For a while, it felt like every conversation about AI and ecommerce was heading in the same direction.
Websites were going to need new infrastructure built specifically for AI. Shopify had Universal Commerce Protocol, or UCP. People were adding llms.txt files to their websites. There was a growing idea that brands needed to make their websites “AI readable” in a completely different way to traditional search.
A few months later, it all feels a lot quieter.
You don’t hear many merchants talking about UCP. llms.txt doesn’t seem to have become a meaningful part of how ChatGPT or other AI platforms find information. And when you actually use these tools, a lot of the experience still feels surprisingly similar to normal search.
So, did all of this infrastructure just die?
Not exactly.
UCP didn’t disappear, it went underneath the surface
Shopify’s Universal Commerce Protocol is probably the more interesting of the two.
UCP was introduced as an open standard for allowing AI agents and commerce platforms to communicate with each other. In practical terms, it gives an AI a structured way to do things like search a product catalogue, create a cart, move through checkout and understand order information.
That’s still useful.
What has changed is that Shopify merchants don’t necessarily need to think about it.
Shopify is increasingly handling this infrastructure at the platform level through products like Shopify Catalog and Agentic Storefronts. Rather than every merchant implementing a new protocol on their website, Shopify can expose structured commerce information directly to platforms like ChatGPT, Google and other AI assistants.
This makes UCP far less visible.
From the merchant’s perspective, it can look like nothing is happening. Underneath that, Shopify is trying to position itself as the commerce infrastructure sitting between millions of stores and AI agents.
It’s also worth noting that there probably won’t be one universal protocol controlling everything.
OpenAI has its own Agentic Commerce Protocol. Shopify has UCP and its broader agent infrastructure. Other platforms will develop their own standards and integrations.
That’s fairly normal at this stage of a new technology.
The important distinction is what these protocols are actually trying to solve.
UCP isn’t really designed to answer a question like:
“What are the best running shoe brands?”
That type of query still relies heavily on search, content, reviews, editorial websites and other information available across the web.
UCP becomes more useful when the question turns into:
“Find me a pair of black running shoes under $200, check if they’re available in my size and add them to my cart.”
That is where structured commerce infrastructure starts to matter.
llms.txt is a different story
llms.txt was pitched as something closer to a robots.txt file for AI.
A website could create a simple file explaining what content was important, where useful information lived and how an LLM should understand the site.
In theory, it made sense.
In reality, there still isn’t much evidence that major consumer AI platforms are relying on it in any meaningful way.
When ChatGPT needs current information, it can search the web, retrieve relevant pages and interpret them directly. Google obviously already has an enormous search index. Other AI search products are building similar retrieval systems.
They don’t necessarily need a website owner to create a separate map explaining what they should read.
There’s also a more fundamental problem.
If I publish an llms.txt file saying that my agency is the best Shopify agency in New Zealand, why should an AI trust that?
It still needs independent signals.
It needs third-party articles, reviews, case studies, relevant content and other sources to work out whether that claim is credible.
This is why I think llms.txt was probably overhyped.
There’s nothing particularly wrong with having one. If it takes five minutes to generate, there’s very little downside.
I just wouldn’t build an AI search strategy around it.
AI search still looks a lot like search
One of the more interesting things about using ChatGPT Search and similar tools is how familiar the underlying behaviour actually feels.
The interface is completely different, but the information still has to come from somewhere.
For a lot of informational queries, the process is still broadly:
Search for relevant information.
Find useful pages.
Read those pages.
Compare different sources.
Then generate an answer.
That changes some aspects of SEO, but it doesn’t suddenly make good websites, useful content or authority irrelevant.
In some ways, it makes them more important.
If an AI is synthesising an answer from five or ten different sources, you want your brand to either be one of those sources or be consistently mentioned by them.
Ecommerce is splitting into two different AI problems
This is probably the distinction that matters most.
There are really two different things happening under the broad label of “AI commerce.”
The first is AI discovery.
Someone asks ChatGPT:
“What are the best skincare brands in Australia?”
“Which sofa brands have good quality but aren’t ridiculously expensive?”
“What are the best Shopify agencies in New Zealand?”
That world still looks heavily influenced by traditional search, editorial content, reviews, authority and brand mentions.
The second is AI transactions.
Someone asks:
“Find this product in a medium.”
“Show me options under $150.”
“Add this to my cart.”
“Can it arrive before Friday?”
This is where structured product feeds, APIs, Shopify Catalog, UCP, ACP and similar infrastructure become much more important.
They’re related problems, but they’re not the same problem.
And I think a lot of the early AI ecommerce conversation blurred them together.
What should ecommerce brands actually focus on?
For most Shopify merchants, I wouldn’t recommend spending a huge amount of time chasing every new AI optimisation trend.
The fundamentals are still more valuable.
Make sure your website can be crawled properly.
Have useful product information.
Use clear product titles, descriptions and attributes.
Keep pricing and availability accurate.
Use good structured data.
Create genuinely useful content around the questions customers are asking.
Build authority outside your own website through PR, reviews, partnerships and third-party mentions.
And make sure your Shopify catalogue is properly structured so Shopify can distribute that information into emerging AI commerce channels.
Then, sure, add an llms.txt file if you want.
Just don’t confuse having an llms.txt file with having an AI strategy.
The bigger opportunity for Shopify
What’s actually more interesting to me is Shopify’s position in all of this.
Shopify already sits underneath an enormous amount of ecommerce infrastructure.
If consumers increasingly start shopping through AI assistants rather than navigating directly to individual websites, Shopify has an opportunity to become the commerce layer powering those transactions.
The merchant may own the product.
ChatGPT or Gemini may own the interface.
But Shopify can still own a large part of the infrastructure in between.
That’s probably why UCP matters even if nobody is talking about it.
It was never particularly important whether merchants knew what the acronym meant.
What matters is whether an AI agent can understand a Shopify catalogue, find the right product, know whether it’s available and ultimately complete a transaction.
So I wouldn’t say UCP has died.
If anything, it’s becoming less visible because it’s becoming infrastructure.
llms.txt, on the other hand, currently looks much closer to an idea that got ahead of actual adoption.