What agentic commerce actually means for your Shopify store

Something changed in how people buy online, and most Shopify stores have not noticed yet. Your customers are starting to hand the shopping to software. They ask ChatGPT to find the best waterproof hiking boot under 200 dollars, or they tell Perplexity to compare three brands and just buy the winner. The person never lands on your homepage. An agent does the browsing, the comparing and, increasingly, the buying.

This is what people mean by agentic commerce: a shift from customers visiting stores to AI agents transacting on their behalf. The agencies shouting about it loudest have wrapped it in acronyms (DTA, UCP, GEO, MCP) to make it sound like a priesthood. It is not. Underneath the jargon is one simple, uncomfortable question for every merchant: when an agent is doing the shopping, does it pick your store, or someone else's?

We build and run Shopify Plus stores for a living, so this is not a trend piece. This is what the shift actually does to a store, which parts are real today, which are still hype, and what we would do about it if it were our revenue on the line.

What agentic commerce actually means

Strip out the branding and there are three layers, from the one that matters today to the one that is still mostly a demo.

Discovery. A shopper asks an AI assistant to find or compare products, then clicks through to buy themselves. This is live, at scale, right now, and it is where nearly all the money is today.

Assisted checkout. The agent fills the cart and pre-fills the details, the human approves the final purchase. This is rolling out fast through the payment and platform players.

Autonomous purchase. The agent buys within rules you set, no human in the loop per order. Real, growing, but still the smallest slice.

The mistake is treating this as a far-off, fully autonomous future you can ignore. The discovery layer is already deciding who gets the sale, and it runs on whether an AI can read, trust and recommend your store.

The numbers are not hypothetical anymore

It is easy to dismiss this as a demo that never becomes revenue. The 2025 holiday data says otherwise, and these figures come from the platforms measuring real traffic and real orders, not from an agency trying to sell you something.

Adobe measured traffic to US retail sites from generative AI sources up roughly 1,200 percent year on year in early 2025, and it kept climbing through the year. The channel went from a rounding error to a real front door in about twelve months.

Across the 2025 holiday season, Salesforce attributed about 262 billion US dollars of global online spend to AI and agents, close to one dollar in every five spent online.

AI-referred visits converted 31 percent more often than other traffic, with more time on site and a lower bounce rate (Salesforce). On peak days like Thanksgiving, Adobe put the conversion lift as high as 54 percent. Agent traffic is not just bigger, it buys.

Here is the number that should decide your next quarter. Salesforce found that retailers running their own branded shopping agents grew holiday sales 59 percent faster than those that did not (6.2 percent versus 3.9 percent year on year). The stores that leaned into this are already pulling away from the ones waiting to see how it plays out.

This breaks the old SEO playbook

For twenty years the game was ranking on a page of blue links a human scrolls. Agentic discovery does not work like that. An AI assistant reads, synthesises and returns one answer, or a shortlist of three. There is no page two. You are either in the answer or you are invisible, and nobody sees the store that came fourth.

The discipline forming around this is being called GEO, or generative engine optimisation. Ignore the acronym and hold onto the principle: you are no longer optimising to be found by a person who will judge your site for themselves. You are optimising to be understood, trusted and recommended by a machine that will judge it for them, in a fraction of a second, and move on.

In practice that rewards things good stores already value and punishes things they already should have fixed. Clear, structured product data an agent can parse without guessing. Genuine trust signals, real reviews and specifics an assistant can cite. Fast, clean pages, because an agent that times out fetching your product simply recommends the one that answered. The stores that win here are not the loudest. They are the most legible to a machine.

The plumbing: feeds, MCP and the agent-ready store

This is where the acronyms get thickest, so here is the builder's version. For an agent to recommend and eventually buy from your store, it needs two things: a clean way to read your catalogue, and a safe way to act on it. Most of that is unglamorous groundwork you can start on today.

The reading layer is structured data. Rich product schema, accurate and complete feeds, and inventory and pricing that are actually current, not a nightly export that is wrong by lunchtime. Agents trust structure. A store whose data is messy gets skipped in favour of one whose data is clean, the same way you would not recommend a supplier whose price list you could not trust.

The acting layer is where MCP and Shopify's own agent tooling come in. MCP (Model Context Protocol) is simply a standard way for an AI to call a tool or a service, a common plug so an assistant can query your catalogue or start a checkout without a custom integration for every model. Shopify is building the commerce side of this into the platform. Our honest read: you do not need to hand-roll an experimental MCP server this quarter to be ready. You need the fundamentals underneath it correct, because when the standardised rails land, the stores with clean data and fast pages switch on in an afternoon, and the ones without spend months catching up.

What we would do this quarter

If this were our store, we would not chase the shiniest acronym. We would get legible to agents in the right order.

1. Audit how an AI sees you today. Ask ChatGPT and Perplexity to recommend products in your category. If you are absent, or worse, misdescribed, that is your baseline and your priority.
2. Fix the data layer. Complete, accurate product schema and feeds. Real-time inventory and pricing. This is the single highest-leverage move and it helps conventional SEO too.
3. Earn citable trust. Specific, structured reviews and product detail an assistant can quote. Vague marketing copy is invisible to a machine that wants facts.
4. Make the store fast. An agent that times out fetching your page recommends a competitor. Performance is now a discovery issue, not just a conversion one.
5. Then, and only then, wire up the agent rails. MCP endpoints and Shopify's agentic checkout tooling matter, but they pay off only once the four steps above are solid.

The Sonder take

Here is the part the acronym-sellers leave out. Agentic commerce is not a project you finish. The models change, the standards move, the way an assistant weighs your store this month is not how it will weigh it next quarter. A store you optimise for agents in July and never touch again is stale by summer, in exactly the way a launched-and-abandoned website has always gone stale, only faster now, because a machine re-evaluates you constantly.

This is precisely why we build living websites. A store that watches its own data, catches the feed that broke overnight, and keeps pace as the rules shift is not a nice-to-have in an agent-led market. It is the difference between being recommended and being skipped. The winners will not be whoever published the cleverest MCP guide. They will be the stores that stay legible, fast and trustworthy to a machine, week after week, without someone remembering to check.

You do not need to panic, and you do not need to buy a priesthood's worth of new acronyms. You need clean data, fast pages, real trust signals, and a way to keep them that way as the ground moves. Get those right and you are ready for whatever the agents do next. Get them wrong and the most sophisticated buyer in your market, the one doing the shopping for thousands of your customers, quietly recommends someone else.

Sources: Adobe Analytics generative AI retail traffic and conversion reporting, 2025. Salesforce 2025 holiday shopping data and Cyber Week figures. Figures are directional and reflect US retail measurement over the 2025 holiday period.