Agentic Commerce Checklist: 40 Readiness Items
Forty checks across your feed, schema, policies, reviews and tracking that decide whether ChatGPT, Gemini and Copilot can read your store, pick your product, and send you a buyer who actually pays.
Hot take: most stores that "aren't showing up in ChatGPT" don't have an AI problem. They have a spreadsheet problem.
This agentic commerce checklist is 40 checks in five groups: your product feed, your schema markup, your policies, your reviews, and your tracking. Get them right and an AI shopping agent can read your catalog, trust your price and return terms, compare you fairly against the next brand, and send a buyer to a checkout that works. Get them wrong and the agent doesn't rank you lower. It just skips you, and you never find out why.
As of October 2026, here's where things stand. On Shopify, agentic storefronts are on by default for eligible stores across ChatGPT, Google AI Mode, Gemini, Microsoft Copilot and Meta. ChatGPT finds your product and sends the buyer to your own checkout. Google AI Mode, Gemini, Copilot and Meta can close the sale inside the chat if direct checkout is on.
Every item below links back to that reality. No speculation, no "the future of shopping" fog. Just the list, the reason each item exists, and the source I checked this week.
What does an AI shopping agent actually need from your store?
Five things, in this order.
It needs to find the product (feed and catalog eligibility). It needs to read it without guessing (schema and clean attributes). It needs to trust the offer (price, shipping, returns written as plain facts). It needs proof that real people bought it and liked it (reviews and ratings). And you need to see what happened after it sent someone over (tracking).
Here's the thing most founders miss. An agent is a very fast, very literal shopper with zero patience for "see details in tab three." A human will scroll and hunt. An agent reads fields. If your return window lives inside a paragraph of brand-story copy, to the agent it doesn't exist.
One Shopify merchant in the community forum summed it up in six words: product data becomes your storefront. Another, in the same thread, complained that the useful data in most catalogs is buried in undocumented metafields. That's the gap this list closes.
An agent doesn't rank a messy product lower. It skips it. Silence is the penalty, and nobody emails you about it.
If you want the deeper story on why agents skip rather than demote, I broke down the 8 data gaps that get products skipped. This post is the working list. Print it, hand it to whoever runs your catalog, and tick boxes.
Where can an agent sell your product right now?
Before the checklist, the map. This changes monthly, so I re-checked every row against the vendor's own page this week.
| AI channel | What the agent does today | Where the buyer pays | Source |
|---|---|---|---|
| ChatGPT | Finds and recommends products from Shopify Catalog | Your store checkout, in an in-app browser on mobile or a new tab on web; no fee beyond processing | Shopify ChatGPT channel |
| Google AI Mode and Gemini | Finds products, and can check out in the chat | Shopify-powered direct checkout inside the conversation, on by default for eligible US stores | Shopify Google channel |
| Microsoft Copilot and Meta | Finds products, and can check out if you activate it | Shopify-powered direct checkout in the channel when activated | Shopify agentic storefronts |
| Google Search with UCP | Buy button on eligible listings, paid through Google Pay | Google Wallet details, merchant tagged with the native_commerce attribute; expanded to main search results on May 5, 2026 | PPC Land |
| Google price tracking agent | Watches a price, then buys on the merchant's site with Google Pay | Your site; launched November 2025 with Wayfair, Chewy, Quince and select Shopify merchants |
Two protocols sit under all of this. OpenAI's Agentic Commerce Protocol says it plainly: start by sharing a structured product feed. The Universal Commerce Protocol, co-developed by Google, Shopify, Walmart, Target, Etsy and others, covers catalog search, multi-item carts, checkout, account linking and order updates. If you're on Shopify, you don't build either one. Shopify Catalog speaks both for you. The full comparison is in Agentic Commerce Protocol vs Universal Commerce Protocol, and the backstory on why ChatGPT stopped selling inside the chat is in what happened to ChatGPT Instant Checkout.
So what's your job? Feed the plumbing clean water. That's the 40 items.
Group 1: Is your product feed readable by every agent? (items 1 to 8)
Every agent starts at the feed. OpenAI's feed spec lists nine required fields: item_id, title, description, url, brand, seller_name, image_url, availability and price. Shopify fills most of these from your admin, which is great news and also a trap, because it fills them with whatever you typed in 2021.
Every product passes Shopify Catalog eligibility. Your store is on the Starter plan or higher, each product has a title, at least one image and a price above $0, and it's published to your online store. Products with Unlisted status or hidden from search engines are out. All of that is on Shopify's Catalog requirements page. Fail this and nothing else on the list matters for that product.
Your account is in good standing. Same page: verified email, two-step authentication on, invoices paid, a low chargeback rate, orders fulfilled promptly. Sloppy operations can quietly pull your whole catalog.
Titles carry the attributes a shopper would type. Brand, product type, the one or two attributes that decide the purchase (size, material, use case). OpenAI caps feed titles at 150 characters. "The Cloud" tells an agent nothing. "Cloud Cooling Bamboo Sheet Set, Queen, 300 Thread Count" tells it everything. My full method is in how to write a Shopify product page title.
Descriptions state facts first, story second. OpenAI's best practices ask for concise, factual copy. Lead with what it is, what it's made of, who it's for and what's in the box. Keep the brand voice, but put it after the facts.
Every product has a barcode where one exists. Shopify's Catalog guide calls out barcode data such as UPC and GTIN for product matching. OpenAI's spec wants a valid 8, 12, 13 or 14 digit GTIN with a check digit. No barcode? Leave it blank. Never fake one.
Variants are real variants with named options. One stable parent id, a unique id per purchasable option, and option names like "Size" and "Color" filled in. When a variant has its own price, image or stock, the feed should show it. This is where catalogs with 40 colorways fall apart; I covered the page side in writing pages for products with many variants.
Type, Vendor, Collections and Tags are filled in and consistent. Shopify lists these classification fields as signals AI channels use to categorize products. "Misc" as a product type is a vote for invisibility.
Availability is true right now. An agent that recommends an out-of-stock item, or a buyer who lands on "sold out," burns trust for both of you. If you run a Google Merchant Center feed too, keep it error-free; Shopify's Google AI Mode channel requires your products to be in Merchant Center. My feed teardown is in the Google Shopping feed guide.
Your feed is a sales rep who never sleeps and never improvises. It says exactly what you typed. Nothing more, nothing less.
Group 2: Can an agent read your product page without guessing? (items 9 to 16)
Feeds get you into the catalog. Your page markup is what an agent or a crawler reads when it visits the page directly, and what Google uses for merchant listings. Here's the list.
Product structured data on every product page, with the required fields. Google's merchant listing documentation requires name, image and an offer with price and priceCurrency. Most Shopify themes output this. Check yours with Google's Rich Results Test anyway; I've seen themes ship with price missing after a customization.
Offers carry availability and itemCondition. Both are recommended properties. An agent comparing three "in stock" products will pass on the one that doesn't say.
shippingDetails is filled in. Google lists it as recommended: shipping rates and delivery times in markup. When a shopper asks for something "that arrives by Friday," this is the field that answers.
Your return policy is marked up at the Organization level. Google recommends nesting MerchantReturnPolicy under Organization and using product-level policies only to override the default. One policy, one place, every product covered.
aggregateRating and review markup match what shoppers see. Recommended properties, and agents lean on them. The number in your markup should match the stars on your page. Mismatches look like manipulation.
gtin and brand appear in markup too. Same identifiers as the feed. Consistency across feed, markup and page is the trust signal.
OAI-SearchBot is allowed in robots.txt. OpenAI's crawler page is blunt: sites that block OAI-SearchBot will not be shown in ChatGPT search answers. GPTBot is the training crawler and is a separate choice. Many security apps block both by default. Check yours.
Key facts live in the HTML, not only in images or tabs. Ingredient panels as JPEGs, sizing charts as PNGs, warranty terms behind a click-to-open widget that loads late. A human sees them. A crawler often doesn't. Put the facts in text on the page.
A quick word on llms.txt, since someone always asks. It's cheap to add, and I don't think it's on the critical path for product discovery yet. My test notes are in whether llms.txt Shopify apps help AI visibility.
Group 3: Do your policies answer the agent's questions before it asks? (items 17 to 24)
Policies feel like legal housekeeping. To an agent, they're product attributes. Shopify's ChatGPT channel requires three policies to be complete: terms of service, privacy policy, and return and refund policy. The Google channel requires the same. Missing one can switch the channel off for your store.
Terms of service, privacy policy and return and refund policy are all filled in. In your Shopify admin, under Settings, Policies. Not the default template with "[Your store name]" still in it.
Your return window is a number. "30 days from delivery." OpenAI's feed spec has fields for accepts_returns and return_deadline_in_days. An agent can't parse "hassle-free returns" into a deadline.
Who pays return shipping is stated. Buyers ask. Agents get asked. Say it in one sentence.
Final-sale items are flagged on the product. If a product can't be returned, say so on that product, not only in the policy page. OpenAI's spec gives return_policy a URL to "return/final-sale terms" for this reason.
Shipping cost and delivery time are written as numbers. "$6.95 flat, free over $75, 3 to 5 business days in the US." This feeds item 11 and answers the most common follow-up question an agent gets.
Your FAQ answers the questions AI channels pull from. Shopify suggests using its Knowledge Base app so AI platforms have FAQs to reference when they answer shoppers. Write the five questions your support inbox gets every week. Answer each in two sentences.
Warranty and guarantee terms are specific. "Two-year warranty against manufacturing defects." Vague promises read as no promise.
Your hero product fits the in-chat checkout. This one surprised me. Shopify says Google AI Mode and Gemini direct checkout doesn't support subscriptions, product bundles, customizable products or local delivery. If your best seller is a subscribe-and-save bundle, the buyer gets routed to your store instead. That's fine, as long as your page is ready for them. Know which path each hero product takes.
I'll be honest. Policies are the item group I see skipped most, because they feel boring. But here's the math on boring: one missing policy can take a store out of a channel entirely. Zero visitors times any conversion rate is zero.
If you're worried that a generous return policy hurts margin, read the return policy conversion myth first. Clear beats generous.
Group 4: Do your reviews give the agent proof it can quote? (items 25 to 32)
Agents compare. When a shopper asks for "the best cooling sheets for hot sleepers," the agent needs a reason to name you over the brand with 4,000 reviews. Reviews are that reason, and the specifics inside them are what the agent quotes.
Every hero product shows a review count and star rating. OpenAI's feed spec has review_count and star_rating fields (0 to 5, two decimals). Google's markup has aggregateRating. Fill both.
Review counts are honest and consistent everywhere. Same number on the page, in markup, in the feed. If your review app and your theme disagree, fix it.
You ask for reviews that mention use cases. "Kept me cool through a Texas August" is gold. "Great product!!" is noise. A post-purchase email that asks one specific question ("What problem did this solve for you?") gets you quotable reviews.
Photo reviews are on. They help human buyers, and some review apps put them in markup. They also make the review feel real when the agent hands the shopper to your page.
Negative reviews get answered. A calm, specific reply to a two-star review tells shoppers, and anything reading the page, that someone's home.
New products get to a minimum review threshold before you push them. How many is enough? I worked through the numbers in how many reviews you need to convert. Push the products that already have proof.
Off-site proof exists. Agents read the open web. A few honest mentions on Reddit, a roundup article, a creator review: these give an agent something outside your own site to cite. You can't fake this, and you shouldn't try.
Review content repeats the attributes in your title. If the title says "cooling" and no review mentions temperature, the claim stands alone. Ask reviewers about the attribute you sell on.
Your reviews are the only part of your catalog written by strangers. That's exactly why an agent trusts them more than your copy.
Group 5: Can you see what the agents sent you, and what it was worth? (items 33 to 40)
This is the group almost nobody does, and it's the one that tells you whether the other 32 mattered.
The opportunity is real. Adobe reported that in July 2026, AI-referral traffic to US retail sites converted at a 60% higher rate than non-AI traffic and generated 53% more revenue per visit, based on more than 1 trillion visits to US retail sites. That's an aggregate across big retailers, not a promise for your store. But it says these buyers show up warm.
AI channel orders are visible in your Shopify admin. Shopify says orders from AI channels show with channel or referrer attribution. Build a saved report filtered to those channels.
Feed URLs carry a tracking parameter. OpenAI's best practices suggest adding parameters like utm_medium=feed to product URLs when you want feed-specific click tracking.
AI referrers are grouped in your analytics. chatgpt.com, gemini.google.com, copilot.microsoft.com, perplexity.ai. Make one channel group so the traffic doesn't hide inside "referral." If Shopify and GA4 disagree on the numbers, here's why Shopify Analytics and Google Analytics never match.
You track conversion rate and average order value for AI traffic separately. Not blended with paid social. These visitors behave differently, so measure them differently.
Your landing page for AI traffic is your best product page. The agent sends the buyer to the exact product. Not your homepage. Not a collection. That page is where the money is made or lost.
Direct checkout settings are a decision, not a default. In Sales channels, then Agentic, each channel has its own toggle. In-chat checkout removes your page from the path, which also removes your bundles and free-shipping threshold. Decide per channel. My settings walkthrough is in the Shopify agentic storefronts setup guide.
You run a monthly prompt test. Ask ChatGPT, Gemini and Copilot the five questions your buyers would ask ("best X for Y under $Z"). Log whether you show up and what the agent says about you. It's crude, it's free, and it catches problems your dashboards won't. If you also sell on Amazon, run the same kind of test on its assistant with this Alexa for Shopping ranking study.
You re-check this list every quarter. The ChatGPT row in the table above changed twice in six months. Put a calendar reminder in for January.
What happens after the agent says your name?
This is the part I care about most, because it's the part that pays.
Agents decide who gets the visit. Your page decides who gets the money. ChatGPT hands every buyer to your checkout. Google hands bundle and subscription buyers to your page. So the most-visited page in your agentic future is the same product page you have today, now receiving a visitor who arrives with a specific question already in their head.
Run the math on a store like this, a hypothetical. A pet supplement brand selling a $60 joint chew for senior dogs. Agents start naming it when people ask for "joint supplements for older dogs that they'll actually eat." Conversion rate on the page 1.5%, average order value $60. That means revenue per visitor is $0.90. On 10,000 AI-referred visitors, that's $9,000.
Now rebuild the page around the question those shoppers arrive with: "Will my picky 12-year-old lab eat this, and how fast will I see a difference?" The taste guarantee up top, the 30-day timeline, a two-tub bundle at $84. Conversion rate moves to 3.0%, average order value to $84. Revenue per visitor becomes $2.52. On the same 10,000 visitors, that's $25,200.
Conversion rate x average order value x visitors. 1.5% x $60 vs 3.0% x $84.
Same agent. Same recommendation. Same visitors. $16,200 more, because the page finished what the agent started.
And this isn't theory for us. For a bedding brand we worked with, conversion rate went from 1.0% to 3.5% and average order value from $125 to $231. Revenue per visitor went from $1.25 to $8.10. On 10,000 visitors, that's $12,500 before and $81,000 after. See the full case study numbers. Real client numbers, not typical results, and not a promise of what your store will do.
The agent is the matchmaker. Your page is the first date. Nobody marries the matchmaker.
How to write that page for a visitor who already asked an AI five questions is its own topic. I wrote it up in how to write a Shopify product page for ChatGPT traffic.
Which agentic commerce checklist items should you fix first?
Forty items is a lot. Don't do them in order. Do them in this order:
- Items 1, 2 and 17 today. Catalog eligibility and complete policies. These are on/off switches. Everything else is a dial.
- Item 15 today. Check robots.txt for OAI-SearchBot. Five minutes.
- Items 3 to 8 this week, on your top 3 products only. Titles, facts-first descriptions, barcodes, variants, classification, stock. Not the whole catalog. The three products that make most of your revenue.
- Items 33 to 36 this week. You can't improve what you can't see.
- Item 37 next. Your hero product page, rebuilt for a buyer who arrives warm.
- Everything else over the next 30 days.
I've watched founders spend a month on llms.txt files and schema plugins while their best seller had "Misc" as its product type and no return policy. Fix the switches first. Then the dials.
The full map of every AI channel, with the reading list for this whole topic, is in the AI commerce guide.
Frequently asked questions
What is an agentic commerce checklist? It's the list of data, markup, policies, reviews and tracking an AI shopping agent needs before it can read your products, compare them and send a buyer to checkout. This one has 40 items in five groups.
Do I need to build anything to sell through ChatGPT on Shopify? No custom build. Agentic storefronts are on by default for eligible US stores, and ChatGPT sends buyers to your own checkout with no fee beyond payment processing. Your job is clean data and complete policies.
Can customers buy inside Google AI Mode or Gemini? Yes, for eligible US stores with a Merchant Center account, through Shopify-powered direct checkout. Subscriptions, bundles, customizable products and local delivery aren't supported there, so those buyers go to your store.
Which item matters most? Catalog eligibility. A product without an image, with a zero price, an unlisted status or missing store policies is invisible to every agent.
How do I know if agents are sending me sales? Use Shopify's channel attribution for AI orders, tag feed URLs, and group AI referrers in analytics. Then watch conversion rate and average order value for that traffic on its own.
What to do next
Open your Shopify admin today and go to Settings, then Policies. If any of the three required policies is blank or still a template, fill it in before you touch anything else on this list. It's the cheapest switch you'll ever flip.
Book Your Profit Audit
Agents can only send buyers to the page you already have, and right now that page decides whether an AI-referred visitor is worth $0.90 or $2.52. Book a profit audit and we'll show you where your hero page is leaking, then rebuild a high-converting product sales page in less than 15 minutes.
Or go here to check it out → revenueflows.ai
P.S. Forty boxes get the agent to say your name. One page gets you paid. Don't polish the matchmaker and ignore the first date.
Frequently asked questions
What is an agentic commerce checklist?
It's a list of the data, markup, policies, reviews and tracking an AI shopping agent needs before it can read your products, compare them and send a buyer to checkout. Ours has 40 items in five groups, and the feed group comes first because every agent starts there.
Do I need to build anything to sell through ChatGPT as a Shopify store?
No custom build. Shopify's agentic storefronts are on by default for eligible US stores, your catalog flows through Shopify Catalog, and ChatGPT sends the buyer to your own checkout with no fee beyond payment processing. Your job is clean product data and complete policies.
Can customers buy my products inside Google AI Mode or Gemini?
Yes, if you're eligible. Shopify says direct checkout in Google AI Mode and Gemini is on by default for eligible US stores with a Merchant Center account, but subscriptions, bundles, customizable products and local delivery aren't supported there.
Which item on the agentic commerce checklist matters most?
Catalog eligibility. If a product has no image, a zero price, an unlisted status or missing store policies, no agent can see it, and the other 39 items don't matter for that product.
How do I know if AI shopping agents are sending me sales?
Shopify shows AI channel orders with channel or referrer attribution, and you can tag feed URLs with a parameter like utm_medium=feed. Then track conversion rate and average order value for that traffic separately from everything else.