What is agentic commerce?

Agentic commerce is a model of ecommerce in which AI agents assist shoppers throughout the buying journey, from discovering products and comparing options to helping complete a purchase.

The word agentic comes from the idea of an AI system that can act.

Traditional AI answers a question. What is a good gift for a child turning ten.

An agentic system takes the next step. Find suitable gifts, compare the options, check availability and help me buy the best one.

The difference is not subtle. The AI is not only providing information. It is helping execute the task.

Shopify describes agentic commerce as letting shoppers discover, compare and purchase products inside a conversation, with platforms such as ChatGPT, Microsoft Copilot, Google AI Mode and Gemini becoming part of the emerging shopping ecosystem.

What changes when an agent shops your store

A human visitor loads your homepage, reacts to the photography, scans a collection page and forms an impression in about three seconds. Everything we normally optimise, which is theme performance, layout, imagery, merchandising and copy tone, exists to serve that moment.

An AI agent does none of that. It queries your catalogue, receives structured data, evaluates that data against the shopper’s stated constraints and returns a shortlist. Your theme is irrelevant to the process. Your product data is the entire product.

This is the single most useful thing to understand about agentic commerce on Shopify. The channel does not reward better design. It rewards better data.

The four layers of the Shopify agentic stack

Layer one: catalogue

Your product information has to reach the AI platforms in a form they can read. Shopify handles the syndication side of this, both through its own catalogue services that feed connected AI shopping experiences and through the feeds you already publish to Google and Meta.

What travels through this layer is not your product page. It is the structured record behind it, which includes title, description, category, price, currency, variants, options, availability, images, shipping terms and product identifiers.

Layer two: the agent interface

An agent needs a way to search your catalogue, ask follow up questions and assemble a cart. Shopify has built storefront capabilities on the Model Context Protocol, the open standard originally released by Anthropic and now widely adopted, which gives assistants a defined way to talk to a store’s catalogue, cart and support content.

In practice this means an assistant can run a constrained search against your products, retrieve the details it needs to compare them, and create a real cart tied to your store rather than a simulated one.

Layer three: cart and checkout

Once the shopper decides, the order has to become a real order. This is where the emerging protocols matter.

Shopify and Google describe the Universal Commerce Protocol as an open standard defining how AI agents interact with commerce systems across cart creation, checkout, payment and the experience after purchase.

Alongside it, the Agentic Commerce Protocol, released openly by OpenAI with Stripe, defines how an agent submits an order and how a delegated payment token replaces a raw card number. Google’s Agent Payments Protocol addresses the same trust problem from the direction of verifiable mandates, which means cryptographic proof that the shopper authorised this agent to spend this amount on this item.

The details differ. The principle is identical across all of them. You remain the merchant of record. The agent receives a scoped, revocable permission rather than access to the shopper’s wallet.

Layer four: merchant controls

Merchants need authority over which channels they participate in, what pricing and promotions apply, which markets are served and what inventory is exposed. Shopify’s current setup allows eligible merchants selling to buyers in the United States to take part in AI shopping experiences involving ChatGPT, Microsoft Copilot, Google AI Mode and Gemini, with the checkout experience varying by platform.

Availability moves month to month, so treat any specific list as something to verify in your own admin rather than something to plan a quarter around.

How a request travels end to end

Here is the full path a single shopper request takes, using a watch retailer as the example.

Step one. The shopper writes: I want a Swiss automatic under ₹1,20,000, around 40mm, on a steel bracelet, that ships within a week.

Step two. The agent decomposes that into constraints. Movement type automatic. Country of origin Switzerland. Price ceiling. Case diameter approximately 40mm. Strap material steel. Delivery window seven days.

Step three. The agent queries connected catalogues. Your store either matches or it does not, and this is where most merchants lose without knowing it. If case diameter, movement type and strap material live only in an image or a paragraph of marketing copy, your watches are invisible to this query even when three of them are perfect matches.

Step four. The agent retrieves details for candidate products. Price, availability, variant options, images, shipping terms and any reviews it can read.

Step five. The agent evaluates and shortlists, usually to three or fewer. Your delivery time and return policy are inputs at this stage because the shopper mentioned shipping.

Step six. The agent presents the shortlist with reasoning. The shopper asks follow up questions. Which of these has a sapphire crystal. Which has a longer power reserve.

Step seven. The shopper decides. Either the agent hands off to your storefront, or on a supported channel it creates a cart and submits the order with a delegated payment credential.

Step eight. Your Shopify store processes the order exactly as it processes any other. Payment, tax, fraud checks, inventory decrement, fulfilment and shipping all run normally.

Step nine. Everything after that, which is tracking questions, exchanges, sizing problems and warranty claims, arrives at your team the way it always has.

Notice where the work sits. Steps three, four and five are decided entirely by data you control and mostly have not finished filling in.

What Shopify handles and what you handle

Shopify handles catalogue syndication to connected AI channels, the agent facing interface, cart creation, payment processing, tax calculation, fraud screening, order creation and the standard fulfilment pipeline.

You handle product titles, descriptions, categories, variant completeness, specification metafields, images, inventory accuracy, price consistency, shipping and return policies written in concrete terms, structured data validity in your theme, crawler access and which channels you enable.

The division matters because merchants often wait for Shopify to ship a feature that will make them discoverable. No feature does that. The platform can only syndicate what you have entered.

Your Shopify admin checklist

This is the practical section. Everything below can be checked or fixed inside a standard Shopify setup.

Products and variants

Every variant needs its own price, availability, SKU and, where applicable, barcode holding a valid GTIN. Half filled variants are the single most common failure we find during audits, and they fail silently.

Set the product category using Shopify’s standard product taxonomy rather than leaving it blank or using a custom label only. That taxonomy value is one of the clearest signals a machine can read about what the product actually is.

Metafields for real specifications

Create structured metafields for the attributes that decide purchases in your category, then render them on the product page.

For fashion: fabric composition, fit, cut, length, sleeve type, occasion, care instructions, model measurements.

For watches: case diameter, case material, movement type, power reserve, water resistance, crystal, strap material, lug width, warranty term.

For jewellery: metal purity, gross and net weight, stone type, carat, certification, dimensions in mm.

For homeware and tapware: dimensions, finish, material, flow rate, certifications, installation requirements.

Use metaobjects where a specification set repeats across many products. The important part is that these values exist as data, not only as a line inside a paragraph and certainly not only inside a specification graphic. An agent cannot read your image.

Structured data in the theme

Shopify themes, including Horizon and Dawn, output product markup in JSON LD form. Customisation frequently breaks it, most often around offers, availability, price currency, aggregate rating and shipping details.

Validate every template type you use rather than one sample product. Check a simple product, a product with many variants, a product on sale, a sold out product and a bundle if you sell them. A store can pass validation on its default template and fail on the templates that carry most of the revenue.

Inventory accuracy

Recommending an item that turns out to be unavailable is the fastest way for an assistant provider to stop trusting a merchant. If you sell across offline stores and online, which many of our clients do, your inventory sync is now a marketing system rather than an operations detail.

Pricing consistency

Site price, feed price and checkout price must agree, and prices must be tax inclusive where that is the local expectation. Mismatches are a trust penalty and are easy to introduce through automatic discounts and market specific pricing.

Shipping and returns written in concrete terms

State the dispatch time in days, the delivery estimate by region, the return window, the conditions, who pays return shipping and the warranty term.

Avoid the phrase as per company policy. It is unreadable to a machine and unconvincing to a human.

Crawler access

Shopify lets you customise robots.txt through robots.txt.liquid in the theme. Review it, and review any bot rules at your CDN or firewall.

Many stores block AI crawlers by accident while trying to block scrapers. Decide deliberately which agents you want reading you, then confirm in your server logs that they are actually getting through rather than assuming it from the file.

Feeds and sales channels

The Google and YouTube channel pushes your catalogue to Google Merchant Center, which feeds more surfaces than most merchants realise. Fix the errors that accumulate there. Missing GTINs, disapproved items and stale prices all reduce your reach.

Review which sales channels are enabled and which markets are configured, since market configuration affects what is exposed and to whom.

Reviews

Many review apps render entirely in JavaScript with no structured markup, so review content exists for shoppers and not for agents. Check what is actually present in the page source, and choose an app that outputs proper markup if yours does not.

Themes: what to check on Horizon

Newer Shopify themes are cleaner structurally than the generation before them, but three things still need attention on any customised build.

Specification rendering. If your metafields do not appear in the page HTML, they may still reach some channels through the catalogue layer, but you lose the reinforcement of having them visible and crawlable. Render them.

Variant handling. Confirm that variant level price and availability update correctly in the structured data and not only in the visible interface.

Sections that hide content. Accordions and tabs are fine when the content is present in the HTML. They are a problem when the content is loaded only on interaction, which happens more often than teams expect on heavily customised builds.

Third party checkout, COD and the India reality

Most writing on this topic assumes a checkout environment that does not exist here.

Many Indian direct to consumer stores run GoKwik, Shiprocket Checkout, Razorpay Magic or similar for COD, address intelligence and RTO control. Agentic checkout paths are generally built against Shopify’s own checkout.

If you have replaced or wrapped that layer, the honest position is that completing a purchase inside an assistant is probably not available to you today. That is not a reason to ignore the channel. Discovery and comparison happen regardless of your checkout, and a shopper who arrives at your storefront already convinced by an AI shortlist is a high intent visitor by any measure.

Two further local points. COD availability becomes a hard filter when a shopper asks for it, so decide consciously whether you want agent driven COD orders given RTO economics. And assistant shopping features have rolled out market by market, usually the United States first, so confirm availability for India before building plans around it.

How to measure agentic traffic in Shopify

Standard analytics undercounts this badly. Referrals from assistants often arrive with missing referrer data, and a shopper who reads an AI answer then types your brand name into a browser appears as direct traffic or branded search.

Five things to put in place.

• Segment referrals from known assistant domains in GA4 and save the comparison so it is easy to check monthly.
• Review server logs for AI crawler user agents to see who reads your store and how often.
• Track branded search volume in Search Console as a proxy for assistant driven brand discovery.
• Add a post purchase survey asking how the customer heard about you. Blunt, but currently one of the more reliable signals available.
• Watch the ratio of new customer direct traffic to total sessions over time.

Expect attribution to stay imperfect for a while. Directional evidence beats none.

Common mistakes we see on real stores

Specifications trapped in images. A beautifully designed specification graphic that no machine can read, on a store with no equivalent text.

Product categories left blank. The taxonomy field is quick to fill and heavily weighted.

One good product page and four hundred thin ones. The flagship gets a rich description. Everything else gets two lines. Agents evaluate the four hundred.

Structured data validated once, at launch. Then six months of theme edits quietly break it.

Blocked crawlers nobody meant to block. Usually added during a scraping incident and never revisited.

Inventory that is roughly right. Roughly right is a different product from right when a machine is making recommendations on it.

Return policies written by a lawyer. Accurate, unreadable, and useless as a ranking input.

A ninety day plan

Days 1 to 30. Audit structured data across every template. Fill variant gaps. Set product taxonomy categories. Fix inventory sync issues. Review robots.txt.liquid and confirm crawler access in logs. Set up referral tracking and the post purchase survey.

Days 31 to 60. Build specification metafields for your top categories and render them on the product page. Rewrite the top fifty product descriptions to answer real buyer questions. Make reviews crawlable. Rewrite shipping and returns pages in concrete terms.

Days 61 to 90. Clear Merchant Center errors and expand feed coverage. Fix brand and product title consistency across marketplaces and listings. Test the full journey for an assistant referred shopper, including your third party checkout. Review what the first data is telling you and set a quarterly cadence.

Frequently asked questions

How does agentic commerce work on Shopify?

Shopify exposes your catalogue, cart and checkout to AI agents through structured connections rather than through your theme. The agent searches your product data, compares options against the shopper’s request, builds a cart and, on supported channels, submits the order. Your store processes it like any other order.

Do I need an app to enable agentic commerce on Shopify?

The infrastructure is provided by the platform rather than by a third party app. Your work is in the admin, which means product data, taxonomy, metafields, inventory, policies and channel settings.

Which AI platforms are connected to Shopify?

Shopify’s agentic shopping experiences have included ChatGPT, Microsoft Copilot, Google AI Mode and Gemini, with availability and checkout behaviour varying by channel and market. Verify the current list in your admin, since this changes frequently.

Does agentic commerce work with GoKwik or other third party checkouts?

Discovery and comparison work regardless. Completing the purchase inside the assistant generally depends on Shopify’s own checkout, so stores that have replaced that layer should plan for a handoff into their own checkout instead.

What is the Universal Commerce Protocol?

An open standard described by Shopify and Google that defines how AI agents interact with commerce systems across cart creation, checkout, payment and the experience after purchase.

Will my theme affect AI visibility?

Indirectly. The agent does not render your theme, but your theme controls whether structured data and specification content appear correctly in the page HTML, so a broken template can cost you visibility.

How long does it take to make a Shopify store agent ready?

For a catalogue of around a thousand products, expect roughly three months of steady work across Shopify development, merchandising and copy.

Is this the same as SEO?

It overlaps heavily but is not identical. Traditional SEO optimises for ranking against keywords. This optimises for being retrieved and correctly understood by systems that answer in sentences rather than links.

Do we need a developer?

Merchandising and copy work can sit with your team. Structured data validation, metafield and metaobject architecture, theme level rendering and feed configuration usually need development support.

Agentic Commerce on Shopify: Building Ecommerce for an AI-First Future