- 15 hours ago
- 6 min read
Most eCommerce merchants are preparing for agentic shopping by optimizing the wrong thing. They're updating homepages. Refreshing product page copy. Thinking about UX.
None of that is wrong, exactly. It's just that AI agents don't browse sites the way humans do. They read your structured data, form a belief about your product, and make a recommendation before a human ever sees your page.
This makes your product feed your primary persuasion layer. In this piece, I’ll explain the importance of managing these data feeds for agentic shopping, and what to do to optimize websites for the agentic web.
First, what is an agentic shopping feed?
A product feed is the structured data file that tells platforms, search engines, and now, AI agents exactly what you sell. An agentic shopping feed is optimized, not for human browsers or search engines, but for AI agents making autonomous purchase decisions on a user's behalf.
How agents are changing product discovery
When a human considers buying something, they browse, read reviews, scroll images, get a feel for the brand…Then, they make a call that’s part emotional and part rational.
AI agents don't do any of that.
An agent processing a purchase query pulls structured data from your feed and forms a probabilistic view of your product. It’s not reading your brand story. It’s not feeling your visual identity. It’s running your structured attributes against a set of criteria, and filtering your product in, or out, of its recommendation. The buyer's journey is replaced by the agent's logic.
The agent is operating in what we'd call validation mode rather than exploratory mode. It arrives at your product with a framework already running, not with an open mind. Your job is to satisfy that framework.
This changes what your feed needs to do. Fields that used to be compliance checkboxes are now persuasion signals. Agents put heavy consideration on attributes like:
Returns language
Warranty data
Delivery certainty
Review volume
These aren't administrative details anymore. They're the inputs an agent uses to decide whether your product is trustworthy enough to recommend.
For merchants looking to get involved with agentic shopping, the early-mover window is genuinely open right now. According to Search Engine Land, 63% of AI agent purchase journeys selected the first search result, highlighting that although position/ranking is difficult, being selected first is vital.
Two protocols are shaping this space. Google's Universal Commerce Protocol (UCP) powers buying experiences in AI Mode and Gemini. OpenAI's Agentic Commerce Protocol (ACP), built with Stripe, powers product discovery and checkout inside ChatGPT. They're separate standards, but the feed quality requirements underpinning both are the same: complete, accurate, structured product data.
Wix is an official signatory for the Agentic Commerce Protocol (ACP) and works closely with Stripe and OpenAI to ensure that Wix merchants are positioned to participate in agentic commerce from the start.
Google UCP requirements
To participate in Google's UCP, Google recommends creating a supplemental data source in Merchant Center rather than modifying your primary feed to avoid formatting issues that could affect standard product ingestion. The key feed-level changes are:
Add the native_commerce attribute to each product you want eligible for UCP-powered checkout. This is a product-level boolean; set it to true to opt a product in, and false or absent to exclude it. Only products with this attribute will display a ‘Buy’ button on Google's AI surfaces.
Define your returns policy in Merchant Center, including return cost, return window, and a link to the full policy. This is a Merchant of Record requirement and will be referenced on the checkout screen.
Also include your customer support information in Merchant Center. It will be used to generate the "Contact Merchant" link on the order confirmation page.
UCP-powered checkout is currently available to eligible US-based merchants, with the onboarding experience rolling out gradually. Google will notify you if you need to update your feed attributes, returns, or account configuration.
OpenAI / ChatGPT (ACP) requirements
OpenAI's approach is worth addressing separately as it’s still moving. To make products discoverable in ChatGPT, merchants provide a structured product feed in CSV or JSON format containing identifiers, descriptions, pricing, inventory, media, and fulfillment options. Price and availability are required fields, while reviews, rich media, and shipping options improve ranking and user trust.
Instant Checkout in ChatGPT is currently available to approved partners, and merchants can apply at chatgpt.com/merchants. That said, OpenAI has been shifting its checkout model away from direct in-chat purchasing toward merchant-built ChatGPT apps, so the implementation path may look different depending on when you are reading this. What has not changed is the underlying requirement which is a well-structured, accurate, regularly updated product feed.
How to audit and manage your data feeds for agentic commerce
All the conversion rate optimization work you've done on your product pages still matters for humans. For agents, what matters is the quality, completeness, and specificity of your structured data. This is where the practical work sits.
Audit your current feed
Work through your top twenty products and ask three questions for each:
Are your returns and delivery fields populated with specific, actionable language rather than a redirect to another page?
Is your delivery range as tight as your fulfillment genuinely supports?
Is review volume sufficient to generate statistical confidence, and is that data feeding into your structured output?
Once you know where the gaps are, you can close them quickly with the tips that follow.
Certainty wins every time
Uncertainty in your data isn't neutral; it reads as a negative signal. A vague returns policy registers as risk to human users and AI agents. Anchoring effects compound this: whichever products appear first in an agent's shortlist, usually those with the clearest, most complete data, tend to stay there.
Be explicit in your returns language
"See website for details" is the fastest way to get filtered out. Agents are looking for certainty, and vague policy language registers as a risk signal. "Free 30-day returns, no questions asked" is parseable. "Please review our returns policy" is not. Audit every product in your feed and replace any vague returns references with specific, structured language.
Your returns policy is visible in Google Shopping results too. Being explicit here benefits every user who sees your listing, not just AI agents.
Commit to a delivery window
It's worth noting that Google Merchant Center's shipping fields require a range rather than a single figure. Make the range as tight as your fulfillment genuinely supports. “Delivered in 3 to 3 working days” signals far more certainty than “delivered in 3 to 7 working days.” The narrower the window, the stronger the signal.
Review volume matters more than review score
Agents treat review volume as a confidence signal, not just social proof. A product with 400 reviews at 4.2 stars will typically outperform a product with 12 reviews at 4.9 stars in agent-mediated recommendations. The logic is probabilistic: more data points mean lower uncertainty. If your feed is thin on review data, that's the first thing to address.
The most reliable way to build volume quickly is a post-purchase email sequence. A well-timed request sent two to three days after delivery consistently outperforms requests made at the point of purchase. Keep the ask simple and direct. Here’s how to add reviews to your Wix website.
Check what your feed is actually outputting
Wix generates structured product data automatically for most pages, but you may want to check and customize the structured data in some cases. Go into your product catalogue and check the fields that agents weigh most heavily: returns policy, delivery promise, and review count.
Write for two audiences simultaneously
Your product titles, descriptions, and attribute fields need to work for both human browsers and agent parsers. That means specific language over evocative language in structured fields. Save the brand voice for your descriptions. In the data fields, be precise, be explicit, and be consistent across your entire catalogue.
Early adoption matters in agentic shopping
The stores already optimized for agentic criteria will be the ones surfacing when the traffic flows.
Meanwhile, merchants who get deprioritized by AI agents won't receive a notification. There's no ranking drop they can point to in a dashboard. Their products will simply appear less frequently in agent-generated recommendations, and the revenue impact will show up weeks later with no obvious explanation attached.
The feed has always mattered. What's changed is that it now matters for an audience that doesn't forgive vagueness, can't be charmed by great photography, and isn't going to give you the benefit of the doubt. So get specific, be explicit, and do it before your competitors do.
