We’ve made significant strides in developing the infrastructure for the next generation of commerce. Over the past six months, we launched the Agentic Commerce Protocol (ACP), an open source specification that lets buyers, AI agents, and sellers transact through APIs; we introduced the first complete agentic solution with the Agentic Commerce Suite and onboarded leading businesses such as URBN (including Anthropologie, Free People, and Urban Outfitters) and Etsy; and we powered AI shopping experiences across AI agents. Most recently, we rolled out support for machine payments, allowing content and data to be monetized directly by agents.
Along the way, we’ve sat down with dozens of businesses, from agile SaaS startups and B2B platforms to major telcos, utility companies, and Fortune 500 retailers. Every sector is preparing for agentic commerce.
In each conversation, businesses ask the same questions: Which channels should I launch first? How do I get discovered by agents? And how do I find the right partner to build this?
These questions ladder up to five main themes:
- Ensuring agents know you exist
- Helping agents understand how to use your site
- Configuring your site to handle nonhuman traffic
- Aligning your organization around agentic commerce
- Establishing AI governance
This guide draws on real-world conversations and Stripe’s deep expertise working with leading businesses preparing for and adopting agentic commerce. Whether you’re exploring agentic commerce for the first time or refining an existing strategy, this technical field guide outlines concrete things you can do today to optimize your business for agentic commerce. This isn’t an exhaustive list; rather, this guide covers key tactics that we’ve seen other businesses implement with success.
1. Ensure agents know you exist
To be recommended by agents, you have to be visible. This requires ensuring your technical infrastructure welcomes agents rather than turning them away and deploying an optimized product catalog feed. It’s still early to formalize a discovery strategy when the models themselves are still evolving, but there are concrete steps you can take to improve how agents parse and understand your website and product catalog.
Confirm that your robots.txt file and firewall settings welcome agents
Some companies, in an effort to protect their data, unwittingly block the very crawlers (like GPTBot or Google-Extended) that power these AI recommendations. If you block the agent, your website is effectively invisible.
To ensure your robots.txt is optimized for agentic discovery without exposing sensitive admin routes, you need to differentiate between verified commerce agents and generic scrapers. A good starting point is to check your current configuration against known agent user agents (such as GPTBot, ClaudeBot, and Google-Extended). You can find updated lists and configuration examples directly in the official crawler documentation for OpenAI, Anthropic, and Google, which allows you to distinguish verified AI agents from malicious scrapers.
You might also want to explicitly allow access to high-value static content (like product detail pages and category archives) while strictly disallowing expensive dynamic paths (like internal search results, cart add actions, or dynamic pricing API endpoints). This guides agents toward your most important pages while protecting your infrastructure from wasteful compute cycles.
There is one major caveat. If your business model is the content itself (e.g., news publishers, specialized data providers, or creative writers), giving it away for free to train an LLM might not make sense. Instead, you might strategically choose what content you expose for free (such as summaries or recaps) and what you keep gated (such as the full article). You can also monetize agents using machine payments. This allows you to provide your data for a fee, turning AI agents into paying customers rather than content scrapers.
Make sure your content is rendered server-side
Many agents struggle to process client-side rendered content because they can’t execute the JavaScript that loads it. If your product pages, prices, or availability information are rendered client-side, agents might miss key details.
Server-side rendering is the recommended approach for agentic readiness. With server-side rendering, your server generates the full HTML before sending it to the requester, which means agents get complete, ready-to-parse content without needing to execute JavaScript. If implementing full server-side rendering across your site isn’t feasible, consider static site generation or hydration.
Optimize your product information for agents with a /llms.txt file
You have a sitemap.xml for Google, but agents can struggle to efficiently parse heavy HTML pages cluttered with navigation menus and ads. You can give them a high-signal text file to help your pages get found.
To achieve this, consider publishing a /llms.txt file. This is a simple Markdown file at your root that acts like a curated index for AI. It should list your core product collections, return policies, and documentation in plain text. By feeding the model clean, structured text, you can increase the probability that agents correctly cite your products and policies.
Stripe uses this approach: our /llms.txt file at stripe.com/llms.txt provides agents with a clear directory of our offerings. We also make all our documentation available in Markdown format for efficient agent reference (e.g., see docs.stripe.com/payments/accept-a-payment.md). This dual approach helps agents understand both what we offer and how to use it.
Visit the official standard at llmstxt.org to learn more and see a directory of live implementations from other tech-forward brands at llmsdirectory.com.
Deploy a product feed
While product feeds are not available for all agents yet, we expect them to be the most important entry point for agents to discover your products. While agents can crawl your website to gather product information, direct data feeds ensure agents receive your product changes immediately, rather than waiting on slow, sometimes unpredictable crawling cycles. They also give you precise control over launch timing—allowing you to set exact dates for when products appear on agentic surfaces—and they can help to eliminate hallucinations by ensuring agents get the most accurate pricing, availability, and specifications.
One challenge: different agents will likely want your data in different formats. One might need an SFTP file drop. Another might want a custom API integration. A third could have its own feed spec entirely. We’ve already seen brands reformat the same product catalog in six different ways to get listed across multiple agents—creating an ongoing maintenance burden that drains time and resources.
You can prepare by partnering with a commerce provider, such as Stripe, that can distribute your catalog anywhere. With Stripe’s Agentic Commerce Suite, you can upload your catalog directly to Stripe or connect your existing syndicator. We’ll distribute your product information to each AI agent automatically.
2. Help agents understand how to use your site
Unlike traditional machines, LLMs operate around semantic understanding. As a result, communicating with agents in ways that mirror natural human language is especially important; they need explicit technical instructions to interact with your business. This requires hosting a suite of specific files on your domain that function as instruction manuals for agent discovery and interaction.
Tell the agent who you are and what your API can do with well-known/ai-plugin.json
This is the entry point for agent discovery. Think of it like a biography for your brand, telling the agent who you are and what your API can do.
Your description_for_model field is particularly important and should be prompt-engineered, meaning it uses clear, actionable language that explains when to use your API and what it can do. Agents perform best with specific capabilities and concrete use cases, not generic marketing language.
For example, don’t just say “We sell shoes.” Instead, consider language like: “Use this website to search for athletic footwear, check real-time size availability, and retrieve shipping estimates,” which better tells an agent how to help a customer shop with you.
Define your brand with manifest.json
When an agent presents your products in a chat window, it needs to render them with your brand identity, including your logo, name, and visual style.
Agents pull this information from your manifest.json file to create product cards and other UI elements. Use this file to define your official short_name and high-resolution icons. This ensures agents render your brand consistently across every interaction.
For implementation details and copy-paste examples, see MDN Web Docs: Web app manifest—the industry standard documentation.
Instruct agents on how to execute tasks with openapi.yaml
Agents need precise instructions on how to interact with your API, such as which endpoints to call, what parameters to include, and how to structure requests. They rely on your openapi.yaml file to understand exactly how to execute tasks such as searching products, checking inventory, or processing orders.
Just like the description_for_model field, treat your summary and description fields like prompts. Instead of summary: “Get Users”, use summary: “Retrieve a list of active users filtered by subscription tier”. Be specific about capabilities, required parameters, and use cases.
For technical specifications, see OpenAPI Specification (OAS) 3.1.0.
3. Configure your site to handle nonhuman traffic
Agents behave fundamentally differently than human shoppers. They read faster, query more products simultaneously, and operate under strict computational budgets and time-out constraints. If your infrastructure is optimized only for human traffic patterns, agents will struggle to access your catalog or might abandon your site when they encounter slow responses, blocks, or expensive-to-process pages.
Here are a few ways to make sure your site can respond to agentic traffic.
Improve agents’ token efficiency with edge computing logic
Agents “pay” (in compute tokens) to read your website. If your product description is 2,000 words, it can become too expensive for agents to process. As a result, agents might deprioritize or skip your site entirely in favor of leaner, more efficient sources.
To solve this, consider using a modern CDN capable of running edge computing logic. Configure an edge function to inspect the incoming User-Agent header. If an agent (like GPTBot) is detected, intercept the request and serve a lightweight, data-only template (Markdown or JSON) instead of your heavy visual site. This can reduce token cost by approximately 90%.
Set rate limits to prevent “agentic bursts”
Human shoppers browse one product at a time, but agents shopping on behalf of a customer might query 50 products in 2 seconds, checking availability, comparing prices, and reading specifications across your entire catalog nearly instantaneously. A standard firewall might flag this as a DDoS attack and ban the IP.
Configure your Web Application Firewall (WAF) to apply a dedicated rate limit rule for legitimate agents (e.g., GPTBot). Instead of a hard “Block,” set the action to Challenge or return a 429 Too Many Requests status code with a Retry-After header. This explicitly tells the agent, “We allow you here, but please wait five seconds.”
Cache your “read-heavy” API endpoints to prevent time-outs
Agents operate on strict time-out budgets. If your API takes two seconds to respond to a stock check, the agent might time out or hallucinate a failure.
To prevent this, aggressively cache your “read-heavy” API endpoints (such as “price” and “inventory”) at the CDN edge using the stale-while-revalidate directive. This allows the CDN to serve cached data instantly while fetching updates in the background.
Read the RFC 5861: HTTP Cache-Control Extensions for Stale Content, the official specification for asynchronous cache updates.
4. Align your organization around agentic commerce
Preparing for agentic commerce requires changes beyond infrastructure, including how leadership teams align around data and which new roles get created. The businesses moving fastest have made deliberate organizational shifts in leadership priorities, team structure, and cross-departmental collaboration.
Explore departmental shifts
We’re observing shifts within marketing and IT teams at businesses preparing for agentic commerce. Marketing teams are moving from persuasion-focused copy to precision-driven content. These teams are prioritizing structured data and clear product specifications over emotional storytelling.
IT and security teams are focused on managing agent throughput. The most prepared organizations have trained their security teams to differentiate between malicious scrapers and legitimate agents, ensuring valid machine traffic gets prioritized rather than blocked.
Coordinate the C-suite around how data is used
In businesses that have successfully prepared for agentic commerce, we’ve observed a common pattern: alignment between the CIO, CTO, and CDO (chief digital or data officer) around a fundamental shift in how data is used. These organizations have moved beyond viewing data purely as an analytics tool and have started treating it as transactional infrastructure.
- The CIO in these organizations has expanded their focus from internal compliance to external API availability. They’ve recognized that back-office ERP systems—once considered internal-only—now need to be accessible to front-line agents making real-time purchasing decisions.
- The CTO has championed composable architecture as a strategic priority. The most prepared businesses have moved to microservices specifically to support high-velocity agent traffic.
- The CDO has taken ownership of data hygiene as a revenue function. These leaders understand that in agentic commerce, bad data can result in lost sales. They’re responsible for the “grounding truth” that prevents agent hallucinations and ensures accurate product information.
Consider hiring an agentic AI product manager
In addition to product managers for web and mobile, we’re seeing leading businesses define a new role: the agentic AI product manager, who leads development of autonomous systems that can reason, plan, and take action across various channels.
These individuals typically come from backgrounds that blend API engineering with merchandising, and they understand both technical implementation and customer experience. Unlike traditional PMs, they don’t own a visual interface. Instead, they own the agentic experience, testing how agents perceive the brand, optimizing the llms.txt “copy,” and monitoring conversion rates of machine-driven transactions.
5. Establish AI governance
Agentic transactions create additional operational considerations that might require new policies. This might include establishing technical controls for managing fraud and agent access, financial policies for handling edge-case transactions, and guidelines for how first-party agents interact with customers.
Manage new fraud and risk patterns
Agentic commerce is reshaping fraud detection. Fraud signals tuned to human traffic are becoming outdated; AI agents lack human variability and can be misflagged as fraudulent. There’s also the possibility for new fraud patterns, as agents can be manipulated by bad actors to place risky orders or bypass normal guardrails. Without the proper systems in place, businesses accepting agentic payments spend more resources fighting fraud, leading to lost revenue, increased chargebacks, and eroded customer trust.
Many businesses are solving this by using Stripe’s Shared Payment Tokens (SPTs), a new payment primitive for AI commerce. SPTs let agents initiate payments using a buyer’s permission and preferred payment method, without exposing credentials. When used on Stripe, SPTs work seamlessly with Stripe Radar and can relay the underlying risk signals using transaction and payment method details, including the likelihood of a fraudulent dispute, card testing, a stolen card, a card issuer decline, and other signals.
Implement technical controls for agent traffic
Agent behavior can sometimes be unpredictable, either due to bugs in the agent’s logic or how a specific agent interprets your catalog or APIs. Businesses are building granular controls into their infrastructure, so they can respond to these situations without disrupting their entire operation. They’re implementing the ability to revoke specific Shared Payment Tokens or adjust access for specific User-Agent strings without affecting human traffic. This gives them precision and flexibility when managing agent behavior.
Set financial policies for edge cases
Finance and legal teams are establishing policies for handling unusual transactions before they occur. Let’s say an agent processes a transaction with unexpected parameters, like an unusual discount code or shipping destination. Having predefined thresholds (such as automated approval for variances under $50 and manual review above that) helps teams respond consistently rather than making case-by-case decisions under pressure.
Ensure brand consistency in agent interactions
Organizations with first-party agents embedded into their site are testing how their systems respond to unexpected user inputs. This includes defining system prompts and implementing output filters (like Azure AI Content Safety) to ensure agent responses stay aligned with brand guidelines, even when users make unusual or off-brand requests.
Build once, get ready to sell everywhere with Stripe’s Agentic Commerce Suite
The Agentic Commerce Suite enables you to sell on AI agents more easily by making your products discoverable, simplifying your checkout, and allowing you to accept agentic payments via a single integration. Leading brands are already onboarding to the Agentic Commerce Suite, such as URBN (including Anthropologie, Free People, and Urban Outfitters), Etsy, Ashley Furniture, Coach, Kate Spade, Nectar, Revolve, Halara, and Abt Electronics.
The Agentic Commerce Suite allows you to:
Make your products discoverable by agents
The Agentic Commerce Suite provides dedicated hosted ACP endpoints to share near real-time product, price, and availability information with AI agents—with minimal changes required to your existing systems. You can upload your product catalog directly to Stripe or connect your existing product catalog from leading product syndicators. We can then syndicate your product information to each AI agent, enabling you to automatically start taking payments across any supported agent.
Retain control over the customer experience
The Agentic Commerce Suite is powered by Stripe’s Checkout Sessions API, which helps with aspects of the checkout, including shipping and taxes. You can choose to have Stripe manage this on your behalf via built-in Stripe products, such as Stripe Tax, or you can use your existing commerce stack to upload tax codes, manage near real-time inventory checks, and set dynamic shipping rates—with minimal changes to your systems.
Once a customer completes an agentic transaction, you simply use your existing in-house order and fulfillment process. As the merchant of record, you also retain all control over customer relationships, including how refunds and disputes are managed.
Accept agentic payments and protect against emerging fraud patterns
To help protect businesses, the Agentic Commerce Suite handles and processes Shared Payment Tokens (SPTs), a new payment primitive for agentic commerce. Every token can be scoped to a specific seller, bounded by time and amount, and observable throughout its lifecycle to prevent unauthorized agent actions and reduce the likelihood of disputes.
When used on Stripe, SPTs can also be powered by Radar to relay the underlying risk signals using transaction and payment method details that help differentiate between high-intent agents and low-trust automated bots. The result is enterprise-grade fraud protection that works without needing weeks of seller-specific data history.
Get tailored support from Stripe
The shift to agentic commerce is complex. You need to audit your schema, refactor your APIs for composability, and implement new protocols such as MCP and ACP. You do not have to navigate this transition alone; Stripe offers distinct paths to help you execute based on your organizational needs.
The partner ecosystem for end-to-end solutions
Stripe’s certified system integrator (SI) partners can help you get started with agentic commerce. They combine strategic insight with the technical rigor needed to build high-performance integrations tailored to your specific vertical.
Stripe professional services for specialized advisory services
For enterprises seeking direct architectural guidance from the source, Stripe’s professional services team offers focused, technical engagement. We work alongside your engineering teams (or your partners) to ensure your implementation follows best practices for security, scalability, and composability.