Are you guys actually trusting ai to handle live order creation yet, or are you still keeping a human in the loop for the final checkout?
3 Likes
Maybe a nope…the fear of data leakage is so big
2 Likes
This question has gotten a lot more concrete since February. Here’s where things actually stand:
The infrastructure is ready. The UCP spec (Universal Commerce Protocol, by Google and Shopify) now covers the full journey: product discovery, multi-item carts, checkout, order management, fulfillment, discounts, and identity linking. Google AI Mode is live with checkout for Etsy and Wayfair. Shopify turned on Agentic Storefronts for all merchants on March 24.
Merchant readiness is mixed. At Shoptalk last week (March 24-26), the tone was surprisingly humble. Executives openly admitting they don’t know what to prioritize. Projections range wildly - eMarketer says 8.8% of ecommerce will be agentic by 2029, Merkle says up to 50%.
On @Christinebeauty’s data concern - the UCP spec actually addresses this. Payment credentials flow one-directionally (platform to merchant only) using opaque tokens. The AI agent never sees raw card data. And for merchants worried about exposing product/pricing data, the Catalog capability lets you control exactly what the agent can query in real time.
The real blocker isn’t security or trust - it’s product data quality. AI agents parse structured data fields, not web pages. Merchants with clean, complete product data (titles under 150 chars, full variant info, real-time inventory) are getting orders now. Everyone else is invisible to agents.
2 Likes
The new Flow
When someone asks ChatGPT “recommend me a good yoga mat under $50,” ChatGPT can now show Shopify products and let the buyer checkout without leaving the chat. This works for Gemini and Copilot too.
Your products are included by default. You don’t need to do anything to “turn it on.”
What it means
4 million Shopify stores are now competing for the same AI recommendations. The AI doesn’t show all of them it picks the ones with the best product data.
Hygiene Check
Detailed product descriptions (not just “Blue shirt - size M”)
Proper product categories assigned (Standard Taxonomy)
Alt text on images (AI reads this to understand your products)
Meta descriptions filled in
Complete variant information (sizes, colors, materials)
Think of it like Google SEO but for AI. The stores with richer, more structured data get recommended first
What you can do to be Agentic AI ready
Go to Products in your admin. Open your top 20 products.
Does each one have a detailed description (50+ words)?
If not, expand them.
Check if you have product categories assigned. Go to any product → the “Product category” field. If it says “Uncategorized,” fix it.
Add alt text to your product images. AI Vision models read this.
Fill in your SEO meta descriptions. Every product should have one.
These 4 things take an afternoon and will put you ahead of 80% of merchants who won’t bother.
Hope this helps!
Rahul
2 Likes
Still keeping humans in the loop for anything that actually commits an order. Too many variables - fraud, inventory, payment auth - for AI to own that fully yet.
Where it makes sense is earlier in the flow: recovery emails, flagging abandoned carts, that kind of thing. Lower stakes if it gets it wrong.
What’s your use case? That’d change the answer a bit.
2 Likes
not about trust. about reversibility.
if an action can be undone, AI runs it. if it can’t, a human approves it first. simple…
order creation on a live store has no undo button. that’s where the human stays.
that’s actually the model we built ShopOS on. agents handle what’s reversible. humans gate what isn’t. the squad decides which is which.