Back to blog
    • Comercio agéntico
    • E-commerce
    • ChatGPT
    • Google
    • Meta
    • SEO
    • Tiendas online

    Google, ChatGPT and Meta now sell for you: what changed this week and what your store must do before the holidays

    In eight days, Meta launched an agent that shops on its own, OpenAI opened ChatGPT ads to 31 European countries and Google put your store's cart inside its search engine. Here's what it means for a normal online store, what applies today outside the US, and where to start without overspending.

    11 min
    Cliente pagando una compra online desde el móvil

    What happened in eight days

    Between September 8 and 16, 2026, the three platforms that decide where people shop announced, almost at the same time, how they're going to sell inside their AI assistants:

    • Meta (September 8) launched Muse, a personal agent that searches for products, enters any store with its own browser, fills in the checkout and pays with Stripe Link once the user approves. On September 10 it connected Shopify's full catalog.
    • Google (September 16) expanded the Universal Commerce Protocol in Merchant Center: shoppers build the cart inside Google (Search, AI Mode, Gemini) and finish in your store with the cart already loaded. Plus AI sales analytics, a brand agent inside YouTube ads and new feed attributes that feed AI Mode answers.
    • OpenAI (September 16) launched Sponsored Agents (your brand's agent inside a ChatGPT ad), campaigns created in natural language and a ChatGPT Ads app for Shopify, in the US from the 16th and in 31 European markets, Canada, Australia and India from September 23.

    If you run an online store, this isn't a tech story. It's a change in the place where your customers will decide what to buy.

    Why it matters even if you don't sell in the US

    Almost everything spectacular (the agent buying on its own, the cart travelling from Google to your store) works only in the US today. It would be easy to conclude "this isn't for me yet". That would be a mistake, for three reasons:

    1. The model is already settled: discover in AI, buy on your site. OpenAI tested paying inside ChatGPT and pulled it in March 2026. What remains is this: the assistant decides which stores it shows, and your product page and checkout decide whether you sell. That's already happening in your country with ChatGPT search, Gemini and Perplexity.

    2. ChatGPT ads reach Europe next week. With a correct product feed you can advertise on ChatGPT from September 23 in Spain, France and the rest of Europe. Whoever has the feed ready that day buys traffic their competitors don't know exists.

    3. All three platforms read the same thing. A catalog feed in Google Shopping format, structured data on every product page, a robots.txt that lets AI crawlers in, and a checkout an agent can complete. Build that once and you're ready for Google, OpenAI and Meta at the same time, and for whatever they open in your country next year.

    The five pieces that work for all three platforms

    After reading all three specifications, this is what they have in common. It's the list I give every client with a store:

    1. One master catalog with stable identifiers

    Each product's id must be the same in the feed, in the page's structured data, in the Meta pixel and in the checkout. Sounds obvious and it's the most common failure: three systems with three different identifiers, and no platform manages to match your page with your product. A feed in Google Shopping format (id, title, description, link, image_link, price, availability, brand, gtin) feeds Merchant Center, Meta's Commerce Manager and ChatGPT Ads with the same file.

    2. Product and Offer structured data on every product page

    Google uses it to verify price and availability; OpenAI and Meta read it when the agent visits the page. With name, images, a description over 150 characters, brand, sku, gtin, price, currency, availability, shipping policy and return policy. That last point is what almost every store leaves out, and without it agents drop you: they won't buy from a site that doesn't say how returns work.

    All rendered on the server, inside the HTML. AI crawlers don't run JavaScript.

    3. A robots.txt that lets in the ones that answer

    OAI-SearchBot and ChatGPT-User (OpenAI), Google-Extended (Gemini), meta-externalfetcher and meta-webindexer (Muse and Meta AI), PerplexityBot, Claude-SearchBot. Many stores block them without knowing because a security plugin or the CDN's anti-bot mode does it by default. An agent that can't load your product page buys from another store. I wrote about this in more detail in how to get ChatGPT and Perplexity to recommend you.

    4. A checkout an agent can complete

    Guest checkout with no mandatory registration, no blocking CAPTCHA or SMS verification before paying, forms with standard autocomplete, final price visible with taxes and shipping, and wallets: Google Pay (what Google uses), Stripe Link (what Muse uses), PayPal. Every bit of friction you put in front of a human you also put in front of the agent, and the agent has no patience.

    5. Content the models cite

    Descriptions with real specifications, not just adjectives. FAQs per product (Google just created a feed attribute, question_and_answer, exactly for this). Real, visible reviews. An "About us" page with legal name, address and contact. An up-to-date sitemap.xml and, optionally, an llms.txt at the root with the catalog map and policies.

    Two-minute test: open ChatGPT or Perplexity and ask "where can I buy [your best-selling product] with shipping to [your city]?". If your store doesn't show up, or shows with a wrong price or availability, you know what the problem is. And your competitors are in the same situation, so the advantage goes to whoever fixes it first.

    What applies today in Europe and Latin America

    So you don't waste time on what doesn't exist in your country yet:

    PlatformAvailable now outside the USStill US only
    GoogleMerchant Center with a complete feed, conversational attributes, structured data, AI performance insights (by country)Cart transfer, native checkout, Business Agent in Search and YouTube
    OpenAIChatGPT Ads with a product feed (from September 23 in Europe), organic citations in ChatGPT search if your site is crawlableSponsored Agents, organic ACP feed (approved partners only)
    MetaCatalog in Commerce Manager, product tagging, Business Agent on WhatsApp, Instagram and Messenger (global since June, pay per use)Muse, Meta AI shopping mode, one-tap checkout in ads

    Look at the left column: it's longer than it seems. A well-built feed, complete structured data, ChatGPT Ads in Europe and a Meta agent that answers and sells on WhatsApp are things you can have running this month.

    The four-week plan

    It's the order I follow with my clients. Each step depends on the previous one; skipping one is the surest way to get feeds rejected.

    Week 1: data. A unique, stable identifier per variant. Normalized catalog with every required and recommended field (shipping, returns, gtin, quality images, grouped variants). XML feed at a stable URL with a 15-minute cache.

    Week 2: pages. Product/Offer JSON-LD with shipping and returns on every product page, robots.txt open to AI crawlers, sitemap with real dates, shipping, returns and privacy policies at fixed URLs. Validate with Google's Rich Results Test until zero errors.

    Week 3: accounts and feeds. Merchant Center verified with the feed connected and no rejections. Meta Commerce Manager with the same feed. ChatGPT Ads account with the feed uploaded (minimum 100 products). Meta Pixel and Conversions API with the same ids. Google's conversational attributes in a supplemental source.

    Week 4: agents and measurement. Meta Business Agent configured with catalog, FAQ, hours and escalation rules to a person. Different UTM parameters in each feed's URLs (utm_source=chatgpt, utm_source=meta_ai) so you know where each sale comes from. Analytics segments by referrer and by agent user-agent.

    What's US-only (agent checkout, cart transfer) gets prepared in a sandbox for when it opens in your country. No rush, but no reason to start from scratch that day either.

    What it costs

    Indicative figures for a store with 200 to 5,000 products on Shopify, WooCommerce, Medusa or a custom build:

    • Readiness audit (exactly what you're missing, with priorities): from €900.
    • Weeks 1 and 2 (data, feed, JSON-LD, robots, policies): €2,500 to €6,000 depending on platform and catalog state.
    • Weeks 3 and 4 (accounts, connected feeds, pixel, Business Agent, measurement): €1,500 to €3,500.
    • Agent checkout (UCP and ACP, for when it opens in your country): €4,000 to €9,000, and today it only makes sense if you sell in the US.

    Compared with what a mid-sized store spends on advertising every month, it's a one-off investment that decides whether you exist in the channel where more and more people ask what to buy. Service details are in online stores that sell and the online store solution.

    If you sell services, not products

    Part of this applies. Product feeds are designed for goods with a fixed price, but a service gets in through three routes: as a catalog item when it has a closed price (a session, a package, a subscription), through conversational agents (Meta's Business Agent on WhatsApp books appointments and qualifies leads, and already works worldwide) and through organic search with Service and FAQPage structured data, a visible price and a declared "book" action on the page. It's exactly what this website has, which is why I wrote a specific guide for service businesses.

    Mistakes I'm already seeing

    Waiting for it to "mature". ChatGPT ads open in Europe on September 23. Stores with a feed ready that day will have data and sales while the rest start asking questions.

    Copying the Google feed to OpenAI as is. OpenAI's visibility and checkout flags don't copy over from Merchant Center; they have to be added.

    Blocking crawlers "to protect the content". I understand the intent. The cost is not existing for whoever asks the AI.

    Neglecting the return policy. It's the field that gets the most feeds rejected and the one that makes the most agents drop you. It's ten lines of text at a fixed URL.

    Ignoring attribution. Purchases made by Muse arrive as "direct" traffic. If you don't put UTMs in the feeds and don't watch direct traffic that converts unusually well, you won't know this channel is already selling for you.

    In short

    This week Google, OpenAI and Meta made clear how they're going to sell inside their assistants, and all three ask for the same thing: a clean catalog, complete structured data, access for their crawlers and a frictionless checkout. Most of that already works in Europe and Latin America, and what doesn't gets prepared now and switched on when it opens.

    If you have a store and want to know exactly what you're missing, book a 15-minute call. I'll tell you what I'd do first, what applies in your country today and what can wait. No commitment, and no selling you what you don't need.

    Carlos Canelón

    Written by

    Carlos Canelón

    Technology partner for business owners

    Twenty years building software for companies. I help owners and executives save time and sell more with automation, AI and custom systems. Working with clients in Spain, France, the United States and Venezuela.

    Share this article

    Contact me on WhatsApp