- 1 day ago
- 12 min read
Sign up for Wix Studio today and prepare your clients for the future of eCommerce.
A shopper lands on your product page with Gemini open in Chrome's side panel. Before they scroll, they tap a saved prompt: "Summarize the reviews, compare this to the top three alternatives, and tell me if it's worth the price." The AI parses your page, pulls context from across the web, and hands them a verdict. You never see the prompt. You never see the comparison. You just see a bounce—or, if the verdict goes your way, a conversion.
That's not hypothetical. It's what a growing share of desktop shoppers are quietly starting to do now that Gemini is native to Chrome, and this browser "co-pilot" flow has a default distribution channel wired into most of the web.
I work on agentic commerce at Wix, and this piece is my best read of where merchants should focus today. I'll be honest up front: nobody has this fully figured out. The agentic protocols, the consumer habits, the checkout plumbing—all of it’s evolving in real time.
From working closely with our merchants, I know the topic can be overwhelming. So, I'll cover what exactly agentic shopping is, what the main flows look like, why you should care about it, and how to practically prepare your store. What follows are the steps I'd take if I were running an online store this quarter, knowing we'll re-learn parts of it next quarter. I'll keep this article updated as the landscape shifts.
What is agentic commerce?
Agentic shopping is the use of AI agents to handle shopping tasks—like product discovery, price comparison, and sometimes, the purchase itself—with varying degrees of autonomy on behalf of a human shopper. "Agentic" means these aren't just recommendation engines. Agents pursue goals, make decisions, take actions, and adapt.
Agentic shopping isn’t one thing. It's a spectrum that runs from AI helping shoppers compare options to AI negotiating a discount to AI checking out on a shopper's behalf.
For instance, AI applications like ChatGPT or Gemini are directly integrated with the structured product feed data of an onboarded merchant's catalog, so it bypasses the store's website entirely. In agentic browsers, it navigates the website directly, reading its pages, reviews, and policies the way a savvy shopper would. Either way, the store's products are being evaluated on how well their information matches what the shopper needs.
What agentic commerce means for merchants
The shift to the agentic web is accelerating faster than most merchants realize, and distribution is the reason. Gemini is now embedded in Chrome's side panel, providing a co-pilot shopping experience reaching across most of the desktop web.
Skills for Gemini in Chrome—as in, reusable, one-click AI workflows, many of which are shopping-focused—are turning ad-hoc AI use into habitual consumer behavior. Browsing in AI Mode inside Google Search does the same thing from the other direction: the shopper starts in AI Mode, opens your page in a side panel, and the page itself becomes part of the AI's context.
The headlines on platform-mediated checkout have been more cautious. OpenAI scaled back Instant Checkout in March 2026 after low adoption and operational issues. But platform-mediated discovery (i.e. when ChatGPT, Gemini, or Perplexity recommend products matching a shopper's intent) and browser-level co-pilot shopping are both scaling aggressively. That split matters for where you invest first.
Gartner predicts that by 2028, 60% of brands will use agentic AI to facilitate streamlined one-to-one interactions. As we'll see, these interactions already happen on different surfaces for different flows. As a merchant, you should be aware of this so you can ensure your products are set up to compete for those agents' attention.
Three primary flows for agentic commerce
Today's agentic commerce is taking shape across three flows. Two are well-established patterns, and a third is emerging.
These flows differ by the protocols that shape them:
UCP and ACP for platform-mediated commerce
WebMCP and NLWeb for browser co-pilot shopping
A2A for agent-to-agent coordination.
They also differ by how involved the human shopper is.
01. Platform-mediated commerce
A shopper uses an AI platform (like ChatGPT, Gemini, or Perplexity) that has direct access to merchant catalogs through feeds, APIs, or agentic protocols like the Universal Commerce Protocol (UCP) and Agentic Commerce Protocol (ACP). The agent discovers, compares, and recommends inside the AI interface. In some cases, checkout happens directly on the AI platform, but most of the time, the shopper is redirected to the merchant's site to complete the purchase.
Human involvement: moderate. The shopper sets the intent and confirms the outcome. Everything in the middle (discovery, filtering, comparison) happens autonomously.
02. AI browser co-pilot shopping
A shopper works alongside an AI agent that lives in the browser, e.g. Gemini in Chrome, Opera's Neon, or OpenAI's Atlas. The agent visits your site, reads your pages, reviews, and policies, and either presents options or proceeds through checkout. Your content, site structure, and checkout experience all matter, not for a human's eyes, but for the AI's ability to parse and act.
Human involvement: highest. The shopper is present throughout, directing, reviewing, and deciding. The AI handles the legwork.
03. Agent-to-agent commerce (emerging)
The shopper's personal agent talks directly to the merchant's agent through protocols like A2A. Imagine a discount-negotiation flow where a shopper’s agent receives an intent structured as a blend of constraints. Hard constraints—specs, price ceilings, or delivery windows—are non-negotiable; soft constraints define the optimization goals, such as finding the lowest price or the highest merchant rating.
The agent identifies candidates satisfying every hard requirement before ranking them against soft preferences. This is where trade-offs surface: a familiar, highly rated merchant might be slightly pricier than a cheaper, unproven alternative. To resolve this, the agent either breaks the tie via a secondary factor like shipping speed or initiates a peer-to-peer negotiation with the merchant's agent.
In this A2A path, the shopper’s agent presents loyalty data while the merchant’s agent evaluates it against a plain-English policy configured during onboarding rather than hard-coded logic. If the terms align, a personalized coupon is automatically applied. The shopper receives a tailored deal, and the merchant never had to write a single line of discount code.
Human involvement: minimal, but not absent. The shopper sets preferences and constraints. The agent may come back to the human for decisions that exceed the agent's authority: approving an unexpected offer, completing a 3D Secure (3DS) challenge, confirming a substitution when the original item is out of stock.
As human involvement decreases, agent autonomy increases. This progression maps closely to McKinsey's "automation curve," which describes six levels of purchasing delegation from "cognitive sidekick" (the co-pilot flow) through "supervised executor" (platform-mediated eCommerce) to "network autonomy" (the A2A vision).
How to prepare your online store for agentic commerce
This part used to be a flat list. It isn't, really. Some of these actions are table stakes—the baseline every merchant needs to do regardless—and some are investments that pay off only if your shoppers are actually coming through a particular flow. I'm splitting them into two tiers for that reason.
Your foundation (every merchant should do this, regardless)
01. Understand your shoppers' profile
Start with what you sell, the single best predictor of who shows up and how they behave.
Routine, task-oriented products (household staples, consumables, replenishment). Shoppers wanting to buy such products often show up in utilitarian mode. Shopping is a task to finish, not a pastime. Delegation climbs the curve quickly here; operational reliability matters more than brand narrative.
Complex, research-intensive products (electronics, travel, insurance). For such products, shoppers often want AI to do the analytical legwork, comparing specs, monitoring prices, and evaluating trade-offs. But they make the final decision themselves. Platform-mediated discovery excels here: AI narrows the field, the human picks the winner.
Identity-driven and experiential products (luxury, fashion, home décor, milestone purchases). Shoppers want to actively discover such products, browse, compare aesthetics, and make a personal choice that reflects who they are. An agent that bypasses that process defies the point of the purchase.
Individual shoppers cross modes: a utilitarian shopper will still agonize over a vacation package, and a recreational shopper still reorders groceries on autopilot. That's fine. If what you sell is toilet paper, utilitarian mode is the dominant persona that matters to you.
Importantly, there's actual research behind this: The "word-of-machine" effect shows that consumers prefer AI for utilitarian purchases, resist it for hedonic ones, and accept hybrid "augmented intelligence" across the board. Consumer-willingness studies show roughly 70% of consumers are willing to use AI agents for shopping, with only about half of those ready to delegate both routine and research-driven purchases.
You’ll need the output of this step to determine the flow your shoppers are most likely to use, in the next section.
02. Structure your product data for AI discovery
Structured data is important regardless of the flow your shoppers will use. To make sure your store’s structured data is optimized for agents:
Ensure your site ships complete structured feeds for every field that agents might need to filter on: price, availability, identifiers, shipping, variants, return policies, and so on.
Choose specific, factual descriptions over marketing copy.
Keep inventory and pricing accurate in real time.
Add schema markup (JSON-LD for Product, Offer, and MerchantReturnPolicy) so agents and crawlers can extract structured data; the same markup feeds NLWeb for natural-language queries.
For platform-mediated and A2A agents, structured data is the only input. These agents never render your pages, so structured data is the only thing they have to work with. This step is table stakes for the flows that can't see your site and a quality floor for the ones that can.
For co-pilot agents, structured data is a quality floor, not a requirement. Co-pilot agents are fundamentally page-reading agents, which means they can work off of:
Rendered DOM and visible text (including prices, specs, and reviews written for humans)
Screenshots + vision models (increasingly capable, they parse layout, images, and even UI state)
Inferred structure from headings, tables, and human-readable patterns.
So, if your product page has a clear <h1> heading with the product name, a visible price, a spec table, and a reviews section, a capable co-pilot agent will handle "summarize the reviews and compare this to three alternatives" just fine. This is part of the whole point of the co-pilot flow: it meets the shopper on pages built for humans.
Nonetheless, if your site allows co-pilot agents to access structured data, it will ensure they work reliably, cheaply, and based on the right price, variant, and availability.
More specifically, the two main reasons co-pilot agents will work better with structured data are:
Disambiguation and accuracy: Vision + DOM parsing is good, not perfect. Variants (size/color), promo vs. base price, "in stock at your ZIP" vs. generic availability, and bundle pricing; these are exactly where unstructured pages mislead agents. JSON-LD and a clean feed resolve ambiguity the agent would otherwise guess at (and sometimes guess wrong).
Speed and token cost: A structured feed or JSON-LD blob is cheaper and faster for the agent to consume than rendering + OCR + reasoning. In a side panel where the shopper is waiting three seconds for a verdict, faster wins.
03. Prioritize based on your most expected flow
With the foundation in place, here's where to invest first:
If most of your shoppers will show up in co-pilot mode, following the steps described in make your site ready for AI agents (below) is your highest-impact next move.
If most will show up through platform-mediated discovery, your foundation already covers the biggest lever. Double down on feed depth and real-time accuracy, then look at how to monitor your AI visibility (below).
If your category is A2A-natural (for example, weekly-reorder groceries), the need to have merchant agent capability moves up the queue sooner for your store.
The items below apply to everyone. The emphasis and sequencing don't.
04. Make your site ready for AI agents
A co-pilot agent will read your site whether you help it or not. The question is whether it reads it precisely and can act on your behalf, or whether it guesses from pixels and hands the shopper a best-effort summary. This section is about moving from the second to the first.
For browser agents and similar tools that navigate websites on behalf of shoppers, your site is the interface. NLWeb turns structured content into a conversational layer agents can query. LLMs.txt gives machine-readable context about your site and grounds agents interacting with your domain. WebMCP, increasingly supported in browsers, lets sites expose structured tools (add-to-cart, check-shipping, apply-coupon) to agents as function calls, so the agent can take action. Keep specifications, reviews, policies, and pricing in crawlable text (not only images). Design checkout to work programmatically with minimal friction.
05. Monitor your AI visibility
Start from the uncomfortable truth: you have almost no visibility into what a shopper's side-panel AI actually says about your products, how it compares you to competitors, or the moment the shopper decides to leave. Traditional analytics don't capture it, because the decision happens inside an interface you don't own.
Generative engine optimization (GEO) is the emerging discipline here: understand how often AI platforms cite your brand, what sentiment they associate with you, and how you compare in AI-generated recommendations. Evaluating your LLM traffic in GA4 is a practical starting point. What you watch depends on which flow you prioritized: ChatGPT and AI Mode citations, browser-referral signals, or eventually agent-to-agent traffic indicators. Attribution in AI search is still incomplete, but the field is measurable and improving.
06. Start building your merchant agent capability (a look at what's next)
For most merchants, a full merchant agent is next-year work—unless your category is A2A-natural, in which case you'll want to be early. The on-site precedent already exists: Amazon's Rufus and Walmart's Sparky are on-site shopping assistants configured by the merchant, and Wix Smart Chat plays the same role for SMBs.
These are on-site assistants today, not A2A agents. But they're the natural on-ramp: a merchant who can configure negotiation rules, returns, and cross-sell logic in plain language now will have a running start when the shopper at the other end is another AI rather than a human.
How Wix prepares your store for agentic commerce
Agentic commerce is multi-channel by nature. Your products may be discovered on Perplexity, purchased through a ChatGPT merchant app, or negotiated by an A2A protocol, all in the same week. Here's how Wix keeps your store visible and transactable across those channels. Most of it’s automatic.
Wix is a proactive agentic partner
Many agentic commerce optimizations are happening at scale. Wix is an official signatory for the Agentic Commerce Protocol and has active partnerships with agentic commerce leaders like Stripe, Google, PayPal, and OpenAI. Wix is creating the infrastructure, partnerships, and features that will bring agentic commerce capabilities to its users as this technology develops.
Wix automatically structures your site data
When you set up your Wix store, Wix automatically structures your site data. Your store product catalog is, by definition, a structured representation of all your products' attributes, e.g. options, prices, stock availability, and so on. Your job is to continuously update and enrich the product catalog and site content; Wix will take care of the rest. Remember, the more you invest in this, the easier it will be for shopping agents to interact with your store.
Additionally, Wix generates information like schema.org structured data markup to provide more context to agents via RAG and tools like NLWeb.
Eligible Wix stores can syndicate products into AI shopping channels
For eligible stores, Wix is already syndicating catalog, pricing, and inventory to AI-powered shopping platforms so products are discoverable on surfaces like Perplexity and PayPal's AI shopping assistant. Wix is also integrating Stripe's Agentic Commerce Suite to add a standardized backend for checkout and fulfillment across additional AI platforms. You stay the merchant of record, your brand stays visible, and new AI channels come online without re-integration.
Wix makes your site readable to AI agents
Every Wix site includes Site MCP, a Model Context Protocol (MCP) server that agents can use to discover products, business details, blog posts, and services. Eligible Wix sites are automatically configured for NLWeb, turning structured content into a conversational interface agents can query (for example, size, price, and availability questions) using your product data. An LLMs.txt file is automatically generated for eligible sites to provide a machine-readable site overview. With these built-in features, agents get multiple reliable paths to understand your store.
Wix provides a merchant agent capability
Wix Smart Chat is an early-stage merchant agent today: any visitor, human or AI, can interact with it through your site's chat. Wix Smart Chat trains on your content and rules, guides buyers, and bridges AI with human handoff. As agentic commerce evolves, this is the foundation for more autonomous interactions across A2A and external AI surfaces.
Wix Analytics helps you monitor your AI visibility
Wix’s AI Visibility Overview lets you see how your sites are performing in AI search. Use it to learn how often your site is cited by AI platforms, monitor brand perception, and see whether competitors are surfacing for the same queries. Read more about Wix and Wix Studio’s GEO features here.
Agentic commerce is still early, and the full picture is still coming into focus. A loyal customer's own agent will one day talk to your store's agent directly, negotiate a personalized discount, and drop a generated coupon in the cart in seconds. You won't write discount logic for it—you'll configure a skill in plain English and let the agent handle the rest. That future is visible from here, but it's not where most merchants need to focus today. Instead, get your foundation right, prioritize the flow your shoppers are already using, and you'll be ready for what comes next.
Sign up for Wix Studio today and prepare your clients for the future of eCommerce.
