Why structured product data will shape how brands grow, how customers find products, and why creative thinking should start in your catalogue
OpenAI and Shopify are working on a native integration that will let users browse and check out products directly inside ChatGPT.
It’s a development worth watching closely. Not because it’s another shiny AI launch - but because it changes something fundamental about how product discovery might work.
For years, Shopify stores have largely relied on interruption-based discovery: paid social, retargeting, email. If you wanted someone to find your product, you had to work out how to get in front of them.
But ChatGPT changes the model. It allows discovery to begin with a conversation - and more importantly, with intent.
What matters in that kind of environment isn’t how catchy your ad is. It’s how well your product data is structured.
Discovery is becoming intent-first
Let’s say a user types:
“I’m looking for a birthday gift under £40 for someone who likes running and hates plastic packaging.”
That’s not a search query. It’s not a category page visit. It’s a conversation - and it’s rooted in a specific set of constraints, preferences, and values.
The only way to answer that kind of request accurately is with data that reflects those kinds of distinctions. That means going beyond just “running accessories” or “gift ideas” and into things like:
- Who the product is for
- What use cases it supports
- What values or constraints it aligns with
- How it’s positioned, both functionally and emotionally
If your store’s data isn’t structured in a way that surfaces that information clearly and consistently, you’re unlikely to appear -no matter how good your product is.
The limits of traditional architecture
This shift creates a real challenge for most stores.Shopify’s admin interface encourages a flat, tag-based approach to product data. In practice, that often means brands end up with hundreds of unstructured tags and very little consistency in how attributes are applied.
The result is that your product data is “technically there”, but almost impossible to use effectively - whether that’s for filters, feeds, search, or now conversational interfaces like ChatGPT.
This isn’t a new problem, but the AI shift makes it harder to ignore.
If a customer describes a context or constraint in plain language - and your store hasn’t captured that in a structured, machine-readable way - your products likely won’t make it into the results.
Brands need to think like systems designers
This is where we believe the real opportunity lies. It’s not about AI as a feature or a bolt-on. It’s about using AI as a forcing function to get your product data systems in order.
That starts with clarity:
- What do we know about each product?
- How consistently is that information captured?
- What use cases, audiences, and values do we support —and where is that reflected in the data?
- Can that information be used across filters, feeds, internal search, schema, and third-party tools?
The brands that succeed in this next phase of commerce won’t be the ones with the most creative campaigns. They’ll be the ones with the cleanest, clearest systems.
A shift in how we think about creative
There’s also a mindset change here. Traditionally, we’ve treated product data and creative messaging as separate things - one for developers or ops, the other for marketing.
But when you look at how LLMs interact with sites, that division starts to break down.
The things we think of as brand storytelling - delivery promises, differentiators, value props, even things like urgency - all need to be structured, repeatable, and portable. In many ways, they need to become part of the product data model.
So we start to ask questions like:
- What’s the best format to describe urgency in areusable way?
- Can we standardise how we explain what makes aproduct sustainable?
- How do we tag products that are good for gifting —not just based on price, but suitability?
This is creative, but it’s creative work done through the lens of data engineering.
Laying the ground work now
Right now, we don’t know exactly how ChatGPT will ingest Shopify product data. But it’s safe to assume structured information will be more useful - and more durable than unstructured copy or inconsistent tags.
The good news is that if your internal structure is sound, remapping is relatively straightforward. You can always rename fields or shift attributes. What’s harder is going back later and trying to clean up messy, inconsistent data at scale.
So the goal isn’t to get everything perfect now - it’s to make sure your structure supports clarity, flexibility, and future adaptability.
Tools that might become part of the stack
We’re still early in this shift, but a few tools are already proving useful as brands start to prepare.
- Matrixify is still one of the simplest ways to make large-scale changes to Shopify product data quickly and reliably.
- Airtable, paired with tools like AirSys, is becoming a lightweight way to enrich and manage product attributes —particularly for teams that want to work outside the Shopify admin.
- Espresso is an interesting option for building structured interfaces on top of Shopify’s product layer, especially if conversational or structured discovery takes off.
We may also see more brands lean on headless-style setups, not for frontend flexibility, but to allow better control over their data architecture. But the key point is this: your current platform doesn’t need to change to make progress - your structure does.
What this all comes down to
We’ve said before that taxonomy isn’t just about navigation - it’s about how your store communicates with machines.
The same goes for tags, attributes, and product detail pages. These aren’t just for internal organisation. They’re part of how your store gets read, interpreted, and surfaced.
And while this new ChatGPT + Shopify integration isstill in its early stages, it’s pointing clearly in one direction: product data is no longer just a backend consideration. It’s the foundation for future discovery.
For brands already thinking this way, it’s a validation. For those still treating data as admin work, now’s the time to rethink.