The 2026 agentic checkout readiness checklist for store owners: a practical find/read/buy audit to learn if your store is agent ready, plus tiered fixes at $0, $50, $500.
AI shopping agents have started finishing purchases, not just recommending them. The enterprise vendors wrote the readiness memo for CTOs. Nobody wrote the plain version for the person running a $50-a-month store on Shopify or WooCommerce. This is that version: a checklist you can run today, scored honestly, with a tiered plan for what to fix at $0, $50, and $500 a month. The goal is not to chase hype. It is to answer one question with evidence instead of vibes: when a shopping agent shows up ready to buy, can your store actually sell to it?
Agentic checkout is when an AI agent completes a purchase for a person, handling the browse, the choice, and the payment without the human touching your checkout flow. A few agents can do pieces of this right now. Visa announced at its June 2026 Payments Forum that users can link a Visa card inside ChatGPT, which lets an agent pay at any Visa-accepting merchant within set limits. Perplexity runs shopping-style answers that surface specific products with buy paths. Chrome's auto-browse operates pages by driving the DOM the way a person would, clicking and typing through a real checkout. Read that precisely: the capability exists and is shipping in pieces, but broad consumer adoption is not here yet. What is announced is not always generally available, and Visa has not made every rollout detail public. The honest framing is "early and accelerating," not "everywhere."
Research from Checkout.com found that a majority of merchants expect consumer adoption of AI shopping to outpace their own readiness. That gap has a precise shape, and it is not the one most owners assume. The failure is rarely "an agent can't read my site." Modern store platforms produce decent HTML, so an agent can usually parse your product name and a price. The failure is the second half: an agent that can read you still can't reliably buy from you.
Picture the agent's job as a relay with a baton. It has to find your product among competitors, read your page well enough to trust the price and stock, then drive your checkout to completion. Drop the baton at any leg and the sale is gone, silently. No abandoned-cart email fires, because there was never a session your tooling recognized as a shopper. This is the agentic readiness gap: the distance between legible and transactable. Your competitor who closed that distance gets the order, and nothing in your analytics tells you why you lost it.
So the checklist below is organized by the three legs of that relay. Work them in order. Discovery first, because comprehension is wasted on a page no agent reached. Comprehension second, because a flawless checkout is wasted on a price the agent couldn't confirm.
Discovery is the first leg, and it is mostly about being legible to the systems that build shopping answers. Four checks:
1. Clean, crawlable product pages. Each product lives at a stable URL with its name, price, and key specs present in the page, not buried behind a tab that only loads on click. If your catalog is a single JavaScript app that paints products after load, a crawler-style agent may see an empty shell. View source and confirm the product text is actually there.
2. A product feed the engines can ingest. The AI shopping surfaces lean on structured product data, often the same Merchant Center feed that powers Google Shopping. If you sell on Shopify or WooCommerce, you likely have a feed already. Keep it current on price and stock. Our ChatGPT shopping product feed guide walks the feed setup end to end.
3. An llms.txt file acting as a site map for agents. A markdown index at your root that points to your catalog, bestsellers, shipping, and returns saves the agent from guessing which of your thousands of URLs matter. It is one file and costs nothing.
4. Discovery on the answer engines themselves. Perplexity runs a merchant program that shapes how products appear in its shopping answers. Getting listed there is a discovery lever, not a payment one. Our Perplexity merchant program guide covers eligibility and setup.
Reading is the comprehension leg. An agent reached your page; now it has to turn that page into facts it trusts enough to act on. Four more checks:
5. Bot access in robots.txt. Confirm you are not accidentally blocking the agents that fetch pages live. The user agents that matter for shopping answers include OAI-SearchBot and ChatGPT-User from OpenAI, PerplexityBot, and Claude-User. A blanket Disallow from an old SEO config or a security plugin can quietly remove you from the pool. This is the most common own-goal we see.
6. Server-rendered price and availability. Price and stock must exist in the HTML the server sends, not appear only after client-side JavaScript runs. A crawler-style agent that can't confirm your price will not guess it. It moves to a competitor whose number was right there in the markup.
7. schema.org/Product structured data. Product, Offer, and AggregateRating markup hand the agent the exact fields it needs: name, price, currency, availability. This is the difference between an agent inferring your price from page text and reading it from a labeled field. Most store platforms can emit it; many themes do it badly. Validate yours.
8. Markdown-friendly, cleanly structured content. Clear headings, real lists, and labeled specs parse better than dense marketing prose and image-only spec sheets. If your size chart or ingredient list is a JPEG, the agent can't read it. Put the facts in text.
Buying is the hardest leg, and the one most stores never test. Reaching this stage means an agent has to act on your store, not just read it. Four final checks, with an honest note on what's live versus forward-looking:
9. A checkout a non-human operator can complete. Browser-driving agents like Chrome auto-browse finish purchases by operating your DOM, filling fields and clicking buttons the way a person does. That only works if your checkout is accessible to a machine: labeled form fields, buttons that are real buttons, no step that depends on a hover or a hard-to-target widget. This is the leg WooCommerce stores most often stumble on, and we cover the fixes in our WooCommerce ChatGPT shopping guide. The deeper, field-by-field version is the DOM-level transactability audit.
10. A payment rail an agent can actually use. This is the leg that changed in 2026. Visa's card-linking inside ChatGPT means that if you accept Visa, you sit inside the addressable pool for those purchases without signing up for anything. Stripe has also been shipping tooling aimed at agent-initiated payments. The practical takeaway: you probably don't need new payment infrastructure, you need a checkout the agent can reach with the credential it already holds.
11. A structured sell layer, looking forward. WebMCP lets a store expose products and checkout as callable tools, so an agent queries your catalog directly instead of scraping rendered pages. Be precise about timing: WebMCP is in the origin-trial era. Browser-native agents and some extensions invoke it, but ChatGPT and Claude do not broadly call site WebMCP tools today. Shipping the snippet is a forward investment that no-ops harmlessly on agents that don't support it yet, not a switch that lights up every assistant this week.
12. Confirmation and guardrails. An agent-completed order should still respect the controls a human order does: clear totals before commit, inventory checks, and a confirmation the buyer's agent can read back. The payment networks add their own limits and approval steps, but your store should not assume an agent will tolerate a surprise fee revealed three steps in. Surprises break trust and abandon carts, human or not.
Twelve checks, three groups of four. Give yourself one point for each item you pass cleanly, zero for partial. Then read the shape, not just the total:
Find (items 1-4). A low score here means agents may never reach you, so nothing downstream matters yet. Start at the top.
Read (items 5-8). A low score here is the most common and the most fixable. These are mostly free, mostly fast, and they unblock the majority of agent purchases.
Buy (items 9-12). A low score here is where ambition lives. Don't invest in the sell layer before find and read are solid, or you'll be building a fast lane onto a road no one can find.
A rough reading: 10-12 means you're ahead of your category. 6-9 means you're legible but losing buyable moments. Below 6 means an agent that wants to buy from you probably can't, and the fixes are cheaper than you think. The point of scoring is sequencing. You want to know which leg of the relay drops the baton first.
Readiness is not a single product you buy. It is a set of fixes that ladder up by budget. Here is the honest tiering.
$0 a month. Most of the find and read checklist is free. Open up robots.txt for the shopping agents. Publish an llms.txt index. Add or fix schema.org/Product markup. Make sure price and stock are in the server-rendered HTML. These cost time, not money, and they close the largest share of the readiness gap. If you do nothing else, do these. The free Agent-Ready Grader tells you which of them you're currently failing.
~$50 a month. This tier buys you eyes and a structured sell layer. Monitoring shows whether agents are actually hitting your product pages, which user agents, and how often, data that Google Analytics filters out by design. Attribution starts connecting AI-referred sessions to outcomes. Crawlytics sits squarely in this band: its Commerce tier at $49.99 a month covers AI bot tracking, traffic attribution, and the WebMCP commerce snippet that exposes your products as agent-callable tools. Honest framing: at this tier you're measuring readiness and shipping the forward-looking sell layer, not buying a guarantee that every assistant transacts with you tomorrow.
~$500 a month and up. This is full agent-commerce tooling: deeper feed management, custom integration work, dedicated payment-agent flows, and the engineering time to maintain a checkout that's continuously tested against browser-driving agents. Most small stores don't need this tier in 2026. It's where a high-volume merchant goes once agent-driven revenue is a measurable line in the P&L, not a bet.
The discipline across all three tiers is the same: fix in order, find then read then buy, and don't spend at a higher tier until the free fixes below it are done. An agent that wants to buy from you in 2026 is a cheap problem to solve. The expensive mistake is not knowing whether one already tried.
Written by Crawlytics Team. Crawlytics tracks AI bots, generates llms.txt, and powers WebMCP commerce, all from one snippet on any stack. See how it works →
Discovery is the first leg, and it is mostly about being legible to the systems that build shopping answers. Four checks: 1. Clean, crawlable product pages. Each product lives at a stable URL with its name, price, and key specs present in the page, not buried behind a tab that only loads on click. If your catalog is a single JavaScript app that paints products after load, a crawler-style agent may see an empty shell. View source and confirm the product text is actually there. 2. A product feed the engines can ingest. The AI shopping surfaces lean on structured product data, often the same Merchant Center feed that powers Google Shopping. If you sell on Shopify or WooCommerce, you likely have a feed already. Keep it current on price and stock. Our ChatGPT shopping product feed guide walks the feed setup end to end. 3. An llms.txt file acting as a site map for agents. A markdown index at your root that points to your catalog, bestsellers, shipping, and returns saves the agent from guessing which of your thousands of URLs matter. It is one file and costs nothing. 4. Discovery on the answer engines themselves. Perplexity runs a merchant program that shapes how products appear in its sh
Reading is the comprehension leg. An agent reached your page; now it has to turn that page into facts it trusts enough to act on. Four more checks: 5. Bot access in robots.txt. Confirm you are not accidentally blocking the agents that fetch pages live. The user agents that matter for shopping answers include OAI-SearchBot and ChatGPT-User from OpenAI, PerplexityBot, and Claude-User. A blanket Disallow from an old SEO config or a security plugin can quietly remove you from the pool. This is the most common own-goal we see. 6. Server-rendered price and availability. Price and stock must exist in the HTML the server sends, not appear only after client-side JavaScript runs. A crawler-style agent that can't confirm your price will not guess it. It moves to a competitor whose number was right there in the markup. 7. schema.org/Product structured data. Product, Offer, and AggregateRating markup hand the agent the exact fields it needs: name, price, currency, availability. This is the difference between an agent inferring your price from page text and reading it from a labeled field. Most store platforms can emit it; many themes do it badly. Validate yours. 8. Markdown-friendly, cleanly st
Buying is the hardest leg, and the one most stores never test. Reaching this stage means an agent has to act on your store, not just read it. Four final checks, with an honest note on what's live versus forward-looking: 9. A checkout a non-human operator can complete. Browser-driving agents like Chrome auto-browse finish purchases by operating your DOM, filling fields and clicking buttons the way a person does. That only works if your checkout is accessible to a machine: labeled form fields, buttons that are real buttons, no step that depends on a hover or a hard-to-target widget. This is the leg WooCommerce stores most often stumble on, and we cover the fixes in our WooCommerce ChatGPT shopping guide. The deeper, field-by-field version is the DOM-level transactability audit. 10. A payment rail an agent can actually use. This is the leg that changed in 2026. Visa's card-linking inside ChatGPT means that if you accept Visa, you sit inside the addressable pool for those purchases without signing up for anything. Stripe has also been shipping tooling aimed at agent-initiated payments. The practical takeaway: you probably don't need new payment infrastructure, you need a checkout the a
Agentic checkout is when an AI agent completes a purchase on a person's behalf, handling the browsing, product choice, and payment without the human clicking through the store's checkout themselves. The person delegates a task like "order more of my usual coffee," and the agent finds the product, confirms price and availability, and pays using a linked credential within limits the person set. It differs from normal online shopping in who drives the funnel: a machine operator instead of a human one, which means your store has to be readable and operable by software, not just by a person with a mouse.
A handful, in pieces, as of mid-2026. Visa announced card-linking inside ChatGPT, which lets an agent pay at Visa-accepting merchants within set limits. Perplexity runs shopping answers that surface specific products. Chrome's auto-browse completes checkouts by driving the page DOM directly. The precise read: these are real capabilities shipping in stages, not a universal feature every shopper uses yet. Rollout details, including regions and whether some are pilots, are not all public. Treat it as early and accelerating rather than fully arrived.
Yes, in specific configurations, though total volume is still small. A browser-driving agent like Chrome auto-browse can complete a checkout end to end when the flow is accessible to a non-human operator. An agent with a linked Visa card in ChatGPT can authorize payment at accepting merchants within limits. What's not yet here is ubiquity: the share of all purchases completed by agents is low, and clean public conversion numbers are scarce. The honest position is that the capability is proven and the footprint is growing, which is exactly why readiness is worth doing now while it's cheap.
Check that your robots.txt isn't blocking the shopping agents, then confirm your price and stock are in the server-rendered HTML. The first is a one-line review that prevents a silent own-goal: an old SEO rule or security plugin disallowing OAI-SearchBot, ChatGPT-User, or PerplexityBot removes you from the pool entirely. The second makes sure that when an agent does reach you, it can read the two facts it needs to act. Both are free and take minutes. Add an llms.txt index and schema.org/Product markup right after for the next biggest gains.
Yes, because the safety controls live mostly on the payment and agent side, not yours. Visa's integration adds spending limits and approval steps; the buyer's agent operates within constraints the person set. For your store, the main job is making sure an agent-completed order respects the same guardrails a human one does: clear totals before commit, real inventory checks, and no surprise fees revealed late. Opening robots.txt to shopping agents and adding structured data doesn't expose you to new risk, it just makes your existing products legible. The WebMCP snippet no-ops on agents that don't support it, so there's no downside to shipping it early.
This page is part of Crawlytics.app. View all pages: llms.txt · llms-full.txt