In late 2025 it looked like we’d be shopping inside ChatGPT within months. In 2026 the story changed several times. For an eCommerce director in Mexico, Colombia or Argentina, the noise is huge and the question is simple: what do I need to do, and when?
This article lays out what happened, separates what matters from what doesn’t and proposes a 90-day plan.
What happened in 12 months
- September 2025. OpenAI and Stripe introduce the Agentic Commerce Protocol (ACP) and, with Shopify, Instant Checkout: buying without leaving the chat.
- January 11, 2026. Google, together with Shopify, Etsy, Walmart, Target and Wayfair, launches the Universal Commerce Protocol (UCP) at the National Retail Federation conference (Findskill).
- March 4, 2026. OpenAI retires native Instant Checkout in ChatGPT. According to the same source, only a dozen Shopify merchants had fully integrated it.
- March 2026. Shopify enables Agentic Storefronts, which connect catalogs to ChatGPT and other models by default (FlowHunt).
- April 2026. OpenAI moves toward models where the merchant controls checkout: the assistant discovers and recommends, and the purchase closes on the merchant’s site or app (Checkout.com).
What it means, in one sentence
AI is already the storefront; checkout is still yours. The value lies in being chosen by the agent, not in having a buy button inside the chat.
The scale is already visible in the US data: Salesforce estimates that AI and agents influenced 20% of global Cyber Week 2025 orders, about $67 billion in sales, mostly through product recommendations and conversational service (Salesforce, December 2025).
That has a practical consequence: the work for 2027 isn’t integrating a new protocol overnight. It’s getting your catalog, prices and stock into a state an agent can read, compare and trust.
LATAM is different
Three differences from the US that change priorities:
- The weight of Mercado Libre. It’s the main channel for many categories, and as of late April 2026 it had not joined UCP or ACP (Findskill). For quite a while, your presence in AI will depend on how assistants read your public listings, not on an integration.
- Local platforms. Tiendanube, the largest D2C platform in the region, had not formally joined either protocol as of that date either (Findskill). If your store runs on Shopify, the path is more direct; if it runs on VTEX or Adobe Commerce, ask your partner about the platform’s plan.
- Payments and trust. Cash payments at retail chains already account for 26% of online payment methods in Mexico (Marketing4eCommerce). A purchase flow delegated to an agent has to live with that reality.
The test we run with every brand: can an agent buy your product today?
To know how ready a brand is, we assess it across six dimensions:
| Dimension | Question | Red flag |
|---|---|---|
| Findable | Does the agent find your product when asked for the need, not the name? | You only show up if the user types the brand |
| Readable | Does your site have Product schema with price, stock and brand? | Listings without structured data, or with the price inside an image |
| Fresh | Are price and availability up to date on your site and feeds? | The agent recommends an out-of-stock pack size |
| Consistent | Does the data match across your site, retailers and marketplaces? | Three different prices for the same SKU |
| Accessible | Does your robots.txt let in the AI crawlers that work in your favor? | Legacy blocks nobody has reviewed |
| Purchasable | Does the path to purchase lead to an official store? | The agent sends shoppers to an unauthorized seller |
Most brands we assess fail on “fresh” and “consistent” before any protocol issue.
90-day plan
Days 1 to 30: data hygiene.
- Review your robots.txt and decide explicitly which AI crawlers you let in.
- Implement or fix Product and Offer schema on all your listings.
- Take an inventory of price, pack size and claims by SKU on your site and your top three retailers, and fix the differences.
Days 31 to 60: feeds and catalog.
- Connect your catalog to Google Merchant Center with price and stock updated several times a day.
- If you’re on Shopify, review your Agentic Storefronts setup. If you’re on VTEX or Adobe Commerce, ask your partner about the platform’s plan for UCP and ACP.
- Complete product attributes on marketplaces: they’re what an agent uses to compare.
Days 61 to 90: measurement.
- Measure your AI Shelf with real shopping prompts in ChatGPT, Gemini and Google AI Mode.
- Log where each assistant sends shoppers to buy your product.
- Name an owner for “agentic commerce” on the team. Today, at almost every company, it belongs to no one.
What we don’t recommend
- Building your own protocol integration before your platform solves it. The standard is still moving.
- Waiting. What weighs most (clean data, complete listings, consistency) serves SEO, the Digital Shelf and AI at the same time. Nothing is lost by doing it today.
In short
Agentic commerce is moving slower and messier than announced, but it’s moving. The advantage will go to brands whose catalog an agent can read and trust.
At Balloon we assess your brand’s readiness across the six dimensions and tell you what to fix first. Get your assessment.
FAQ
What is agentic commerce?
It's a purchase made by an AI agent on the user's behalf: it searches, compares and completes the transaction. Today it coexists with a more common model, where AI recommends and redirects to the merchant's site to close the sale.
Which protocol should my brand adopt, ACP or UCP?
It depends on your platform. Shopify and Google push UCP; ACP comes through Stripe. In practice, most brands in LATAM will handle it through their eCommerce platform or payment gateway, not with their own integration.
Does this apply if I sell mostly on Mercado Libre?
Yes, but differently. As of late April 2026, Mercado Libre had not joined UCP or ACP. In the meantime, what you do control is the quality of your listings, your product data and your own store.