# How Shopify Stores Are Preparing for AI Shopping Agents Beyond Just Orders > How Shopify stores are preparing for AI shopping agents: what's automatic, what needs a toggle, and why the real work is product data, not settings. Source: https://arbyn.app/blog/how-shopify-stores-are-preparing-for-ai-shopping-agents-beyond-just-or Published: 2026-07-18 --- AI shopping agents on Shopify grew from a hypothetical to a measured, reported growth line inside of two quarters. AI-driven traffic to Shopify stores grew 8 times year over year in Q1 2026, and orders originating from AI-powered searches grew nearly 13 times, according to Shopify's own reporting. New buyers arriving through AI channels place orders at nearly twice the rate of buyers from other channels. The question for most Shopify stores isn't whether to prepare for this. Enrollment in the core mechanism is already automatic. The real question is what preparation actually means beyond checking a settings toggle. Half the stores I talk to think being ready for AI shopping agents means turning something on. It mostly means fixing your product data so an agent can actually read it. Odera Joseph Echendu, Founder, Arbyn What's Already Automatic, and What Still Requires a Decision As of March 2026, Agentic Storefronts activated by default for eligible Shopify stores selling to US buyers, syndicating product catalogs automatically into ChatGPT with no separate setup step. Google AI Mode and Gemini, and Microsoft Copilot, by contrast, require a manual toggle in Settings, Sales Channels, Agentic Storefronts, meaning a store that assumes it's fully enrolled everywhere because ChatGPT enrollment happened automatically may still be invisible on the other two surfaces until someone actively flips those switches. The checkout model itself settled into something more conservative than the earliest coverage suggested. OpenAI's original Instant Checkout, launched in September 2025, charged participating stores a 4 percent fee on purchases completed directly inside ChatGPT. That model was discontinued in March 2026, with OpenAI stating it didn't offer the flexibility the company wanted to provide. The replacement model that launched by default for all Shopify stores redirects buyers to the store's own checkout to complete a purchase, with no additional transaction fee beyond standard payment processing. Google's AI Mode and Gemini, and Microsoft's Copilot, both support checkout completing natively inside the AI interface itself, powered by the Universal Commerce Protocol, a distinction worth knowing since it means the three major AI surfaces don't behave identically even though all three are nominally "agentic storefronts." Shopify's own executives have been explicit that this doesn't change who controls the transaction layer. President Harley Finkelstein has framed the shift as a distinction between front-end and back-end infrastructure: the AI interface may evolve and multiply, but Shopify's checkout, payments, and fraud prevention systems remain the layer actually processing the sale, regardless of which AI surface a customer discovered the product through. Speaking at the Upfront Summit in March 2026, Finkelstein noted that only about 18 percent of US retail purchases currently happen online at all, framing agentic shopping as a potential new front door that could grow that share rather than simply redistributing existing online sales between channels. He specifically called out that this could help smaller, less-discovered brands get surfaced to customers who wouldn't have found them through traditional search, not just benefit already-dominant retailers with existing SEO advantages. This matters practically because it means a store's existing checkout configuration, payment methods, tax setup, and fraud rules don't need to be rebuilt for each new AI channel individually. The open Model Context Protocol (MCP) is also gaining traction as a separate, complementary standard in this space, and multiple 2026 technical guides note that store owners optimizing their underlying product data once, clean attributes, accurate structured data, tend to serve all of the competing protocols simultaneously rather than needing protocol-specific work for each one, since the underlying requirement, machine-readable, accurate, complete product information, is the same regardless of which AI platform is doing the reading. The Real Preparation Work Is Data Quality, Not Settings Every source covering how to prepare for AI shopping agents converges on the same underlying point, stated in different words: an AI agent doesn't see a store the way a human shopper does, it parses structured data fields, and incomplete or inconsistent product data is the most common reason a product doesn't get recommended, entirely within a store's own control to fix. Specific, recurring recommendations across multiple 2026 guides include clear, specific product titles without vague naming, complete structured specifications, dimensions, materials, compatibility, use cases, spelled out rather than implied, consistent pricing across every collection page and product page, and policy clarity on shipping, returns, and refund windows written in plain language rather than buried behind an accordion menu or a JavaScript element AI crawlers struggle to parse. The technical layer underneath this, JSON-LD schema markup, is worth naming specifically for any store with development resources to invest here. Shopify provides basic Product schema by default, but several 2026 technical guides recommend expanding this with additionalProperty fields covering every attribute an agent might query, aggregateRating data, and complete availability information rather than relying on the default markup alone. This is genuinely technical work, not copy editing, and it sits at the more advanced end of preparation, worth prioritizing after the more accessible fixes, clear titles, consistent pricing, plain-language policies, rather than before them, since the basic data hygiene work benefits both human and AI shoppers immediately while the schema expansion work specifically targets machine readability. Real, named brands are already participating in this specifically, which grounds the abstraction in something concrete: Glossier, Spanx, Vuori, Stanley, and Steve Madden are cited by Shopify as active Agentic Storefronts participants as of early 2026. These are established brands with existing catalog and content discipline already in place before this shift, which is worth noting honestly, a smaller or newer store without that same baseline of clean, structured product data has more foundational work to do before AI agent visibility becomes a meaningful channel, rather than assuming enrollment alone puts it on equal footing with a brand that already had rigorous product data practices. Complex Query Handling and Why This Isn't a Google Search Replacement Everywhere AI shopping agents handle a specific kind of query meaningfully better than traditional search: comparison-heavy, high-consideration questions where a shopper wants an assistant to narrow options based on multiple stated criteria at once, rather than a single keyword. A query like finding a lightweight running jacket under a specific price that delivers by a specific date is the kind of multi-constraint request an AI agent can process in one exchange, where traditional search would require several separate searches and manual comparison across results. This is genuinely different from, and complementary to, traditional search rather than a wholesale replacement, since research-oriented, local, and heavily branded searches are likely to remain search-dominant for longer, according to multiple 2026 analyses, while comparison-heavy categories shift meaningfully toward agent-mediated discovery. FAQ coverage matters specifically because AI agents actively pull from a store's stated policies to filter recommendations. A shopper instructing an agent to only consider brands with free returns will exclude a store's products entirely if that policy isn't stated clearly and accessibly, even if the store's actual return policy would satisfy the request. This is a different failure mode than a product simply ranking lower, it's exclusion from consideration entirely because the agent couldn't confirm a stated requirement was met. Customer reviews and product Q&A content matter for a related but distinct reason: multiple 2026 guides note AI agents actively use this content to assess product quality and relevance when deciding whether to recommend an item, not just the store's own product description. A store with thin or absent review content is giving an agent less signal to evaluate against a competing product with richer, more established review history, independent of whether the underlying product quality is actually comparable. This is worth flagging specifically because it means review generation, already a priority for App Store credibility as covered elsewhere on this blog, now has a second, AI-discovery-specific reason to matter beyond just human shopper trust. Shopify's own Knowledge Base app, covered in more detail elsewhere on this blog, exists specifically to give a store a controlled way to feed FAQ and policy content to AI agents rather than relying on however cleanly an agent can parse a storefront's own pages. Using it doesn't replace the underlying need for clean, accurate product and policy data everywhere else it lives, it's an additional channel, not a substitute for getting the fundamentals right on the storefront itself. A store treating the Knowledge Base app as a complete solution, without also fixing underlying product page data quality, is solving only the half of this problem that's easiest to fix, while leaving the harder, more foundational catalog cleanup work undone. The Skeptical Case, Which Deserves a Fair Hearing Not every source covering this topic assumes AI shopping agents will dominate discovery, and the more measured takes are worth including rather than only citing the growth numbers. One 2026 analysis frames the honest range as AI agents capturing 10 to 20 percent of discovery traffic over the next few years for high-consideration categories like electronics, home goods, and fashion, with lower adoption for impulse-purchase categories where human-led browsing likely stays dominant longer. The same analysis lays out both the base case and the more skeptical bear case directly: the base case assumes agentic discovery genuinely takes meaningful share in comparison-heavy categories over the next few years, while the bear case holds that agents stay marginal because shoppers continue to prefer human-led browsing and discovery for most purchase decisions, with AI playing a smaller supporting role rather than becoming a primary discovery channel. The same analysis notes explicitly that the preparation work, clean structured data, modern checkout, accurate policies, improves human conversion regardless of how large AI-driven traffic ultimately becomes, which is a genuinely useful framing: none of the recommended preparation is wasted effort even under a conservative adoption scenario. This is worth emphasizing because it changes the risk calculation for a store deciding how much time to invest in this now. A store that spends a few hours cleaning up product titles, specifications, and policy pages hasn't wasted that time even if AI shopping agents end up capturing only the lower end of the projected range, since the same cleanup improves how human shoppers experience the site and how traditional search engines index it. Attribution and Measurement, a Genuine New Gap Orders originating from AI-referred traffic flow into a Shopify store's admin with channel attribution tagging which AI platform drove each sale, which lets a store compare performance across channels the same way it already tracks other marketing channels. The gap worth naming directly: most stores' existing analytics dashboards, built around traditional channels, don't natively surface this new attribution category prominently, meaning a store owner has to actively look for it rather than have it appear automatically alongside familiar traffic sources. A store that hasn't specifically checked its channel attribution reporting for AI-referred orders since this rolled out may be underestimating how much of its current volume is already arriving this way. This attribution gap compounds with a measurement discipline several sources flag as easy to skip but important to maintain: confirming that any change made to product data or policy pages actually produced a measurable effect on agent-driven recommendations requires isolating that change from ordinary variation, since AI answers themselves shift over time independent of anything a store does. Drawing a conclusion from a single before-and-after comparison, without a control question left unchanged and a reasonable time window, risks mistaking normal model drift for the effect of a specific fix, which is the same discipline any A/B test requires, applied to a newer and less familiar measurement surface. What to Actually Do, In Order The practical sequence worth following, rather than attempting everything simultaneously: first, confirm actual enrollment status across all three major surfaces individually, since ChatGPT's automatic enrollment doesn't mean Google or Copilot are also active without a manual toggle in Settings, Sales Channels, Agentic Storefronts. This takes a few minutes and immediately clarifies whether the "we're already enrolled" assumption is actually true across every surface or only the one that enrolled automatically. Second, audit product data specifically for the failure modes covered above, vague titles, missing specifications, inconsistent pricing between collection and product pages, before assuming a low AI-referral rate reflects low customer interest rather than a data quality problem an agent couldn't parse past. A useful starting exercise is picking five best-selling products and reading their titles and descriptions as if seeing them for the first time with no other context, the way an agent parsing structured fields would encounter them, rather than as a human already familiar with the brand and catalog. Third, make policy information, shipping, returns, refunds, explicit and easy to find in plain text, not buried in an accordion or JavaScript-rendered element, since several sources specifically flag this as a common technical barrier to agent readability even when the policy itself is perfectly reasonable. Fourth, check channel attribution reporting specifically for AI-referred order volume, rather than assuming it would already be visible in existing dashboards built around traditional marketing channels. Fifth, and lower priority than the first four for most stores, consider expanding JSON-LD schema markup beyond Shopify's default Product schema if development resources allow, since this is the more technical, longer-lead-time improvement that benefits from the foundational data cleanup already being done first. Where This Points None of this preparation work is something Arbyn does on a store's behalf, since it operates downstream of discovery, once a support or sales conversation begins, regardless of which channel produced the customer. What connects directly to Arbyn's own positioning is the volume consequence covered elsewhere on this blog: a store whose AI-referred traffic and orders are genuinely growing at the reported 8x-to-13x range should expect a proportional increase in the support conversations that follow those orders two to five days later, WISMO questions, sizing questions, policy questions, on the store's own channels, not inside whichever AI surface originated the sale. A store that treats the growth numbers above purely as a marketing win, without a corresponding plan for that support volume, is solving discovery and leaving the harder operational half of this shift for later. The honest summary for a Shopify store trying to prepare for AI shopping agents is that the preparation work is almost entirely about data hygiene a store already controls, not a technical integration requiring new development. The stores gaining early advantage aren't the ones with the most sophisticated agentic tooling. They're the ones whose product titles, specifications, and policies were already clean enough for an AI agent to trust before the growth curve made that cleanliness matter, and the ones that connected that same growth to a matching plan for the support conversations it generates, rather than treating discovery and support as two separate, unrelated problems. --- ## Pricing - **Arbyn Starter** - $0/month, permanently free. 150 conversations / month. Resets 1st of each month. - **Arbyn Agent** - $99/month flat, unlimited conversations. Or $990/year (2 months free, saves $198, 17% off). - **There is no trial.** Billing starts immediately on the Agent plan. The free Starter plan is permanent. - The conversation cap is the only difference between plans. There is no feature gating. ## Channels Live today: **support email** and **on-site live chat**. That is the complete list. SMS, Instagram DMs, Facebook Messenger, WhatsApp and Voice are on the roadmap and are NOT live. Arbyn does not edit orders or change line items. Money-moving actions (cancel, refund, discount, gift card, reship, return) require the store owner's approval, and then Arbyn performs them. Running them fully autonomously is a beta authorization and is in development. Shipping address changes are already autonomous. ## What Arbyn does on a Shopify order - **Change the shipping address**: Live. Arbyn does this on its own. Arbyn updates the shipping address on the Shopify order itself, inside the conversation, and writes the change to the order timeline. - **Cancel an order**: Live. You approve it, then Arbyn cancels the order. Anything that moves money waits for the store owner's approval. That is a deliberate control, not a missing feature. Once you approve, Arbyn fires Shopify's order cancellation itself and confirms it to the customer. - **Issue a refund**: Live. You approve it, then Arbyn issues the refund. Arbyn prepares the refund against the original payment method and sends it to you. On approval it files the refund in Shopify. You can cap the value it is allowed to prepare, per channel. - **Apply a discount**: Live. Arbyn creates a real Shopify discount and applies it to the cart, handing the shopper a checkout with the code already on it. It can also issue a discount code on an order once you approve it. - **Send a gift card, or reship an order**: Live. You approve it, then Arbyn does it. Arbyn creates the gift card, or raises the replacement order, in Shopify once you approve. - **Start a return**: Live. You approve it, then Arbyn opens the return. Arbyn opens the return in Shopify on your approval. - **Look up a gift card or store-credit balance**: Live. Arbyn does this on its own. "Do I have store credit left?" is a question most support tools answer with a human. Arbyn reads the balance itself, for a verified customer or from the code they give you, and reports the masked card, the balance and the expiry. If there is no card, it says so rather than guessing. - **Handle a subscription question**: Live. You choose what it does. Arbyn knows which of your products are sold as a subscription, shows that on the product card in the conversation, and sends a subscriber to their subscription management page to pause, skip or cancel. It answers how your subscriptions work from your own knowledge, but it does not read an individual customer's contract, so it will not state their renewal date or status. Most cancels are a customer with product piling up, and the fix is getting them to the page where they can slow the cadence down. Reading the contract itself is on the roadmap. - **Answer support email and live chat**: Live. Arbyn reads every inbound support email and every chat, works out the intent, pulls the live Shopify context, and replies in your brand voice. Money-moving actions (cancel, refund, discount, gift card, reship, return) require the store owner's approval, and then Arbyn performs them. Running them fully autonomously is a beta authorization and is in development. Shipping address changes are already autonomous.