E-commerce
August 26, 2026
Since early 2026, AI buying agents are no longer just prototypes: Google, Adyen, and other major players have released tools to allow machines to compare, select, and buy without ever opening a browser. You may have heard of agentic commerce, but what do you really need to change in your store so that these agents can buy from you? The answer lies in three technical projects: making your catalog readable by a machine that never loads a page, identifying these agents without blocking them like scrapers, and authorizing a delegated payment by the cardholder without them being present. Everything else flows from these three foundations. This article details each of them, the consequences if you neglect even one, and the steps to prepare your platform without getting lost in the marketing hype.
Summary
Why three changes are enough
Agentic commerce is based on a simple observation: an AI purchasing agent does not behave like a human visitor. It does not load a page, click on anything, or wait for any animation. It directly queries your product data, requests access to your catalog, and presents a payment authorization delegated by its user. If you block any of these three steps, the agent stops there. It will never see your prices, or it will compare but abandon at the time of payment.
These three changes affect systems that your marketing team generally does not control: the product data pipeline, the application firewall, and the payment stack. This is why agentic commerce is as much a technical project as a commercial one, and why it requires coordination between catalog, security, and finance. If you do not align these three building blocks, you will open one door only to immediately close the next.
Convert over 2,000 customers on average per month with Qstomy.
The world’s 1st Shopify AI dedicated to customer conversion
Empowering 200+ e-commerce merchants
Machine-readable catalog: beyond the classic product feed
An agent needs price, availability, variants, and identifiers without ever viewing a page. This means your product feed and your structured data on the page become the product experience itself, not a mere copy. Comparing two items requires a stable identifier like a GTIN or a SKU, an up-to-date unit price, real-time stock, variant attributes such as size, color, material, and the delivery and return conditions that determine the total cost.
The schema.org Product and Offer types carry these properties. Product feed specifications carry them in XML or delimited files. When these two sources diverge, the agent cannot know which one is correct: it acts on the one it read first. Any data present only in the rendered HTML, in an image, or in a graphic element is invisible. A fast shipping badge drawn in CSS, a discount applied to the cart, a return window written in the footer: none of this reaches a customer who reads the feed and the structured markup. To learn more about optimizing your product data for search engines, see our guide on SEO for e-commerce sites.
Freshness and completeness: the two pitfalls of the agentic catalog
Freshness is the hard part. A feed refreshed every night displays an obsolete price as soon as an item moves during the day, and the agent will place an order at this incorrect price. Completeness becomes the new merchandising surface: an agent filtering on a field you left blank silently excludes you, with no error message and without letting you compete on the attributes you did fill in.
Which fields are critical, and what happens if any are missing? This depends on the protocol and the client, but expect to have to document material, dimensions, estimated delivery time, return policy, and complete variants. The discipline that makes your storefront readable by LLM crawlers is exactly the same that makes it readable by a buying agent. If you are looking to structure your data to appear better on advertising channels, read our article on Google Shopping for e-commerce.
Bot Policy: How to recognize a legitimate agent without opening the door to scrapers
A legitimate buying agent looks identical to a malicious scraper. It queries many product pages quickly, runs without a graphical browser, produces no human interaction signals like mouse movement or dwell time, and often arrives from cloud address ranges that reputation systems treat as suspicious. Your current anti-bot defenses, designed to block price scraping and credential stuffing, will stop a buying agent in exactly the same way.
Four identification methods exist today. The agent declares itself in its user-agent string, which anyone can copy. The operator publishes its IP ranges and supports reverse DNS verification, which is verifiable but binds the agent to a fixed infrastructure. Requests carry a signature that you verify against a public key, proving the origin. Or you add a named partner to a whitelist via commercial agreement, which is enforceable but difficult to scale. Declaration alone is not enforceable, and this creates an asymmetry: anyone can claim to be a buying agent, and the cost of a false clearance is paid in scraped prices rather than a missed purchase.
Delegated payment: authorizing a transaction without the cardholder's presence
A purchasing agent does not hold a card registered in their own name. They present an authorization delegated by the cardholder, who is not present at the time of the transaction. Your payment flow must therefore accept an authorization request that does not come directly from the cardholder, while remaining compliant with the security and liability rules of the card networks.
Adyen launched Adyen Agentic in June 2026 specifically to handle this use case. Other payment players are preparing similar solutions. The challenge is to prove that the cardholder has indeed delegated the authorization, while minimizing friction for the agent and protecting the merchant against fraud and chargebacks. If your current checkout process requires the cardholder to validate each payment in person, you will block the agent at the final step, after they have chosen your products and validated their cart. To understand how to optimize the entire purchasing journey, explore our article on how e-commerce marketing works.
Making a single change without the others: the silent failure
Opening a machine-readable catalog behind a firewall that blocks all agents is like building an invisible storefront. Allowing access to the catalog without offering a delegated payment pathway produces an agent that compares, selects, and abandons at the moment of payment. Accepting delegated payment without a readable catalog leaves the agent blind to your prices and your stock.
None of these failures trigger a loud alert. The agent does not warn you that it cannot read your catalog or that it has been blocked. It simply moves on to the next merchant. You lose potential sales without even knowing that agents tried to buy from you. This is why it is essential to treat these three projects as an indivisible whole, and to test the complete journey before communicating about your agentic compatibility.
Concurrent protocols and emerging standards
In early 2026, several competing agent-merchant protocols emerged. Google released tools in January 2026 for retailers preparing for the era of agentic commerce. Other players announced their own standards. To date, no single protocol has established itself as the universal benchmark. This means you may need to support multiple feed formats, authentication methods, and payment paths if you want to be accessible to the widest possible range of agents.
The question is no longer whether platforms will support agent-driven purchasing, but which protocol they will choose and how quickly they will converge. Read vendor announcements as statements of intent rather than proof of technical requirements. Measured performance data is not yet public, and best practices are evolving as deployments roll out. If you are wondering how Google structures its e-commerce ecosystem, our article on Google's e-commerce platform will shed some light.
Merchandising for agents: the attribute becomes the message
In a world where the agent reads your product attributes without ever seeing your page, every empty field is a missed opportunity. Rich product sheets, evocative descriptions, polished visuals: all of this disappears if the agent filters on a property you haven't filled in. Material, weight, dimensions, origin, certifications, compatibility: every attribute becomes a potential selection criterion.
Agentic merchandising therefore consists of filling in, verifying, and keeping all of your product metadata up to date, with the same rigor you bring to your advertising campaigns. An incomplete catalog makes you invisible to agents who filter on missing fields. An outdated catalog makes you sell at the wrong price or promise stock you no longer have. To go further on the technical optimization of your catalog, consult our guide on e-commerce SEO and how it works.
Failure modes that no one has solved yet
Several failure scenarios remain without a proven solution. What happens when an agent places an order at the exact moment your stock drops to zero? How do you handle a delegated payment that is declined after the order is confirmed? What liability do you assume if an agent compares your prices based on an outdated feed and then disputes the billing?
Current protocols do not specify a real-time stock reservation or cancellation mechanism. Delegated payment flows have not yet accumulated enough history to calibrate fraud detection models. General terms and conditions of sale do not always cover purchases made by an automated third party on behalf of a customer. You will need to document these edge cases with your legal, logistics, and payment teams before declaring yourself ready for agentic commerce. To better understand order management at scale, read our article on the role of an order management system.
Where to start: sequencing the three projects
Start by auditing the freshness and completeness of your catalog. Identify missing fields in your product feed and structured data, then set up a process to keep them updated continuously. Next, review your bot policy: test whether a declared agent can reach your product pages without triggering blockages, and evaluate the identification methods you are willing to accept.
Finally, engage with your payment provider to understand which delegated payment options they offer, which partners they already support, and their timeline for upcoming integrations. Deploy the three changes only when you can test the complete journey, from catalog reading to order confirmation. A partial deployment creates more confusion than value.
Qstomy and Agentic Commerce: Conversion and Human Support Above All
Qstomy is a Shopify AI agent that guides your human visitors towards making a purchase: personalized recommendations, contextual upselling and cross-selling, cart assistance, parcel tracking, and automated customer service. More than 100 merchants are already using it to increase their conversion rate and reduce abandonment. Unlike shopping agents that read your catalog to compare and order without human interaction, Qstomy accompanies your customers throughout their journey on your store, answers their questions in real time, and helps them finalize their order.
Agentic commerce and Qstomy address two complementary needs. Shopping agents automate the decision and execution for an end user who delegates their recurring purchases. Qstomy optimizes the experience and conversion for visitors who browse, compare, and make their own choices. Both approaches share the same requirement: a structured, fresh, and complete catalog. By preparing your store for shopping agents, you also improve the quality of recommendations that Qstomy can make, since both draw from the same product data.
Checklist, brief summary, and frequently asked questions
Agentic Commerce Checklist:
Catalog: Complete, fresh, and consistent product feed and schema.org structured data.
Bots: Access policy that identifies legitimate agents without blocking, via IP, signature, or agreement.
Payment: Delegated payment flow enabled with your provider, terms and conditions updated.
Testing: Full end-to-end journey validated before public announcement.
In short: Agentic commerce requires three inseparable technical changes: a machine-readable catalog, a bot policy that recognizes agents without blocking them, and a delegated payment flow. Neglecting any of these three pillars results in a silent failure that costs you sales without warning. Start with the catalog, test bot access, then enable payment, in that order.
Frequently Asked Questions
Can a buying agent really purchase from me without ever opening my store?
Yes, if your catalog is machine-readable, your firewall allows it, and you accept delegated payments. Otherwise, it will be blocked at one of these three stages.
How do I know if an agent tried to buy from me and failed?
You will not know automatically. Failures are silent. You must actively test the agent journey and monitor your firewall and payment processor logs.
Do I need to choose an agent-merchant protocol right now?
Several protocols coexist in early 2026. Prioritize the standards supported by your payment and platform partners, and prepare to support several of them if you aim for broad coverage. To delve deeper into the difference between marketing and advertising in this context, read our article on e-commerce marketing and advertising.
Enzo
August 26, 2026
