AI Shopping Agents Are Coming for Your Storefront: A Readiness Checklist
Agentic commerce has stopped being a slide-deck concept. Google has rolled out agentic checkout across Search and Gemini, which means a buyer can describe what they want, watch an AI assemble the options, and approve a purchase without ever opening your storefront. ChatGPT, meanwhile, runs shopping research for buyers who would once have spent an evening with twelve browser tabs open.
The shift underneath is simple and uncomfortable: the primary consumer of your product data is increasingly a machine. Your product page was built to persuade a person — the lifestyle photography, the social proof bar, the carefully sequenced scroll. An agent reads none of it. It reads your structured data, your feed, your policy markup, and it makes a decision in milliseconds.
Most Shopify stores are not ready for that reader.
of online shoppers using AI agents by 2030
That's Morgan Stanley's projection: nearly half of online shoppers using AI shopping agents by 2030, routing around a quarter of their spend through them. It's a projection, not a certainty. But even a partial version of it changes who gets found.
What an AI shopping agent actually does
Take a concrete request: trail running shoes, decent grip, under £150, delivered by Friday. A human turns that into an afternoon of browsing. An agent turns it into a constraint-satisfaction problem and works it in three moves.
Interpret
Parses intent into structured constraints: category (trail running shoes), attribute (outsole grip), budget cap (£150), delivery deadline (Friday).
Compare
Queries structured offers across merchants (schema.org Product and Offer data, shopping feeds, checkout APIs) and ranks every candidate against the constraints.
Execute
Carries the winning offer through checkout on the buyer's behalf, or hands back a pre-filled cart that's one tap from purchase.
The agent loop: interpret intent, compare structured offers, execute the purchase. Merchants win or lose at step two.
Notice where you actually compete. Not at execution — the checkout rails are being standardised for you. The competition happens at the comparison stage, and it's binary in a way human shopping never was. If your £139 trail shoe doesn't expose its price, availability and delivery window in machine-readable form, the agent doesn't weigh it less favourably. It never sees it.
You don't lose the argument. You're not in the room.
The good news for Shopify merchants specifically is that the platform already holds your catalogue in structured form. Variants, prices, inventory, shipping profiles: it's all in there. The gap, in almost every store we audit, sits between what Shopify knows and what your storefront actually exposes. Closing that gap is configuration and theme work, not a replatforming project.
How big this gets, read honestly
Alongside the Morgan Stanley adoption figure, McKinsey estimates that three to five trillion dollars of global retail spend could be redirected by agentic commerce by 2030. Both numbers deserve the same treatment: they are projections from analysts whose record on adoption curves is mixed, not measurements. We'd be suspicious of anyone presenting them as settled fact.
of global retail spend that McKinsey estimates could be redirected by agentic commerce by 2030. An estimate, not a guarantee, but a range large enough to take seriously at either end.
What isn't a forecast is the infrastructure. Google and OpenAI have already shipped production shopping surfaces — that part is fact rather than projection — and platforms don't build checkout rails for fun. Whether agents end up mediating 10% of discovery or 40%, the merchants who are legible to machines will take a disproportionate share of whatever flows through them.
There's a second-order effect worth naming for mid-sized brands. Agent-mediated shopping flattens some traditional advantages — brand recall matters less when a machine assembles the shortlist — and sharpens others, like price accuracy and delivery reliability. A £2m DTC brand with immaculate data can sit beside a household name in the same comparison set. That has never been true of shelf space, and it was only ever half-true of search.
We've covered the discovery half of this shift in our guide to answer engine optimisation for ecommerce. Shopping agents are the transactional end of the same change: first the machines answered questions about products, now they buy them.
The agent-readiness checklist
This is the audit we now run alongside every subscription and CRO engagement. Six checks, in roughly the order an agent encounters them. An agent exits at the first failure.
Six checks before an agent will recommend you
-
Complete Product and Offer structured data
Every product and every variant carries schema.org
ProductandOffermarkup: price, currency, availability, SKU and GTIN where you have them. This is the index entry that gets you into the comparison set at all. -
Accurate live price and availability
The price in your markup matches the price in the cart, every hour of every day. Stale sale prices and "in stock" flags on dead variants are how machine trust dies.
-
Machine-readable shipping and returns
Delivery speed and return terms expressed as
OfferShippingDetailsandMerchantReturnPolicy, not prose buried in an FAQ accordion. "By Friday" gets resolved from this data or not at all. -
Stable, fast pages
Agents operate on timeouts. A product page that takes eight seconds to hydrate, or renders differently on every load, reads as unreliable infrastructure rather than charming jank.
-
Consistent variant data
The same variant carries the same identifier, price and stock state in your page markup, your shopping feed and your sitemap. Disagreement between sources is read as a data quality failure across the lot.
-
Accessible semantic markup
Real buttons, labelled forms, headings in order. Agents that fall back to navigating your pages do so the way assistive technology does. Accessibility work and agent-readiness are largely the same job.
The six-point agent-readiness audit for a Shopify storefront. Each check gates the next: incomplete data means the price never gets compared, and an unstable page means the checkout never gets reached.
What to fix first on a typical Shopify store
Having run this audit against a few dozen Shopify storefronts, the failure pattern is consistent enough to prescribe an order. Start with structured data, because it's where the most damage hides. Many themes render Product markup for the default variant only, so a store with 40 products and 400 variants is invisible at the variant level, which is exactly the level agents shop at. Worse, SEO and review apps frequently inject their own competing JSON-LD blocks, leaving Google to pick between two contradictory prices. We've written a full breakdown of structured data for AI agents if you want the implementation detail.
If you sell on subscription, there's a second layer: check whether your selling plan pricing appears anywhere in your markup. In our experience it almost never does, which means an agent comparing your subscribe-and-save price against a competitor's one-off price is comparing the wrong numbers — usually against you. Our guide to structured data for subscription products covers how to express recurring offers properly.
Second, shipping and returns markup. This is typically a few hours of theme work, and because machine-readable policies are still rarer than they should be, it's a genuine differentiator for now. A delivery deadline is a hard constraint in most agent queries; merchants who can't answer it programmatically forfeit every query that includes one.
Third, speed and stability: not because they matter less, but because the first two are usually broken in ways performance work can't compensate for. Once the data layer is sound, our Core Web Vitals guide for Shopify is the place to start.
A quick self-test that costs nothing: run a product URL through Google's Rich Results Test, then ask an LLM with browsing enabled to list your variants, prices and delivery promise from the page. If either comes back wrong or empty, you've found your starting point.
Do them in that order, even though the temptation is to start with speed because it has the prettiest dashboards.
What agents punish
Some practices that merely irritate humans are disqualifying for machines. Three come up again and again.
Cloaked pricing
Prices revealed only in the cart, member-only pricing with no markup, surprise fees appearing at checkout. An agent that can't verify the final price before execution will not execute.
Popup walls
Newsletter modals, interstitials and cookie walls that block the page. A human sighs and finds the close button. An agent may simply fail the page and move to the next merchant on the list.
Inconsistent feeds
The product page says in stock, the feed says out. Markup says £42, checkout says £47. Surfaces don't argue about discrepancies; they quietly downrank or exclude the offer.
Three disqualifiers in agent-mediated shopping. None of them produce an error you'll ever see, only absence from results.
"An agent doesn't bounce, complain or leave a one-star review. It crosses you off a list you never knew you were on."
To make that concrete: a brand we audited this spring had a genuinely well-optimised product page sitting on top of a shopping feed that hadn't synced properly since a platform migration eighteen months earlier. Around a fifth of the catalogue showed phantom availability. For human traffic the cost was a trickle of support tickets. For anything reading the feed, it marked the entire catalogue as unreliable — the well-maintained 80% was being judged by the broken 20%.
The boring work wins
Here's the part we find genuinely optimistic: almost everything on the readiness checklist is work you should have done anyway. Complete structured data earns rich results today. Fast, stable pages convert humans better today. Accessible markup is a baseline obligation, not a tactic. Consistent feeds make your existing Shopping campaigns cheaper. Agent-readiness isn't a new workload so much as a deadline attached to an old one.
Our position is that the projections will be wrong in the specifics and right in the direction, and that the sensible response is the same either way. The brands doing unglamorous data hygiene now will be over-represented in agent recommendations for the same reason early technical SEO adopters dominated early Google: the machines reward whoever makes their job easiest, and they reward them for years.
The alternative is retrofitting under pressure, in 2028, while a competitor's variants sit in every comparison set you're missing from. We'd rather audit your storefront before the machines do.