Something odd started showing up in client calls this year. Traffic flat or slipping, rankings holding steady, ad spend unchanged, and nothing obviously broken anywhere in the funnel.
Then someone on the team opens an AI assistant, asks it to recommend a product in their own category, and watches three competitors come back by name. Their own store, thousands of SKUs deep, never gets mentioned.
Quick answer: AI assistants skip ecommerce stores whose product data they can’t read or trust. To prepare your store for AI shopping, complete your product attributes, validate your Product schema, expose returns and shipping as structured data, keep price and stock current, and confirm you aren’t blocking AI crawlers. Protocols come later.
Getting this right is a plumbing problem more than a marketing one. Your product feed and your structured data are what these systems read, and your beautifully designed category pages mostly aren’t. It’s the least glamorous work a Magento development company in New York does, and right now it’s the highest-leverage. Run the eight checks further down this page, fix whatever fails, and worry about protocols after that.
Wait, Didn’t AI Shopping Already Flop?
Fair question, and plenty of people quietly wrote this off in the spring.
OpenAI launched Instant Checkout in ChatGPT in September 2025, letting US shoppers buy from Etsy sellers without leaving the chat, with a million Shopify merchants promised as next. By March 2026 it was gone. Barely a dozen Shopify merchants had ever gone live with it, and OpenAI pulled the in-chat purchase flow to focus on product discovery instead.
A lot of people read that as the end of the story. It was closer to the opposite.
The cash register part collapsed. The part where an assistant decides which three brands get named has only grown, and buying went back to where it always was, which is your site. So AI now runs your discovery, and your storefront closes the sale for traffic you never touched, from a shopper who’s already been shown two competitors alongside you.
Google went the other way and doubled down. In January 2026 it launched the Universal Commerce Protocol, an open standard for agent-led shopping built with Shopify, Etsy, Wayfair, Target and Walmart, and backed by more than twenty other companies including Stripe, Visa, Mastercard and Best Buy. Google also rolled out a branded Business Agent that shoppers can chat with directly in Search, plus a set of new Merchant Center attributes built for conversational shopping.
Here’s where the two standards sit today.
| ACP (OpenAI) | UCP (Google) | |
| Launched | September 2025 | January 2026 |
| What it does now | Product discovery inside ChatGPT | Agent led discovery and checkout across Google surfaces |
| Where the buyer pays | Your own checkout, since March 2026 | Your own checkout, with you as seller of record |
| Who’s behind it | OpenAI with Stripe, Etsy, Shopify | Google with Shopify, Etsy, Wayfair, Target, Walmart and 20+ others |
| What you need first | Product data an assistant can read and trust | A Merchant Center feed plus the new conversational attributes |
Notice what both columns have in common. Whatever happens with the protocols, the money lands on your site and the entry ticket is clean product data.
What These Systems Read
Not your homepage, and not the redesign you signed off on last year.
They read structured product data: title, attributes, price, stock, shipping, returns, compatibility, and whatever review signals they can pull from elsewhere. If that data is thin, stale, or contradicts what your site says, the assistant does the sensible thing and recommends a merchant whose data it can trust.
The upside is real. Adobe’s analytics team reports that generative AI drove a 693.4% increase in traffic to retail sites over the 2025 holiday season. Those referrals converted 31% better than other traffic sources, AI driven revenue per visit climbed 254% across the season, and the shoppers who arrived that way spent 45% more time on site.
Traffic from AI tools is still small next to your other channels and worth considerably more per visit. That gap is the argument for moving early.
The Five-minute AI readiness Audit
Grade yourself. Eight questions, honest answers.
1) Is every product in your feed complete? Not just title and price. Material, dimensions, compatibility, what it’s used for. Missing attributes are the most common reason a product never gets shortlisted.
2) Does your product schema validate? Run a handful of product pages through a structured data test. Broken markup is invisible to you and fatal here.
3) Are your returns and shipping terms machine-readable? Buried in a policy page written in 2019 doesn’t count. Assistants weigh delivery certainty heavily.
4) Do price and stock update in near real time? Recommending an out-of-stock item burns the assistant’s credibility, so these systems learn to avoid feeds that do it.
5) Are you accidentally blocking AI crawlers? Check robots.txt and your CDN rules. We’ve found plenty of stores blocking the exact bots they were trying to attract.
6) Can anything query your catalog live? A static feed handles discovery. Anything past that needs an API surface.
7) Does your storefront load quickly on mobile? The shopper an assistant sends you arrives with two competitor names already in hand and very little patience.
8) Can you see agent traffic separately in analytics? If AI referrals are pooled into direct or organic, you can’t tell whether any of this is working.
Reading Your Score
Failed one to three, mostly data questions. You’re in the normal range. This is a catalog hygiene project, and it’s the highest return work available to you this year.
Failed four or more. Your products are likely invisible to these systems today. Start at question one and ignore everything you’ve read about protocols until the data is clean.
Passed everything. Rare, in our experience. Stop building and start measuring which queries you appear for.
Start with the Boring Fix
Everyone wants to talk about protocols. The answer is almost always attribute hygiene.
Most catalogs we inherit have decent data on the top hundred sellers and thin data everywhere else. The long tail is where AI discovery pays, because those are the specific, high-intent questions shoppers type into an assistant instead of a search box. “Waterproof work boots, size 13, wide fit, under $200” is a query your category page can’t answer and your attribute data can.
It’s unglamorous work. As an Adobe Commerce development agency, we spend more project time on attribute mapping and feed logic than on anything that looks impressive in a demo, and it’s the part that moves the numbers.
Why Page Speed Matters More After an AI Recommendation
You already know slow sites lose sales. What’s changed is who’s arriving.
The shopper arriving from an AI recommendation has less loyalty to you than any visitor you’ve ever had. They didn’t find you. Something else did, and it handed them two alternatives in the same breath. A four-second load gives them an easy reason to go check option two.
Frontend weight has been the standing problem on Magento for years, which is why so much Magento Hyva theme development work has moved from nice-to-have into the default recommendation. Stripping the legacy JavaScript stack out of the storefront tends to do more for Core Web Vitals than another round of caching tweaks.
Why Your Magento Version May Be the Blocker
If you’re on an older Magento build, most of what’s above isn’t available to you at any price.
Older versions can’t expose the API surface these systems expect. Unsupported PHP blocks the extensions you’d use. Every patch turns into a negotiation with custom code someone wrote in 2018 and didn’t document. Teams end up paying twice, once to bolt something on and again to undo it during the upgrade they were putting off.
This is the conversation we have most often, and it’s never the one the client booked. Nobody wants to hear that a Magento migration service is the answer to an AI question. It frequently is.
What Adobe Is Doing, and What It Leaves To You
Adobe committed Adobe Commerce to supporting UCP and ACP in February 2026, building on earlier support for Google’s Agent Payments Protocol. The intent is that merchant catalogs, pricing and inventory become readable by AI agents without a custom integration per platform.
Worth reading the fine print, though. Those capabilities are arriving across Adobe’s cloud products first, and what that means for you depends entirely on how you’re hosted.
| Your setup | What Adobe is shipping you | What your team builds |
| Adobe Commerce as a Cloud Service | Protocol support as it rolls out | Feed quality and attribute work |
| Adobe Commerce on cloud | Support arriving, timing varies by release | The same, plus upgrade planning |
| Magento Open Source or self-hosted | Nothing announced | The endpoints, the feed, all of it |
If you’re on that third row, there’s no module sitting in a queue with your name on it. Someone on your side builds the endpoints, and that someone needs to understand both your catalog logic and the protocols. That gap is where a lot of mid-market merchants will quietly lose 2026, then arrive in 2027 competing against stores whose data these systems have already learned to trust.
Who Should Act Now, and Who Can Wait
If you sell forty items of one-off inventory, protocol readiness isn’t your bottleneck. If your catalog is small, your margins are thin, and your repeat rate is healthy, your time is better spent almost anywhere else. If you sell into a category where buyers already know exactly which brand they want, AI discovery changes less for you than the headlines suggest.
The merchants who should move now are the ones with deep catalogs, comparison heavy categories, and buyers who describe a problem rather than name a product. Tools, parts, appliances, furniture, B2B consumables. If a customer has ever emailed you asking which model fits their setup, an assistant is answering that question for someone else right now.
FAQs
Why don’t AI assistants recommend my store?
Because they can’t read your product data or don’t trust it. Assistants shortlist merchants whose attributes are complete, whose schema validates, and whose price and stock are current. Thin or stale data gets passed over in favor of a competitor the system can verify.
How do I prepare my ecommerce store for AI shopping?
Start with product data. Complete your attributes, validate your Product schema, publish returns and shipping as structured fields, and keep price and inventory synced in near real time. Then check that you aren’t blocking AI crawlers in robots.txt or at your CDN.
Do AI assistants replace my checkout?
Not since March 2026, when in-chat purchasing was pulled back and checkout returned to merchant sites. These systems now send buyers to you. Being skipped happens earlier, at the moment an assistant decides which brands to name.
Does site speed affect whether AI assistants recommend a store?
Indirectly. Speed affects how reliably crawlers render your pages, and it affects the shopper who lands after a recommendation already holding two competitor names. A fast storefront won’t rescue an incomplete product feed.
How much does it cost to prepare an ecommerce store for AI shopping?
Treat any figure as a planning range rather than a quote. Catalog and schema work for a mid-size store is usually measured in developer weeks and lands in the low five figures. A platform upgrade sits in a different bracket and should be scoped separately. Any Magento migration company quoting a firm number before auditing your extensions is guessing.
Can older Magento versions support AI shopping?
Usually not without work. The question is whether your build can expose live inventory and pricing to an external system securely, and on older releases the answer is often no. Unsupported PHP and undocumented custom code are the two most common blockers.
