchases where the request can be stated up front and checked against product data. Each use case below leans on a different part of your store's information, so it shows where gaps cost you the sale.

| Use case | What the shopper asks for | What the agent needs from your store |
| --- | --- | --- |
| Routine reordering | The same product again, or the current version of it | Stable product identifiers (GTIN, MPN), live stock and price |
| Subscriptions | A refill on a set schedule | Clear subscription options, prices per delivery and cancellation terms |
| Budget-bound gifts and bundles | Several items that fit one total budget | Accurate prices, variants, delivery dates and gift options |
| Trade and B2B buying | Parts or supplies that meet a specification | Full technical attributes, pack sizes and trade terms |

These examples are illustrative, not a forecast of demand. Start with the use case closest to how your customers already buy, and check that the fields in its row are complete on the page, in the feed and in your structured data.

## Where AI shopping agents meet shoppers today: ChatGPT, Google AI Mode and others

AI shopping agents meet shoppers in ChatGPT and Google AI Mode, and in other AI assistants such as Perplexity. Each owner sets its own rules on which merchants appear, how checkout works and which countries are covered, so read each programme's own pages before you plan work around it.

### ChatGPT

ChatGPT shows product results to users worldwide. OpenAI's merchants page says its merchant feed programme is live for US users so far, and that purchases that start in ChatGPT finish on the merchant's own website or app (read 13 November 2026). For now, ChatGPT agentic shopping is a discovery and recommendation step that sends the buyer to you. The [merchants page](https://chatgpt.com/merchants) is the primary source to watch for changes.

Our guide to the [ChatGPT shopping product feed](/blog/chatgpt-shopping-product-feed) covers the feed itself and how to keep it accurate.

### Google AI Mode

Google AI Mode answers shopping prompts with product listings. Merchants supply product data to Google through Google Merchant Center, and Google also reads the pages themselves. For checkout, Google describes the Universal Commerce Protocol as its open standard for agentic commerce.

Google's own pages state which markets and features are available, and that list changes, so check it there. For how AI Mode differs from AI Overviews, see [Google AI Mode vs AI Overviews](/blog/google-ai-mode-vs-ai-overviews).

### Perplexity, Microsoft Copilot and others

Other AI assistants, including Perplexity and Microsoft Copilot, also answer shopping prompts, each under its owner's own rules and merchant terms. Bing's webmaster team links the freshness of its index to AI answers, which is one reason its IndexNow advice, covered in the readiness steps below, is worth acting on.

| Dimension | ChatGPT | Google AI Mode | Other AI assistants |
| --- | --- | --- | --- |
| Where product data comes from | Merchant product feeds and web pages | Google Merchant Center data and web pages | Each owner's own programme |
| Checkout | Finishes on the merchant's own website or app | Google describes the Universal Commerce Protocol; availability on Google's pages | Varies by programme |
| Regions | Product results worldwide; merchant feeds live for US users so far (read 13 November 2026) | Listed on Google's own pages | Listed on each owner's pages |
| Where to start | OpenAI's merchants page | Google Merchant Center | Each owner's merchant pages |

## What agentic commerce changes for online stores

Agentic commerce changes who reads your store first. An AI agent reads your product data, policies and reviews, decides whether to shortlist you, and the shopper may see your brand as one product card among several. Your website still matters: the agent reads it, and in ChatGPT the purchase currently finishes there.

### Product data becomes the sales pitch

A shopper browsing a category page sees photos, badges and layout. An agent sees fields: title, brand, price, availability, size, material and GTIN (Global Trade Item Number, the barcode number that identifies a product across retailers). A missing attribute makes it harder for the agent to match the product to a shopper's constraint, and the agent may fall back on the page text instead, however good the photography.

### Policies become comparison data

Delivery cost, delivery time and the returns window are facts an agent can compare directly. When two products look alike on specification, these may be the only differences the agent can state.

### Duplicate copy costs visibility

Many stores reuse the manufacturer's description, so the same paragraph appears on dozens of retailers. [Bing's webmaster team](https://blogs.bing.com/webmaster/2025/12/Does-Duplicate-Content-Hurt-SEO-and-AI-Search-Visibility/) addressed this on 19 December 2025: "Duplicate content doesn't trigger search penalties on its own, but it does reduce visibility by diluting authority, confusing intent", and it slows how updates reach search engines and AI assistants.

### Traffic looks different

When an agent does the comparing, fewer shoppers may browse your category pages before buying. Visits that do come from AI assistants show up as referrals, which you can separate out by following [how to track AI traffic in GA4](/blog/track-ai-traffic-ga4).

## How to get your store ready for agentic commerce, step by step

To get your store ready for agentic commerce, choose the products and prompts that matter, make product data complete and consistent, let AI crawlers read product and policy pages, mark them up with structured data, keep feeds fresh and write descriptions that answer comparison questions. Then confirm checkout options with your platform or payment provider.

### Step 1: choose products and prompts

Start with the 10 to 20 products that earn most of your revenue. For each, write the prompts a buyer would type, including the constraints they add: budget, use, size and delivery. [Prompt research](/blog/prompt-research) covers how to find and phrase them.

### Step 2: complete the product data

Check that each title names the brand, the product type and the attribute buyers filter on, such as size or capacity. Fill in GTIN or MPN (manufacturer part number), variants, materials and dimensions. Price and availability must match exactly between the page, the feed and the structured data.

### Step 3: let AI crawlers read product and policy pages

Open your robots.txt and confirm it does not block crawlers such as OAI-SearchBot (OpenAI's search crawler) or Googlebot from product, collection and policy pages. User-triggered agents such as [Google-Agent](https://developers.google.com/crawling/docs/changelog), which Google added on 20 March 2026, [generally ignore robots.txt rules, according to Google](https://developers.google.com/crawling/docs/crawlers-fetchers/google-user-triggered-fetchers), so product and policy pages also need to be accurate for them.

On Shopify, refund, shipping and terms pages live under /policies/, and some stores' robots.txt files disallow that path, so agents may not read those pages. Check yours at /robots.txt. If the rule is there, Shopify's [developer documentation](https://shopify.dev/docs/storefronts/themes/seo/robots-txt) shows how to change the rules by editing robots.txt.liquid. The other fix is to put your shipping and returns details on a standard /pages/ URL and link it from your product pages.

Then view a product page with JavaScript switched off. Price, availability and description should still be in the HTML. [How AI crawlers read your website](/blog/how-ai-crawlers-read-your-site) explains why server-rendered text matters.

### Step 4: add structured data

Mark up each product page with schema.org Product and an Offer carrying price, priceCurrency and availability. Add OfferShippingDetails for delivery cost and time, and MerchantReturnPolicy for the returns window. Only mark up ratings that are genuine and visible on the page.

### Step 5: keep feeds and the index fresh

Submit your product feed to Google Merchant Center and follow OpenAI's merchants page for ChatGPT. Update feeds whenever price or stock changes. For Bing and Microsoft Copilot, [Bing's webmaster blog](https://blogs.bing.com/webmaster/2026/2/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview/) (10 February 2026) says IndexNow "helps keep information fresh across search and AI experiences by notifying participating search engines whenever content is added, updated, or removed."

### Step 6: write descriptions that answer comparison questions

Rewrite manufacturer copy in your own words, and cover what an agent will be asked: who the product suits, how it compares with the next one up in the range, what is in the box and how returns work. Plain, specific sentences are easier for an AI assistant to quote than marketing lines.

### Step 7: confirm checkout options

Ask your ecommerce platform and payment provider which agentic checkout standards they support, such as ACP or UCP, and in which countries. If neither supports them yet, steps 1 to 6 still pay off, because ChatGPT currently sends buyers to your own website to finish.

### Worked example: a readiness worksheet (illustrative)

The worksheet turns steps 2 to 4 into seven checks you can run on one product with a browser, your feed export and Google's Rich Results Test.

| Check | Where to look | Passes when |
| --- | --- | --- |
| Title names brand, type and key attribute | Product page and feed | Same title in both, with the attribute buyers filter on |
| GTIN or MPN present | Feed and Product markup | Value present and identical in both |
| Price and availability match | Page HTML, feed, Offer markup | All three agree today |
| Delivery cost and time stated | Shipping page, OfferShippingDetails | Stated on the page and in markup |
| Returns window stated | Returns page, MerchantReturnPolicy | Page crawlable, window in markup |
| Description is your own | A sentence searched in quotes on Google | No other retailer shows the same sentence |
| Visible without JavaScript | View source | Price and description appear in the HTML |

Illustrative example: a fictional coffee grinder store runs the worksheet on its 10 best sellers, 70 checks in all. Suppose 12 fail: 6 missing GTINs, 4 with no shipping markup and 2 with copied descriptions. Fix identifiers and shipping first, because those are the hard facts an agent filters on.

To verify the fixes, run each product URL through Google's Rich Results Test, check Merchant Center diagnostics for feed errors, and repeat the worksheet a week later. The expected output is a sheet where every row passes for every product you chose in step 1.

## How to tell whether AI agents recommend your products

You can tell whether AI agents recommend your products by running the same set of shopping prompts in ChatGPT and Google AI Mode on a schedule. Each time, record whether your products appear, in which position, which merchants appear instead and which pages are cited. One run is a snapshot; the trend over several weeks is the answer.

### A manual method

1. Pick 10 to 20 shopping prompts from step 1 of the readiness plan.
2. Run each in ChatGPT and in Google AI Mode, in a fresh session, from the country you sell to.
3. Record whether your product appears, its position, which merchants appear instead and which pages are cited.
4. Repeat on the same day each week and compare.

A simple log keeps the runs comparable:

| date | AI assistant | prompt | our product shown (y/n) | position | merchants shown | pages cited |
| --- | --- | --- | --- | --- | --- | --- |
| 11 November 2026 | ChatGPT | quiet burr grinder for espresso under $200 | n | - | Merchant A, Merchant B | review article, retailer page |

Answers vary between runs and between people, because AI assistants personalise and update their results.

### Tracking it on a schedule with RankX AI

RankX AI's AI Shopping module is opt-in and switched on per website. Once on, it checks each shopping prompt you choose on ChatGPT Shopping and Google AI Mode and reports the share of answers that showed products. It shows which merchants' products appeared and which slot yours took.

The [pricing page](/pricing) puts the cost at "5 credits per shopping prompt, per surface, per run". As a calculated example, 10 prompts on ChatGPT Shopping and Google AI Mode is 20 checks, or 100 credits a run.

Every RankX AI shopping finding becomes a proposal you approve, and the [AI Shopping feature page](/features/ai-shopping) says "Nothing is applied on its own." When the checks point to a product-page problem, a [website audit](/features/website-audit) writes a fix plan from the findings, and one click files it as tracked Tasks, deduplicated against work already open. On WordPress, every edit defaults to a dry run, [applying is a second, explicit step](/integrations/wordpress), and an article is published only once someone has approved it.

On WooCommerce and other websites, plans come with a 7-day free trial and 1,000 credits; Shopify stores install the app, which has a free plan and paid plans with a 7-day trial. Plans are set out on the [pricing page](/pricing).

## Risks and open questions for merchants

The main risks of agentic commerce for merchants are a weaker direct relationship with the shopper, errors in data an agent repeats, unsettled rules on authorisation and disputes, and coverage that differs by country. Several are open questions, because the standards are new and their owners are still changing their programmes.

### Who owns the customer

When the agent handles discovery and, in some flows, checkout, the shopper's relationship can sit partly with the AI assistant. Stripe says ACP keeps the business as the merchant of record, but email consent, accounts and repeat-purchase data depend on each programme's terms. Read those terms before you opt in.

### Your claims, repeated by an agent

An agent repeats the data you give it, so a wrong price or stock level travels further. In the UK, the CMA has enforced consumer law directly since April 2025, and its current [unfair commercial practices guidance (CMA207)](https://www.gov.uk/government/publications/unfair-commercial-practices-cma207), first published on 4 April 2025 and updated on 18 November 2025, sets out the rules under the Digital Markets, Competition and Consumers Act 2024.

On scarcity, the [CMA's blog of 31 March 2023](https://competitionandmarkets.blog.gov.uk/2023/03/31/urgency-and-price-reduction-claims-are-your-online-tactics-legal) warned: "Claims that are untrue because the business in fact has plenty of stock are misleading." Our reading: an agent passing that claim on does not make it any less yours.

### Authorisation, fraud and disputes

Agent-initiated payments raise open questions: what counts as proof the shopper approved a purchase, and who carries the loss when an agent buys the wrong item. Google's [Agent Payments Protocol (AP2)](https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol) addresses the first question with signed mandates that serve as proof of a user's instructions, while dispute and liability rules are still unsettled. We do not know how every dispute type will be resolved. Ask your payment provider how it handles agent-initiated orders and chargebacks.

### Coverage differs by country

ChatGPT shows product results to users worldwide, while OpenAI's merchants page says its merchant feed programme is live for US users so far. Stores selling elsewhere can still prepare data and pages now, because the same work serves Google, other AI assistants and ordinary search. Check each owner's page for changes to market coverage.

### Results move

AI answers change between runs, and no AI assistant publishes how it ranks merchants.

## Where to start with agentic commerce

Start with agentic commerce by fixing what every AI shopping agent reads: complete, consistent product data, crawlable product and policy pages, and accurate structured data. Then set a baseline of shopping prompts in ChatGPT and Google AI Mode and check it on a schedule, so you can see whether the work changes which products agents recommend.

If you want that baseline checked on a schedule for you, [Start your free trial](/features/ai-shopping) and switch on the AI Shopping module for your website.

## Check one product from prompt to checkout

Pick one best-selling product and one shopping prompt from step 1. Run the seven worksheet checks on that product, then run the prompt in ChatGPT and Google AI Mode and record whether the product appears, which merchants appear instead and which pages are cited. Keep the date with both results so the next run compares like with like.

Fix the failed checks an agent filters on first: identifiers, price and availability, delivery and returns. Repeat the worksheet and the same prompt a week later. A passing worksheet shows the product data is ready for an agent to read; it does not guarantee that an agent shortlists the product, and a sale needs its own evidence in your store's orders.

Next, [see how ChatGPT product feeds work](/blog/chatgpt-shopping-product-feed) or [check what AI crawlers can read](/blog/how-ai-crawlers-read-your-site).

## What is the Agentic Commerce Protocol?

The Agentic Commerce Protocol (ACP) is an open standard that Stripe developed with OpenAI so AI agents and businesses can complete purchases together. Stripe says the business remains the merchant of record. In ChatGPT today, OpenAI's merchants page says ACP supports product discovery while purchases are completed on merchant-owned websites or apps, and its merchant feed programme is live for US users so far. Whether you can use it depends on your ecommerce platform and payment provider, so ask them before planning any build.

## What does agentic commerce mean for Shopify stores?

For Shopify stores, agentic commerce means AI agents read your product data, theme markup and policy pages alongside your shoppers. Shopify's own UCP page (shopify.com/ucp) says Shopify co-developed the Universal Commerce Protocol with Google. Some Shopify stores' robots.txt files disallow /policies/, so check yours at /robots.txt; if the rule is there, edit robots.txt.liquid to allow it, or put shipping and returns details on a standard /pages/ URL linked from your product pages. Check Shopify's own announcements for which AI assistants and markets your store can sell through.

## How do I optimise my store for AI shopping agents?

Make product data complete and identical across the page, the feed and the structured data, keep product and policy pages crawlable, and write your own descriptions. Then check a fixed set of shopping prompts in ChatGPT and Google AI Mode on a schedule to see whether the changes affect what agents recommend.

## Is agentic commerce available outside the US?

Partly. ChatGPT shows product results to users worldwide, but OpenAI's merchants page says its merchant feed programme is live for US users so far (read 13 November 2026), and Shopify's ChatGPT channel needs a store that sells to US customers. Elsewhere, availability varies by service: Google and the other owners list their covered markets on their own pages, and those lists change, so check each one directly.

## Is agentic commerce the same as AI shopping?

AI shopping is the wider term: any product recommendation inside an AI assistant. Agentic commerce adds the agent taking steps for the shopper, up to checkout. The preparation for both is the same: accurate product data and pages agents can read.
