What I learned reading every Shopify Agentic Storefronts doc.

I'm building Flockd, a tool to help Shopify store owners get their products discovered on ChatGPT, Perplexity, Gemini, and other AI shopping channels. I'm doing this with little to no coding experience, mostly evenings and weekends, with a lot of help from Claude.

Step one of building anything is understanding what you're building on top of. So this past week I read every Shopify document I could find about Agentic Storefronts. The announcement post, the help center articles, the developer docs at shopify.dev/docs/agents, and a handful of third-party explainers from people who clearly know more than I do.

Here's what I learned. Some of it surprised me. Some of it I'm still not sure about. I'm writing this down partly to organize my own thinking, and partly because other store owners might find it useful. There's a lot of marketing talk around AI commerce right now and not a lot of plain explanation.

Five things that surprised me

1. Shopify Catalog is doing more work than I realized.

When Shopify activated Agentic Storefronts in March, they didn't just plug your products into ChatGPT and Gemini directly. They built a layer in between called Shopify Catalog. It's a single global index of every Shopify product, and it uses LLMs to automatically categorize products, extract attributes, consolidate variants, and cluster identical items.

That last part matters. If five different Shopify stores all sell the same Hydro Flask, Catalog clusters them together so AI agents see them as variants of the same product rather than five separate options. Which means your differentiation has to come from somewhere other than just listing the product. It has to come from your specs, your context, your brand information, your reviews. Things that distinguish you within the cluster.

2. AI shoppers convert at 31 to 42 percent higher rates than your normal traffic.

I expected AI traffic to be a long-term play. Invest now, see returns in two years. The numbers are way better than that. Shopify's own data shows 11x growth in AI-attributed orders since January 2025, and AI shoppers convert at 31 to 42 percent higher rates than human shoppers. Microsoft Copilot Shopping converts at 194 percent higher than standard search.

That last number should be a big green flag for all online merchants to make sure they are not missing out. People who arrive at your store from a Copilot recommendation buy at almost three times the rate of someone who lands from a Google search. They've already been pre-qualified by the AI. They're showing up ready to checkout.

3. The protocols are abstracted away from merchants, but they exist.

There are now eight competing standards for agent-commerce communication. Shopify's UCP (with Google), OpenAI's ACP (with Stripe), Visa's TAP, Mastercard's Agent Pay, Klarna's Agentic Product Protocol, and a few more. As a merchant on Shopify, you don't have to think about any of this. Shopify abstracts it all away through Agentic Storefronts.

But these protocols exist, and they evolve. UCP was updated in March 2026 to add multi-item carts and live catalog queries. ACP added new merchant capabilities last month. Knowing what each protocol supports tells you what's possible on each AI channel, even if you don't have to write any of the integration code. It's the difference between knowing why some channels can complete checkout in-app and others can't.

4. Marketing copy is actively bad for AI discoverability.

This was the most surprising finding for me. I always thought "make your product descriptions sing" was good ecommerce advice. It still is, for human shoppers. But AI agents do something different.

When a shopper asks an AI to find a lightweight rain jacket under $200 that packs into its own pocket, the AI extracts factual specifications from product data and matches them against the query. "Luxuriously soft premium cotton" tells the AI nothing useful. "100% GOTS certified organic cotton, 200 GSM, machine washable, OEKO-TEX certified" gets matched.

This means the descriptions you wrote to convert humans are working against you in AI search. The fix isn't replacing them. It's adding structured facts alongside them, in metafields and schema, so AI has the data it needs to surface your products.

5. Liquid themes have real structural limitations for AI.

Most Shopify stores, around 95 percent, run on Liquid, the older templating system. A small minority run on Hydrogen, Shopify's React-based headless framework. For everyday human shoppers, the difference is mostly invisible.

For AI shoppers, it's significant. Liquid themes have limited control over JSON-LD schema output, can't easily add the 20+ contextual schema properties AI agents prefer, and have constraints on robots.txt and meta tag behavior. Hydrogen gives full control. Most $10k to $100k a month stores will never migrate to Hydrogen since it's a major rebuild, but they need to know that their theme is leaving signal on the table compared to what's possible.

This is where tools like Flockd matter. We can inject structured data via theme app extensions and app proxies, getting around some of the Liquid limitations without forcing a re-platform.

What I'm still figuring out

I'd be lying if I said I now understand everything. Here's what I haven't figured out yet.

How much of the ranking is determined by Shopify Catalog versus each individual AI platform. When ChatGPT recommends three Shopify products, did Shopify pre-filter them to the top three based on its own ranking signals, or did it send a longer list and ChatGPT chose? The answer affects how merchants should optimize. The docs are vague on this. I'll be running experiments to figure it out.

The Knowledge Base App. Shopify mentions a Knowledge Base App that stores brand policies, FAQs, and brand voice for AI agents to reference. I can't tell yet whether this is a separate Shopify product, a subset of metafields, or a planned feature. Either way, it's clearly important. AI agents need brand context to answer customer questions correctly during a conversation.

How fast ranking signals propagate. If you fix your structured data today, when does it actually show up in ChatGPT recommendations? A day? A week? A month? The cycle time matters enormously for measurement. Nobody seems to have a clean answer yet.

The attribution dashboard's real coverage. Shopify says orders flow into the admin with full AI channel attribution. But "AI traffic that didn't convert" or "AI conversations where you were considered but not picked" probably isn't data that exists yet at the merchant level. Which means proving your AI shopping investment is paying off is harder than the marketing makes it sound.

What this means for store owners reading this

If you run a Shopify store doing $10k a month or more and you've heard "AI shopping" thrown around but haven't done anything about it, here are five things you can check or fix this weekend.

Check if Agentic Storefronts is enabled in your admin.

Go to Settings, then Sales Channels. If you sell to US shoppers and your store is eligible, you should see an Agentic Storefronts option. Verify which AI platforms are toggled on (ChatGPT, Copilot, Perplexity, Google AI Mode). Most US stores have it enabled by default but not everyone has confirmed.

Audit your top 10 products' descriptions.

Read them as if you were an AI agent extracting facts. Are there specific specifications, materials, dimensions, certifications, use cases? Or is it mostly adjectives like "luxurious," "amazing," "the best"? If it's the latter, those products are leaving AI ranking on the table.

Check your metafields.

In Shopify admin, go to Settings, then Custom Data. Are your products using metafields for things like material composition, sizing, ingredient lists, technical specs? If most of your products have empty metafields, AI agents are flying blind.

Verify your JSON-LD output.

Open one of your product pages, view the page source, search for "application/ld+json". If you don't see a JSON block with full Product schema (price, availability, brand, identifier, etc.), your structured data is incomplete or missing.

Test it yourself.

Open ChatGPT and ask it to recommend products in your category. Are you in the results? If not, who is? What do their listings look like? This is the single most useful exercise you can do.

What's next

I'll keep posting weekly. Next week I'm digging into the OpenAI and Perplexity sides of this. How they actually rank products, what their docs say (or don't say) about merchant signals, and what the differences mean for stores that want to show up in both.

If you want to skip the research and just see how your store currently scores against these signals, that's what Flockd is for. Install it on Shopify — 14-day free trial, no credit card required.

Otherwise, see you next week.