E-commerceAI Automation

AI Shopping Agents and Your NZ Online Store: What Changed in 2026

Kage Works20 August 20267 min read

On 16 February 2026, OpenAI opened "Buy it in ChatGPT" to more than a million Shopify merchants, including SKIMS, Spanx, Glossier and Vuori, letting a customer discover a product and complete the purchase without leaving the chat window. Three weeks later, OpenAI quietly pulled it. By the time reporting caught up with what happened, only around a dozen of those million-plus sellers had built the integration to make an in-chat purchase work (Value Added Resource).

That's the part of the AI shopping story that didn't survive contact with real shoppers. A quieter shift is already moving real traffic and real orders through small stores that have nothing to do with ChatGPT specifically, and it's the one worth your time this month.

OpenAI shipped it, then quietly pulled it back

Instant Checkout launched in September 2025 with Etsy as the first partner, built on the Agentic Commerce Protocol (ACP), a payments standard OpenAI developed with Stripe and open-sourced. Walmart followed. The February 2026 expansion to Shopify's merchant base was the mainstream moment: any of the platform's million-plus sellers could, in principle, let a ChatGPT user buy from them without a browser tab.

By March 2026, OpenAI had pulled it. The company's own explanation was narrow and specific: "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we're allowing merchants to use their own checkout experiences while we focus our efforts on product discovery" (Exploding Topics). Reporting on the decision pointed to the underlying problem: shoppers kept asking ChatGPT what to buy and comparing options, then wanted to finish the purchase somewhere with their saved card and order history already sitting there, and OpenAI hadn't solved fraud detection, real-time inventory sync or sales tax compliance at the speed a live checkout demands.

If you'd read the February headlines and started building an in-chat checkout flow, you'd have spent March ripping it back out. That's the risk of building for the feature getting all the headlines rather than the one still standing a quarter later.

The half of the plan that's still running

Discovery didn't get paused. ACP is still active, and Target, Sephora and The Home Depot have since integrated for product discoverability inside ChatGPT rather than in-chat purchase. A separate, broader standard launched the same year: Google's Universal Commerce Protocol (UCP), announced by Sundar Pichai at NRF's January 2026 keynote and co-developed with Shopify, Etsy, Wayfair, Target and Walmart, with Visa, Mastercard, Stripe and American Express among more than 20 endorsing partners. UCP is open-sourced under Apache 2.0 and wired into Google AI Mode, the Gemini apps, and Microsoft Copilot Checkout (Shopify Engineering).

Both protocols do the same underlying job from different directions: they give an AI agent a structured, machine-readable way to find out what you sell, what it costs, and whether it's in stock, so the agent doesn't have to guess by scraping your page. That's the part that's proving durable. Checkout-in-chat is the part that turned out to be harder than the demo suggested.

Shopify's own numbers tell a narrower story

Shopify reported AI-referred traffic to its merchants up roughly 8 times year over year in Q1 2026, with AI-referred orders up nearly 13 times, and new-buyer orders arriving through AI channels at close to twice the rate of traditional organic search. By Q2 2026, both traffic and orders from AI channels had tripled year over year again. Half of AI-referred sessions landed a shopper directly on a product page, two and a half times the rate for traditional search, which sends more people through a homepage or category page first (TechCrunch).

One figure in that report matters more than the growth rate for a small NZ retailer specifically: 75% of AI-attributed purchases fell outside Shopify's top 100 product categories. Google's front page rewards the biggest, most-searched category leader. An AI agent asked to find "a wool baby blanket made in New Zealand" is doing something closer to a specific product match than a popularity contest, a different game for a niche or specialist retailer than the SEO race most have been losing to bigger competitors for years. Shopify's president, Harley Finkelstein, has framed the shift as additive rather than a replacement for search: traditional search sessions are still up 1.3 times over the past two years and hold roughly a third of all storefront sessions.

Read the two data points together and the picture is narrower and more useful than "AI shopping is here." Agents are increasingly finding specific products for people who already know roughly what they want, and small, specific catalogues benefit disproportionately from that kind of query. They are not yet reliably completing the purchase without a human finishing it in a normal checkout.

Your data matters more than your design right now

Shopify's data shows catalogue-powered AI searches, meaning searches where the agent reads structured product data rather than scraping a rendered page, convert at roughly twice the rate of AI searches relying on scraped or outdated information (Shopify). If you're on Shopify, this mostly happens without you doing anything: Shopify Catalog automatically shares your product data with ChatGPT, Gemini and Microsoft Copilot. If you're on a custom-built or non-Shopify store, that syncing doesn't happen on its own, and the work involved is markup rather than layout: schema.org Product and Offer data covering price, availability, ratings, shipping and returns, plus a GTIN on every item so an agent can match your product to the right listing elsewhere.

An agent that reads a price from your structured data and finds a different price at checkout won't flag the discrepancy or give you the benefit of the doubt. It stops recommending you. Roger McSaveney, general manager of Auckland agency Forge Creative, put it to industry press this way: "an agent that finds a price in your structured data which does not match the price at your checkout will quietly drop you" (Newswire). Treat that as one practitioner's read rather than measured data. The underlying mechanism matches how these agents are built to work: they trust structured data until it's wrong once.

New Zealand businesses are moving fast on AI generally and slow on the governance around it, which is worth knowing before you bolt anything onto a live store. The Employers and Manufacturers Association surveyed more than 300 member businesses in July 2026 and found 83% already using AI in some form, against only 13% with an actual AI policy in place (EMA). That survey covers AI use broadly, not shopping agents specifically, but the gap it describes, adopting fast and checking the details later, is the same gap that left OpenAI cleaning up a checkout flow three weeks after launch.

This month's checklist

If you're on Shopify, confirm Shopify Catalog is syncing your current data (price, stock, variants) correctly rather than assuming it's automatically right because it's automatic.
If you're not on Shopify, audit your product pages for schema.org Product and Offer markup, and add GTINs where they're missing. This is a data project, not a redesign.
Make your checkout price the single source of truth. Whatever feeds your structured data should read from the same place your checkout does, not a separate spreadsheet or a marketing page that gets updated on a different schedule.
Don't build a bespoke in-chat checkout integration on the strength of a headline. The one that shipped first got pulled within a month.
Recheck this in six months. Both ACP and UCP are under a year old and the merchant requirements are still moving.

None of this replaces a normal, working checkout on your own site, and nothing here suggests you need a chatbot bolted onto your storefront. The businesses benefiting right now are the ones whose product data was already accurate, and who didn't have to do anything special when the agents started reading it. If your store's catalogue was built or is maintained as part of a larger platform decision, our comparison of Shopify against a custom-built store covers where that data-ownership question sits either way.