Checklist
AI agent readiness checklist
Every check, from robots rules to checkout, with how to test your own site.
AI agents
Agentic commerce: what it is and what to change on your site
AI agents can now search, compare and, on some platforms, buy for the people using them. This guide explains how that works, which standards matter, and what a site owner should change first. It is written by a conversion consultant, not a payments vendor, so it stays on your pages, your data and your checkout.
Short answer
Agentic commerce means AI agents shopping for people: searching, comparing options against the buyer's budget, size or delivery rules and, where platforms and merchants allow it, placing the order and paying. For a merchant, it adds a second kind of visitor, so products, prices, policies and checkout must be readable and usable by software.
- An AI agent takes the buyer's goal and limits, finds candidates, reads the facts, compares and shortlists, then hands over or completes the purchase.
- ACP, UCP and AP2 cover the checkout and payment step; they do not decide whether an agent picks you.
- Start by letting the agents you want reach your pages and putting every buying fact on the page as text.
- Losing a place on an agent's shortlist rarely shows in analytics, so track AI referrals and check server logs for AI user agents.
- Fraud, wrong orders, blocking good agents and weak attribution are the main risks, so start with low-risk changes.
What is agentic commerce?
Agentic commerce is shopping in which an AI agent acts for a buyer: it searches for products or services, compares them against the buyer's rules, such as budget, size or delivery date, and, where the platform and merchant allow it, places the order and pays. The person sets the goal and approves; the agent does the legwork.
It sits one step beyond the AI shopping assistants most people have used. An assistant answers questions and suggests options. An agent acts: it opens pages, reads prices and policies, fills in forms and, on platforms that support it, completes checkout. For a site owner, the practical change is simple to state and harder to act on. Some of the visitors deciding whether to buy from you are now software, and they read your site differently from people.
How agentic commerce works, step by step
An AI agent buying for someone follows five steps: it takes the goal and limits from the person, finds candidate products, reads the facts it needs, compares and shortlists, then hands over or completes the purchase. Your site can win or lose the sale at every step, usually without anyone at your company seeing it happen.
-
The person sets the goal and the limits
"Find a carry-on suitcase under a set budget that fits this airline's size rules and arrives by Friday." The rules the agent works to are the buyer's, not your marketing's.
-
The agent finds candidates
It uses search results, the AI platform's own index, merchant feeds and sometimes marketplaces. A product that is not crawlable, not in a feed or blocked by bot protection may never make the list.
-
It reads the facts
Name, price, currency, variants, stock, delivery cost and time, returns, warranty. Agents read the page HTML, the accessibility tree, screenshots and structured data. Facts that are missing, hidden in images or only shown after a click count against you.
-
It compares and shortlists
The agent checks each option against the buyer's rules. Unclear shipping costs or a returns policy it cannot find are reasons to drop you, in the same way a person abandons a cart.
-
It hands over or completes the purchase
On platforms that support agent checkout, the agent pays through a protocol with the buyer's approval. Elsewhere it sends the person to your page to finish, or tries to complete your checkout in a browser. Either way, your checkout and forms now have two kinds of user.
Agentic commerce vs traditional ecommerce
In traditional ecommerce a person browses, compares and pays. In agentic commerce an AI agent does some or all of that for them. The products and the checkout may be the same; what changes is who reads your pages and what they need from them.
| Traditional ecommerce | With AI agents | |
|---|---|---|
| Who browses | The buyer, page by page | An agent, often across many stores at once |
| What persuades | Design, images, copy, reviews, offers | Facts the agent can read and check against the buyer's rules |
| Where the decision happens | On your site | Often inside the AI assistant, before anyone visits you |
| Checkout | Your cart and payment page | Your checkout, a browser agent using it, or a protocol on the AI platform |
| How you lose the sale | A bounce or an abandoned cart you can see in analytics | Being left off the shortlist, which you usually cannot see |
| What to fix | Clarity, trust, friction, speed | All of that, plus access, readable facts and accurate structured data |
The agentic commerce protocols: ACP, UCP and AP2
Three open protocols now define how AI agents check out and pay: the Agentic Commerce Protocol (ACP) from OpenAI and Stripe, the Universal Commerce Protocol (UCP) from Google and retail partners, and Google's Agent Payments Protocol (AP2). They cover the checkout and payment step. They do not decide whether an agent picks you in the first place.
| Protocol | From | What it covers | What it means for a merchant |
|---|---|---|---|
| Agentic Commerce Protocol (ACP) | Developed by Stripe and OpenAI; open source under the Apache 2.0 license | Checkout between buyers, AI agents and businesses, including securely passing payment credentials | You stay the merchant of record. OpenAI is the first platform to implement it, in ChatGPT, and each platform runs its own process for which businesses take part |
| Universal Commerce Protocol (UCP) | Google, with Shopify, Etsy, Wayfair, Target and Walmart; announced 11 January 2026 | The commerce journey from discovery to checkout and order management, through APIs, Agent2Agent (A2A) or the Model Context Protocol (MCP) | Lets businesses sell through AI Mode in Google Search and the Gemini app while remaining the merchant of record |
| Agent Payments Protocol (AP2) | Google; announced 17 September 2025 | Payments started by agents, using signed "mandates" as proof of what the user asked for and approved | Mostly handled by payment providers. UCP is built to work with it |
For most site owners, the protocol decision is made with your ecommerce platform or payment provider, not by your own developers from scratch. Ask them which protocols they support and what you need to switch on. The work that is fully in your hands, and that matters whichever protocol wins, is the subject of the rest of this page: letting agents in, and giving them facts they can read and trust.
How an AI agent reads a product page
An agent reads your page as data: the HTML your server sends, the accessibility tree, structured data and sometimes a screenshot. A page that looks complete to a person can be missing the very facts an agent needs to recommend or buy.
PERSON VIEW
Product page
Trail running shoe
$129
Choose your size. Free returns (see details).
Illustrative example
AGENT VIEW
- product.name
- Trail running shoe
- PASS
- offers.price
- 129.00
- PASS
- offers.availability
- InStock
- PASS
- size
- null
- FAIL: SCRIPT-ONLY PICKER
- shippingDetails
- null
- MISSING
- hasMerchantReturnPolicy
- null
- FAIL: POPUP ONLY
Illustrative example: the person can pick a size and open the returns popup. An agent reading the page's data finds the price but cannot confirm sizes, shipping or returns, so it may drop this store from a buyer's shortlist.
What to change on your site for agentic commerce
To prepare a site for AI agents, work in this order: let the agents you want reach your pages, put every buying fact in plain text on the page, make structured data match it, keep feeds current, make policies quotable, make checkout usable by software, settle protocols with your platform, and measure. The first two come first because nothing else works without them.
-
Let the right agents in
Check your robots.txt and, often overlooked, your CDN or firewall bot settings. AI companies publish their crawler names: OpenAI, for example, documents OAI-SearchBot for search and ChatGPT-User for actions a person asks for. Decide which you allow, and make sure your bot protection is not quietly blocking them.
-
Put every buying fact on the page as text
Name, price and currency, variants, stock, delivery cost and time, returns and warranty, in the HTML your server sends. Not in an image, not only in a popup, not only after a script runs. Missing facts cost you human buyers too: a shipping cost nobody can find is a classic reason to abandon a cart.
-
Make structured data match what people see
Product and Offer markup with price, currency and availability, plus shipping details and a merchant return policy. It must agree with the visible page. A price in the markup that differs from the price on screen is worse than no markup.
-
Keep product feeds current
Agents and AI search lean on merchant feeds. Google's own guidance for its AI features points merchants to Merchant Center and Business Profile data. Out-of-date stock or prices in a feed send agents away, or send them in with the wrong expectations.
-
Write policies an agent can quote
Shipping, returns, guarantees and contact details as short, plain statements on a crawlable page, linked from each product page. "Free returns within 30 days of delivery" is quotable; "hassle-free returns" is not.
-
Make checkout and forms usable by software
Real buttons and links, labels tied to their fields, no invisible overlays, a stable layout, clear error messages and a guest checkout. Google's web.dev guidance for agent-friendly sites lists the same basics, and every one of them also helps people using assistive technology.
-
Decide on protocols with your platform
Ask your ecommerce platform and payment provider whether they support ACP or UCP and what switching on costs you in control. You remain the merchant of record under both.
-
Measure what you can
Track referral visits from AI assistants, check server logs for AI user agents, and use the generative AI performance report in Google Search Console. Then treat agent traffic like any other segment: does it convert, and where does it drop?
Who agentic commerce affects first
Agentic commerce affects first the purchases that are easy to describe as rules: repeat orders, products compared on specifications, and quotes gathered from several providers. How it shows up differs by business model.
| Type of site | What an agent does | What to check first |
|---|---|---|
| Online stores | Compares products, prices, delivery and returns across stores; on supported platforms, checks out | Product facts in HTML, structured data, feeds, return policy |
| Lead generation and services | Shortlists providers, reads pricing rules and service areas, fills in quote forms | Plain-text service descriptions, how pricing works, form labels and errors |
| SaaS | Compares plans and features, checks limits and terms, starts trials | Plan details as text rather than only in a comparison graphic, sign-up flow |
| Affiliate and content sites | Reads reviews and comparisons to build its answer, sometimes without a visit | Original testing and data an agent cannot get elsewhere, clear sourcing |
What agentic commerce means for conversion rate optimization
Agentic commerce adds a second kind of visitor to conversion rate optimization. CRO has always asked why people leave without buying. Now it also has to ask why an AI agent left you off a shortlist, which is harder, because that loss rarely shows up in analytics as a bounce or an abandoned cart.
The good news is that most fixes overlap. A price that is visible, a returns policy stated plainly, a form whose fields are labeled and a checkout that does not jump around help a person on a phone and an agent reading the page. Convertica's positioning, "Conversion for humans and AI agents.", comes from that overlap. What is new is the order of priorities: access and readable facts move to the top of the list.
Two cautions from experience. First, do not strip persuasion out of your pages to suit software: people still make or approve most purchases. Second, measure before you rebuild. Look at how much traffic arrives from AI assistants and how it converts before you put a large project behind it. For the other meaning of the phrase, using AI tools to do CRO work, see AI conversion rate optimization.
Agentic commerce risks and limits
The main risks of agentic commerce for merchants are fraud and authorization, wrong orders and returns, blocking good agents along with bad bots, and weak attribution. None is a reason to wait, but each is a reason to start with low-risk changes.
- Authorization and fraud. A payment started by software raises the question of who approved it. AP2's signed mandates and the token-based payments in ACP and UCP exist to answer it. Ask your payment provider how they handle agent transactions and disputes.
- Wrong orders. An agent working from vague product data can buy the wrong size or variant. Clear variant data is cheaper than the returns.
- Blocking the wrong bots. Bot protection tuned for scrapers can also turn away the agents you want. Review what your firewall blocks, not only your robots.txt.
- Attribution. A sale influenced inside an AI assistant may arrive as direct traffic, or never visit at all before checkout. Expect your reports to undercount agent influence.
- Moving standards. Protocols, crawler names and platform programs are changing month to month. This is a young field with little long-term data, including Convertica's, and we say so.
How convertica.org is built for AI agents
Convertica's own site follows the advice on this page. The checks below are things you can verify yourself by viewing the source or fetching the files.
- Everything is in the HTML on load. Pages are static HTML. Prices, offers, FAQs and policies do not need a script to appear; FAQ answers sit in the page, not behind a request.
- AI crawlers are named and allowed. robots.txt names OpenAI's, Anthropic's and Perplexity's crawlers and Google-Extended, allows them, and excludes only the sign-up endpoint and an internal style guide.
- An llms.txt that cannot go stale. /llms.txt is generated from the same data as the pages, and lists a page only once it exists. Read our llms.txt guide for how and why.
- One structured data graph per page, built from the visible content. FAQ markup comes from the same data as the FAQ you can read. There is no review rating markup, and no price markup, because no price is shown.
- Prices an agent can read honestly. Convertica publishes no price figures. The pricing rule is stated in plain text on every relevant page and in llms.txt: the CRO audit is free, and CRO advisory is priced after your audit.
In this section
Guides on conversion for AI agents
Practical guides for site owners. Each one covers what to change, how to check it, and what not to waste time on.
Guide
llms.txt
What the file is, how to write one, and an honest answer on whether you need it.
Guide
AI conversion rate optimization
Where AI tools help CRO work, where they do not, and how to use them safely.
Service
Generative engine optimization services
Convertica's AI search visibility service, sold through CRO advisory.
Guide
AI shopping assistants
How shopping assistants on your site and on AI platforms affect conversion.
Guide
AI ecommerce personalization
What AI personalization can do for a store, and how to test it.
Guide
AI for A/B testing
Using AI to plan, build and read A/B tests without fooling yourself.
Get help preparing your site for AI agents
Start with the free CRO audit: an app with eight checks for people and for AI agents, including agent readiness. Enter your website and email, and watch it run live on your page. If you want help acting on it, CRO advisory is led personally by Kurt Philip, with Convertica's CRO team, and is priced after your audit. Your team makes the changes; we help decide which ones matter. If you would rather not build them, full implementation has Convertica's team build the fixes from your audit. AI search visibility is part of that work too: see generative engine optimization services, and how pricing works.
Agentic commerce FAQ
What is agentic commerce?
Agentic commerce is shopping in which an AI agent acts for a person or a business: it searches, compares options against the buyer's rules and, where the merchant and platform allow it, places the order and pays. The person sets the goal and the limits. The agent does the browsing, and sometimes the buying.
What is an example of agentic commerce?
A person asks an AI assistant to find running shoes in their size under a set budget with free returns. The agent reads product pages, feeds and policies from several stores, shortlists two pairs and, on a platform that supports checkout, completes the purchase after the person approves it. The store that wins is the one whose facts the agent could read and trust.
What is the difference between agentic AI and agentic commerce?
Agentic AI is the general term for AI systems that plan and take actions toward a goal. Agentic commerce is one use of it: AI agents that shop, compare and buy. A coding agent or a support agent is agentic AI but not commerce.
Do I need to implement ACP or UCP?
Not as a first step. The Agentic Commerce Protocol (from OpenAI and Stripe) and the Universal Commerce Protocol (from Google and retail partners) handle the checkout step on the platforms that support them, and your ecommerce platform or payment provider usually decides how you connect. Before that, make sure agents can reach your pages and read your products, prices, stock and policies.
Do AI agents need an llms.txt file?
No. llms.txt is an optional Markdown summary of a site for AI tools. Google says its Search features do not use it. It is cheap to publish and some tools check for it, but the facts on your pages and in your structured data matter far more to an agent deciding where to buy.
How do I know if AI agents visit my site?
Check your server or CDN logs for the user agents AI companies publish, such as OAI-SearchBot and ChatGPT-User from OpenAI, and look for referral visits from AI assistants in your analytics. Google Search Console has a generative AI performance report for Google's AI features. None of these shows every agent, so treat them as a floor.
Will agentic commerce replace my website?
No. Agents still read websites, feeds and policies to decide, and many purchases still end on the merchant's own checkout. What changes is who reads your pages: some of your visitors are now software acting for a buyer, and they need plain facts rather than persuasion alone.
Does the free CRO audit check how AI agents read my site?
Yes. AI agent readiness, whether an agent can read what a page sells, the price and the policies, is one of the eight checks in Convertica's free CRO audit, an app you can run now. In CRO advisory, Kurt Philip reviews it with you.