If you're investing heavily in your storefront in an effort to win over AI shoppers, you're spending in the wrong place. In all actuality, the inputs that decide whether an agent finds and recommends your product are:
- catalog taxonomy
- attribute depth and consistency
- product content
Your theme isn't on that list.
Here's what that means for your store, in the order the work should happen.
Your variants have to be real
Agents navigate a catalog through structured queries, not visual browsing. Taxonomy depth, variant naming, and metafield completeness decide whether those queries return anything at all.
The Universal Commerce Protocol (UPC), which Shopify co-developed, is an open standard for how agents discover, negotiate with, and transact with any merchant. The design principle the team emphasized: merchants declare the capabilities they support, including bespoke functionality, and agents discover and negotiate against those declarations.
For example, we helped build evo’s search/discovery layer, including the catalog taxonomy that decides which product and variant fields become filterable facets.
Pro-tip: If a shopper can choose a finish on your site but that finish isn't a distinct, buyable variant underneath, your configurator is describing a product your data can't confirm. An agent hits the same wall, and quietly recommends someone else.
Your attributes have to agree with each other
Inconsistency reads as unreliability. If material is a metafield on some products, a tag on others, and buried in the description on the rest, there's nothing stable for a query to land on.
This is hardest for multi-brand retailers, because the inconsistency arrives with the data. Our work with evo included variant-level attributes like normalized color and trip type, plus dynamically-sourced recommendation modules filtered by product metafields.
Shopify Catalog acts as the structured layer under agent discovery: products standardized into a universal taxonomy, with pricing and inventory verified in real time and syndicated to connected AI channels.
Keep in mind: Catalog will infer what you didn't declare (likely less accurately than you would).
Normalizing attributes during a migration costs a fraction of retrofitting them afterward. If you're replatforming, this is the moment to do so.
“Category taxonomy and metafield population are the first things I check for. They are the fastest improvements a merchant can make for optimizing their catalog.”
–Tyler Shockey, Solutions Engineer and Developer, ehouse
Your content decides who gets chosen
Structure gets you found. Content gets you chosen.
Product content that contains a unique product selling point, keywords, and specs is going to help the Catalog sort and recommend products. This can be accomplished by modifying the product description or mapping a metafield to description in the catalog mapping tool.
An agent answering the query, "what will this moisturizer actually do for my skin?" can only work with what you wrote or what others said about you. Content that lists ingredients without explaining outcomes gives an agent nothing to recommend with.
The bigger opportunity is the content you've probably buried behind a footer link: returns policy, guarantee, warranty terms, sourcing, care instructions, and the reasons you do things differently. That's exactly what an agent reaches for when it compares two options. It's also one of the few places a smaller brand can outshine a larger one, because specificity beats scale here.
Shopify's Knowledge Base app is a concrete place to start. It surfaces the FAQs agents use to answer questions about your store, shows how often your store information is requested, and lets you write custom answers.
Pro-tip: the auto-generated answers are inferred from your settings and policies, and reviewers have reported real errors, including wrong shipping coverage. Fill it in deliberately rather than trusting the defaults.
FAQs from merchants
1. Does this replace SEO?
No, it changes what the work consists of. Keyword placement matters less while structured, server-rendered product data and a consistent point of view matter more.
2. Is this a project or a posture?
Both. The catalog and taxonomy work is a scoped project. Keeping attributes consistent as you add products is an operating habit, and that's the part that decays.
3. We're a small brand, are we already outgunned?
Not on this. Specific, honest, detailed content beats a larger competitor's generic listing, because the agent is reading for substance rather than authority alone.
4. Do we need UCP set up now?
For discovery, mostly defaults. For negotiation and cart management, someone has to map your business logic to the protocol. Worth knowing where you sit before a customer asks.
