Agentic commerce
for brands, in order.

Agentic commerce is what happens when an AI assistant shops on a buyer's behalf. The buyer tells the agent what they want. The agent searches a catalog of structured product data, assembles a shortlist, and sends the buyer to complete purchase. As of Shopify's Winter '26 edition, every Shopify store is discoverable this way by default through Agentic Storefronts. That handoff lands on your store, through Shopify Checkout, with you as merchant of record. What brands should do right now: audit catalog identifiers and structured attributes first, confirm checkout is fast and works on mobile, then work the list below in order. Skip the catalog step and nothing else matters.

For two years the question every brand asked me was some version of "how do I rank in ChatGPT." It was the wrong question, but it was pointing at the right anxiety. People could feel that the front door to discovery was moving and they did not own the new door. The door has now moved for real, and the order you act in is where most brands get it wrong.

What changed,
without the drama.

Here is the mechanic, stripped down. An AI agent gets asked to find someone a pair of trail runners under $140. It queries a catalog of products it can read. It assembles a shortlist. The buyer picks one. The agent hands them off to complete the purchase. Under Shopify's Agentic Storefronts, that handoff lands on your store, through Shopify Checkout, with you as merchant of record. The plumbing underneath that handoff is standardizing fast, with Visa and OpenAI defining agent tokens for checkout. You keep the customer relationship, the order data, and the lifetime value.

Contrast that with the model that just failed. OpenAI's ChatGPT Instant Checkout, launched in September 2025, let buyers complete the purchase inside ChatGPT. By late March 2026 OpenAI pulled it. Roughly twelve Shopify merchants had gone live in that whole window. Twelve. That is not a rounding error, that is a verdict. I covered the post-mortem in what happened when ChatGPT Instant Checkout was pulled.

The strategic picture is clear. The agents are real and growing. The winning architecture keeps the transaction on your turf. Your job is to be the product an agent can find and trust.

Those two questions determine everything. If the answer to either is no, you are out of the conversation before it starts. AI agents do not tolerate ambiguity the way a human shopper does. A human will click through a confusing title to see if the product is right. An agent will skip your listing and move to the next one.

Where do you
actually stand?

Before getting into the mechanics, it helps to see the full picture. Here is the readiness checklist I run through with every brand I work with. The items are ranked by how quickly each one affects whether an agent can surface your products.

Run through that list honestly. Most brands have at least two red flags, and they are almost always in the catalog data rows. Checkout and post-purchase issues get fixed fast once someone finds them. Data issues get deferred because they look boring. That deference is where brands get beaten.

Boring data wins
the agent game.

This is the part nobody wants to hear and everybody needs to do first. Agentic Storefronts is built on Shopify's Catalog API. The agent does not browse your beautiful product detail page. It reads structured fields. If those fields are thin, wrong, or missing identifiers, you are invisible no matter how good your brand is.

The Catalog API gives agents access to your product data through a machine-readable format. When an agent handles a query for "a trail runner under $140 with a drop under 10mm," it is not scraping your PDP looking for that spec. It is querying structured fields. If drop is buried in a paragraph of description prose, the agent cannot use it. You get skipped.

Identifiers first

GTIN and MPN are the identifiers that let agents cross-reference your product against other data sources: reviews, retailer comparisons, spec databases. A product without a GTIN is harder to identify and trust. For every variant in your catalog, open the product in Shopify admin and confirm the barcode and SKU fields are populated and unique. Brand name should be in a dedicated field, not only in the title.

This gets messy fast for brands with large catalogs. Prioritize your top-selling SKUs and work down. The products agents are most likely to surface are the ones buyers are already asking for, so those are the ones where clean identifiers pay off fastest. For a deeper look at how the API processes this data, the Catalog API and AI agents breakdown is worth a full read.

Attributes out of prose

Size, color, and material are the easy ones. Most brands have those as variant options already. The harder ones are category-specific specs: weight, drop, waterproof rating, fabric weight, thread count, battery life, voltage. These are the attributes an agent uses to match a specific buyer query. If they live in your description as a paragraph, the agent cannot reliably extract them.

Shopify metafields are the right place for these. Set up a metafield namespace for your product category, add the fields, and fill them. This is grunt work. It pays back in agent visibility, search performance, and your own catalog hygiene. Do it once and keep it current when you add new products.

Titles that say what the product is

Your product title has two jobs now. The first is the one it always had: help a human understand what they are looking at. The second is newer: tell an agent what category this product belongs to. A title like "The Original" or "Classic II" or "Summit Series" tells an agent almost nothing. A title like "Trail Running Shoe, Waterproof, Low Drop" is immediately useful.

You do not have to kill your brand names. "Pegasus Trail Runner, Men's, Waterproof" keeps the brand identity while giving the agent a category word to work with. Go through your top 50 products and ask: if an agent read only the title, would it know what category this is? If not, the title needs a revision.

I dug into why some brands surface consistently while others vanish in winning or invisible in ChatGPT. The pattern is consistent: brands showing up have cleaner structured data. The brand with the better product but messier catalog loses.

Keep the transaction
on your turf.

The single most important architectural decision is one Shopify already made for you with Agentic Storefronts: checkout completes on your store, and you stay merchant of record. Do not undo that by chasing closed-loop checkout experiences that take the customer relationship away from you.

There was a reported fee, around 4%, on ChatGPT-originated checkout sales. Think about what you get for that: maybe more discovery, on a surface you do not own, with a customer relationship that belongs to the platform. Pay nothing and own the relationship, the data, the email, the post-purchase flow, and the lifetime value. When the agent hands a buyer to your Shopify Checkout, you get all of that. Agentic Storefronts, explained walks through exactly how the handoff and merchant-of-record status work.

What owning checkout actually requires

It sounds simple. It requires attention to three things that a lot of brands have quietly let slip.

First: mobile checkout has to work. An agent sending a buyer to your store does not know whether they are on a phone or a desktop. The buyer is probably on their phone. If your checkout has any friction, any step that is slow to load, any form field that misbehaves on iOS, that buyer abandons. Test your checkout on a real phone, not a browser's device-emulation mode. Do it today. Store speed and conversion covers the technical baseline worth hitting before you rely on any agent-referred traffic.

Second: post-purchase capture has to fire correctly. An agent-referred order comes through Shopify Checkout like any other order. But if your flows are triggered by a specific UTM source or a custom tag that only gets applied through your own marketing channels, those flows may not fire on agent-sourced orders. Check your email platform setup. Confirm the order confirmation, the winback sequence, and the review request all trigger correctly regardless of attribution source. Klaviyo flows for Shopify brands covers the setup patterns worth auditing here.

Third: inventory accuracy is not optional. An agent that sends a buyer to a sold-out product learns from that experience. Not because it has feelings about it, but because buyers who land on sold-out products abandon. Poor outcomes feed back into which products get surfaced. Keep your inventory feeds live and accurate. Inventory management for DTC brands covers the feed hygiene worth getting right.

The conversion rate angle

Agent-referred traffic may convert differently than your paid social traffic. The buyer has already done comparison shopping inside the AI interface. They arrive further along in their decision. That is a higher-intent visitor than most of your paid traffic. Your product page still has to close the deal, but the psychology is different. What a visitor who has already seen three competing options needs is confirmation: clear differentiation, strong social proof, and a fast path to checkout. A product page audit can identify friction points before they become agent-era problems.

What twelve merchants
taught the rest of us.

I want to sit on the ChatGPT Instant Checkout failure for a second, because the lesson is more useful than the obituary. It launched with real fanfare. It had OpenAI's distribution behind it. And it got about twelve Shopify merchants live before it was pulled in March 2026.

Why so few? My read, from talking to operators: the model asked brands to give up the thing they value most (the direct customer relationship) in exchange for a discovery surface that was unproven. The math did not work. You were paying a fee and handing over the customer to maybe show up in an answer. Smart operators waited. They were right to wait.

There is a broader pattern here worth naming. Every few years a platform offers brands a shortcut to distribution in exchange for some combination of fees, customer data, or relationship ownership. TikTok Shop has a version of this tension. Amazon has had it for twenty years. The answer is never "never participate." The answer is "understand exactly what you are giving up and what you are getting." In this case, the Instant Checkout trade was customer ownership for unproven discovery. It failed fast. The difference between user-controlled and agent-controlled purchasing gets into the architecture of why some of these models survive and others do not. If you build apps rather than run a store, the same shift decides which Shopify apps survive when an agent is the buyer.

The order
of operations.

Do first, this quarter. Audit catalog identifiers (GTIN, MPN, brand) on every variant, starting with your top 50 SKUs. Move size, color, and material out of description prose and into structured fields or metafields. Fix stale price or inventory feeds. Confirm Agentic Storefronts is enabled. Test your checkout on a real phone. This is the work that determines whether agents can even find you. Everything else is secondary.

Do next, this half. Revise product titles so they lead with category and specifics, not slogans. Add category-specific attribute metafields for the specs buyers actually search on. Add a clean primary image per product on a neutral background. Confirm your post-purchase flows fire correctly on agent-sourced orders. Start monitoring which products surface in AI answers. AI-referred traffic and what it converts at gives you a benchmark to compare against.

Do later, or deprioritize. Chasing any one AI platform's closed checkout. Paying for placement in a single assistant before you have proven the channel converts. Rebuilding your whole PDP for agents (they read your structured data, not the page). Spending on AI-specific features before you have cleaned the underlying catalog data that all of those features depend on.

One thing worth doing in parallel: think about your answer engine visibility. Agents pull from multiple sources, not just catalog data. If you write nothing and publish nothing, agents have less to work with when a buyer asks about your category. Answer engine optimization for commerce brands and GEO vs SEO cover how organic content visibility feeds into what agents recommend.

Questions I get
every week on this.

The brands that win agentic commerce will not be the ones with the cleverest AI strategy deck. They will be the ones whose product data is clean and whose checkout they own. Get the boring part right first. The discovery follows. If you want a read on where your catalog stands, start with Agentic Storefronts explained, then come find me at the inquiry page.

Questions I keep
getting asked.