Agentic commerce is buying and selling where AI agents research products, compare options and take purchase actions on someone's behalf, within permissions they set. An agent might prepare a cart for approval or complete an authorised order. The distinction is permission. A shopping recommendation alone doesn't mean an agent can spend money, and a product feed doesn't switch on checkout.
For enterprise sellers, the opportunity starts before payment: can an agent find your offer, understand its limits and recommend it correctly?
In summary
- AI shopping covers discovery and comparison; agentic checkout adds the ability to act on an authorised purchase.
- OpenAI, Google, Visa and Mastercard address different parts of that process. Their announcements aren't interchangeable checkout integrations.
- Google now names Australia and the US among eligible UCP markets, but participation remains limited. Check the actual merchant requirements.
- B2B sellers need accurate specifications, account permissions and procurement approvals, not just a payment button.
- Search visibility, accepted orders and completed sales need separate measurement. A citation isn't a purchase.
I'm Brenden House, founder of Searchmaxxed. We do SEO and AI search optimisation. Our work helps buyers discover your business and understand what it offers. Payment authorisation needs separate engineering and security work, because a website that explains an offer clearly cannot establish a buyer's authority to spend.
How is agentic commerce different from ordinary online shopping?
Agentic commerce delegates shopping and purchasing actions to software acting for the buyer. The customer sets the boundaries, including which suppliers are acceptable and which changes require somebody to review the order before it proceeds.
Think about replenishing industrial components. A buyer could specify an approved manufacturer, exact part number, required certification and delivery deadline, then authorise an order under existing purchasing rules. That's a delegated task. An explanation alone completes no purchase.
| Experience | What the software does | Who controls the purchase? |
|---|---|---|
| Traditional ecommerce | Displays products and checkout | The buyer selects and orders |
| AI-assisted shopping | Answers questions and recommends options | The buyer follows a link or continues manually |
| Agentic checkout with approval | Prepares or submits an order after confirmation | The buyer approves the proposed purchase |
| Delegated purchasing | Acts when predefined conditions are met | The buyer sets permission and spending limits beforehand |
These distinctions describe who can act on the buyer's behalf, rather than suggesting that every platform uses the same names for its shopping features. Stripe's agentic commerce guide describes an approval-based flow; Google's AP2 announcement also describes purchases made under advance permission.
For Searchmaxxed's website work, the immediate question is whether the product page answers the selection criteria. A readable product page gives the agent evidence for comparison, including whether the item meets the specification the buyer supplied at the start. It doesn't confer purchasing authority.
How do AI agents make purchases?
AI agents make purchases through connected merchant and payment systems, or supported website interactions, using authority supplied by the buyer. The merchant can still decline.
An authorised purchase needs a sequence that connects the original request to the order the merchant accepts, rather than assuming each handoff succeeded:
- The buyer supplies requirements, exclusions and permission limits.
- The agent finds products and compares relevant specifications, availability and delivery terms.
- The merchant provides a current offer or checkout session.
- The buyer confirms the purchase, or the system checks an existing mandate.
- The merchant processes payment and returns an order confirmation.
- Fulfilment, tracking, returns and support continue through the agreed merchant process.
Each handoff can fail. A product can be recommended but unavailable, or paid for without meeting the buyer's intended specification unless the system checks it.
An API connects software systems. A payment token substitutes for sensitive payment credentials within a defined payment flow, so the purchasing system needn't pass around the original credentials. Neither fixes inaccurate product information.
OpenAI's Instant Checkout announcement describes the merchant accepting or declining an order, processing payment and handling fulfilment. It also limits payment tokens to specific amounts and merchants with user permission.
That's why our SEO work checks the facts a buyer uses before checkout. Commerce teams validate the transaction separately.
What have OpenAI, Google, Visa and Mastercard announced?
OpenAI and Google have introduced commerce protocols and shopping experiences; Visa and Mastercard have announced payment capabilities for agent-initiated transactions. They address different problems within a purchase, which is why a payment announcement should not be mistaken for a new product discovery channel.
| Announcement | Published | What it introduced | What merchants should distinguish |
|---|---|---|---|
| Mastercard Agent Pay | 29 April 2025 | Agentic Tokens, verified agents and consumer purchase controls | Payment capability is separate from product discovery |
| Visa Intelligent Commerce | 30 April 2025 | Access to Visa's network for developers building AI commerce | A network announcement doesn't prove your checkout is connected |
| Google Agent Payments Protocol | 16 September 2025 | Signed records of user intent and purchase approval | AP2 addresses transaction authority, not search ranking |
| OpenAI Instant Checkout and ACP | 29 September 2025 | In-chat purchasing and an open commerce standard developed with Stripe | The launch began with US users and US Etsy sellers |
| Google Universal Commerce Protocol | 11 January 2026 | A shared standard covering discovery, checkout and post-purchase support | The launch described eligible US retailers; current guidance is broader |
What does OpenAI's current merchant guidance say?
OpenAI's merchant page now describes product discovery with purchases completed on the merchant's site or app by default.
OpenAI now says it is moving away from standalone Instant Checkout and prioritising shopping discovery and merchant-owned checkout. The launch announcement describes its earlier approach. OpenAI's product-feed specification says checkout requires a separately enabled integration.
Decide on discovery feeds separately. The project scope should say whether you need catalogue discovery, an approved checkout integration or both, instead of calling the entire thing agent-ready.
Is agentic checkout available in Australia?
Google's current UCP help page names the United States, Canada and Australia for participating merchants and partners. Access remains selective.
The page says rollout happens in phases and only eligible product listings display the checkout button. It doesn't promise access for every Australian merchant.
OpenAI's merchant page names US users. Confirm Australian availability for the specific ChatGPT shopping feature you plan to use.
Check buyer and merchant eligibility separately. Payment-provider availability needs checking as well.
What's the difference between ACP, UCP, AP2 and MCP?
ACP and UCP describe commerce interactions, while AP2 records payment authority and MCP connects AI applications with tools and data. Knowing the name isn't an implementation plan.
| Term | Plain-English meaning | The question it helps answer |
|---|---|---|
| Agentic Commerce Protocol, ACP | OpenAI and Stripe's open standard for AI commerce | How can an agent and merchant exchange commerce information? |
| Universal Commerce Protocol, UCP | An open standard connecting agents, businesses and payment providers | How can discovery, checkout and support work across systems? |
| Agent Payments Protocol, AP2 | A protocol using signed mandates to record authorised purchasing intent | What exactly did the buyer permit? |
| Model Context Protocol, MCP | A way for AI applications to connect with external tools and information | Which data or actions can this application access? |
A mandate is a verifiable record of the buyer's instructions or approval. Google describes an Intent Mandate for the request and a Cart Mandate for the purchase details.
Permission can be narrow. An agent might read a catalogue without permission to place orders, or prepare a cart without permission to pay.
UCP is designed to work with AP2 and MCP, according to Google's announcement. They're not all competing replacements for the same thing.
Searchmaxxed's structured-data work helps machines interpret your public pages. Bring your platform team into transaction design, because authority, payment handling and order processing need controls beyond the information shown on a public page.
What should merchants and B2B sellers prepare first?
Fix the offer information first. B2B sellers need to distinguish public product facts from account-specific terms and purchase permissions, especially when a logged-in customer sees a different offer.
Start with a real order. Choose an offer customers already buy, then trace what somebody needs to know before committing.
| Business | Information the agent needs | Where a person may need to intervene |
|---|---|---|
| Manufacturer or industrial distributor | Exact part, revision, compatibility, certification and delivery conditions | Substitutions or safety-critical specifications |
| B2B SaaS company | Supported integrations, deployment options, product limits and procurement documents | Security review, contract changes and implementation scope |
| Enterprise retailer | Correct variant, availability, shipping and returns | Exceptions, unavailable stock or unusual delivery requirements |
| Public-sector supplier | Eligibility, evidence and purchasing-route requirements | Formal procurement approval and contract award |
These are planning examples only.
For Searchmaxxed, this becomes concrete page work: product and industry pages, comparison content, relevant schema and internal links that explain the offer without contradictions. We check crawl access and rendering too.
The technical detail lives in our agent-ready website guide. For catalogue formats and platform requirements, use our AI shopping product-feeds guide.
How would you test an industrial-parts purchase?
Use an approved spare part as a hypothetical test, then change the conditions that could make the order wrong. Start without live purchasing enabled.
Use this acceptance checklist with your product, commerce and security teams:
| Test condition | Expected behaviour | Evidence to retain |
|---|---|---|
| Exact part available from an approved supplier | Prepare the correct item and purchasing unit | Part number, revision and supplier |
| Similar part offered as a substitute | Stop unless substitution is expressly allowed | Original requirement and proposed difference |
| Public page and stock system disagree | Revalidate availability before submitting | Availability response and timestamp |
| Cart changes after approval | Recheck permission against the changed order | Approved cart and final cart |
| Supplier requires a negotiated contract | Hand off to the authorised buyer | Quote or contract reference |
| Order response is unclear | Check order status before trying again | Request identifier and merchant order record |
The last row deserves attention. Silence doesn't establish failure. Repeating a purchase request without checking the merchant's order records can create a duplicate order, even though the first attempt appeared to fail.
We address the public page facts. Your engineering team owns transaction tests; your procurement team sets the purchasing rules.
What are the main risks of agentic commerce?
Agentic commerce risks include incorrect orders, excessive permissions, exposed customer data, fraud and unclear responsibility when something goes wrong. Put controls around the action itself, so a purchasing limit is enforced by the system rather than relying on the agent to remember it.
Clear wording helps. A clear instruction is useful context, but the transaction system must still check whether the proposed purchase falls within the permission the buyer supplied.
| Risk | Practical control to require |
|---|---|
| Wrong variant or unsupported claim | Validate the selected item against authoritative product data |
| Spending beyond the buyer's intent | Enforce allowed merchants, purchase limits and approval conditions |
| A fake or compromised agent | Authenticate the agent and validate its authority |
| Private terms or credentials exposed | Restrict access and share only necessary information |
| Duplicate or disputed orders | Retain approval and order records; check retries against existing transactions |
| Confusion over refunds and support | Name the seller and preserve a clear returns process |
These are implementation checks, not a legal opinion or a claim that any protocol removes fraud. Google's AP2 announcement addresses authorisation and authenticity through signed mandates. Mastercard also describes verified agents and controls.
The seller still matters. OpenAI's launch and Google's UCP guidance both keep the merchant responsible as merchant or seller of record.
For enterprise and public-sector purchases, involve security, procurement and legal teams before enabling delegated spending. Searchmaxxed can fix misleading website claims; we don't substitute for those approvals.
Does agentic commerce replace SEO and AI search optimisation?
Agentic commerce still needs product discovery, so SEO and AI search optimisation address an earlier problem than checkout. An agent can't choose an offer it never finds or misunderstands.
Being retrieved is only the start. Our own AI-search research shows the distinction.
We put 300 Australian agency-buyer questions to ChatGPT three times each, with search on. It pulled 17,870 search results into those 900 answers and kept 4,389 as sources. Being found is the easy part. About three in four results it reads never make the answer.
That was Searchmaxxed's research on 2 October 2026. It measured agency recommendations, not shopping transactions, so it isn't a forecast of agent purchase behaviour.
The useful lesson is narrower: track each step separately. Website access, inclusion in an answer, purchase approval and a completed order are different observations.
Our AI search optimisation service covers the discovery work: clear product facts, technical SEO, relevant content and matching structured data. We also track what AI answers say about the business.
For a SaaS seller, that could mean correcting an integration claim before it misleads a shortlisted buyer. Manufacturers need readable compatibility details.
How should businesses measure agentic commerce?
Measure completed orders and qualified enquiries alongside AI discovery, then connect the records only where you have evidence that they describe the same purchase. Agent activity on its own isn't revenue.
The report should let your marketing and commerce teams distinguish what the assistant found from what the customer approved and the merchant eventually fulfilled:
- Which buyer questions show your business, and which answers describe it correctly?
- Which product pages or feeds supplied the information?
- Did the buyer visit your site, approve a cart or submit a qualified enquiry?
- Did the merchant accept the order, fulfil it and retain the sale after returns?
- Which parts of that sequence can you join through recorded identifiers?
Don't invent the missing joins. A visit from an AI tool shows a referral; a server request might only be retrieval. Neither establishes delegated purchasing.
Searchmaxxed connects search measurement with website conversion signals. If an order completes away from your site, include the merchant's transaction records rather than expecting website analytics to explain everything.
Is Searchmaxxed right for your agentic commerce work?
Searchmaxxed fits companies that need their products and services found, understood and recommended in Google and AI search. We aren't a payment network or a ready-made procurement platform.
That distinction sets the scope.
Our Growth System covers ongoing page releases, technical SEO, schema, internal links and AI visibility tracking, with implementation handled as part of the work. Agentic Search Engineering expands that scope. It adds comparison, industry and integration page families, alongside tracking across ChatGPT, Gemini, Perplexity and AI Overviews to see which answers name you.
Custom & Enterprise can include multiple markets, dedicated engineering, custom reporting and original first-party research, depending on the scope agreed for your business. Any commerce integration needs an explicit scope; the package name doesn't establish support for a payment protocol.
The business keeps its website, repository, content and data. You work directly with the people doing the work, without an account-manager relay.
| Searchmaxxed is a good fit if | Look elsewhere if |
|---|---|
| Your enterprise, SaaS or manufacturing buyers struggle to find or compare your offer | You only need payment or checkout integration |
| You want technical SEO, content and website changes implemented | You want strategy advice with no changes to your site |
Frequently asked questions
Can AI agents buy things without approval?
An agent can act under permission given in advance, rather than asking again at checkout. Define the permitted spend and conditions. Google's AP2 describes both live approval and delegated purchases, using signed records that connect the buyer's instructions with the details of the proposed order.
Is a shopping chatbot agentic commerce?
A chatbot that only answers questions isn't completing a delegated purchase. It becomes part of an agentic commerce flow when it can take authorised actions through connected systems.
Does a product feed enable agentic checkout?
No. Product feeds supply catalogue information. OpenAI's product-feed documentation says checkout requires a separately enabled integration, while Google's UCP checkout also has participation and implementation requirements.
Who handles refunds after an AI agent buys something?
The merchant remains responsible under the merchant-of-record arrangements described by OpenAI and Google. Check the seller's returns process and the payment provider's dispute procedures before enabling purchases.
Can B2B SaaS companies use agentic commerce?
B2B SaaS companies can prepare product information for agent-assisted comparison and consider authorised purchasing for suitable transactions. Negotiated contracts, security reviews and approval requirements still need an agreed process.
Does accepting Visa or Mastercard make a website agent-ready?
Ordinary card acceptance doesn't establish a verified agent integration. Ask your payment provider what it supports, how authority is validated and what records you receive.
Where should you start?
Start with the offer your buyers struggle to find or understand, then decide whether checkout automation solves a real problem. Keep the buying rules intact.
Bring that offer, your target markets and the current website to Searchmaxxed. We'll discuss the SEO and AI search work needed to make the business easier to find and compare.