Shopify Just Gave Your Store an AI File. Don't Overwrite It.

In May 2026, Shopify auto-generated three files for eligible stores: /agents.md, /llms.txt, and /llms-full.txt. On May 28, they shipped the ability to customize them.

The advice already circulating in founder communities is: go customize it. Add your brand story. Tell the AI what makes you different.

That advice is wrong. The kind of wrong that leaves you feeling productive while quietly making things worse.

TL;DR: Shopify auto-generated an AI-readable file for your store. Customizing it means fully replacing it, no merge possible. For most stores the right move is to leave the default alone. For app builders, the real opportunity is making your own site readable to AI agents, and thinking about what operational context your app could contribute to merchant files.

Who this is for: Shopify merchants who are tempted to customize their new /agents.md, and app founders thinking about AI discoverability.

Core problem: The “go customize it” advice circulating in founder groups causes merchants to replace a Shopify-maintained file with hand-written copy they’ll have to maintain forever, for no concrete benefit.

What these files actually are

AI agents that shop, compare, and recommend need a machine-readable way to interact with a store. The /agents.md file is Shopify’s answer. It sits at yourstore.myshopify.com/agents.md.

Go read one and you’ll find it isn’t marketing. It’s a protocol map. It points agents to Shopify’s Universal Commerce Protocol endpoints (a discovery endpoint at /.well-known/ucp and an MCP endpoint at /api/ucp/mcp), walks through the steps an agent follows to search, build a cart, and check out, and spells out the rules, including that checkout requires human approval. It also lists the read-only routes for pulling your catalog as JSON.

Here’s the part worth knowing before you touch it: that file is identical on every Shopify store. Allbirds and a one-person coffee brand serve the same boilerplate, with only the domain swapped. It contains nothing specific to your products. Its entire value is the protocol wiring, the part that lets an agent transact with you at all.

/llms.txt and /llms-full.txt exist for the same reason. Despite what the names imply, both point to the same content by default. There is no “full version” with deeper data. They’re just two entry points to the same file.

The llms.txt standard was proposed by Jeremy Howard and the team at Answer.AI in September 2024, specifically to give language models a reliable format for understanding sites. Shopify adopted that convention for /llms.txt and /llms-full.txt.

/agents.md is a separate convention, not derived from the llms.txt standard. It’s Shopify-specific.

Before you read another word of this, go check your own file. yourstore.myshopify.com/agents.md. Read what Shopify generated for you. That file is exactly what the advice going around tells you to throw away.

The replace problem

When you create a custom Liquid template at templates/agents.md.liquid, it replaces the Shopify default for that path.

Not merges. Not extends. Replaces.

The fallback hierarchy works like this: path-specific template first, then your agents.md template if a path-specific one doesn’t exist, then Shopify’s generated default if neither exists. The moment you add a custom template, the default is gone for that path.

So if your custom file says something like “We’re a sustainable apparel brand focused on ethical sourcing and conscious consumerism,” that brand paragraph is now the entire thing an agent sees. The UCP discovery endpoint, the MCP endpoint, the checkout flow, the read-only catalog routes, all of it is gone, and there’s no way to pull it back. The agent that could have bought from you now reads a mission statement and moves on.

There is no Liquid object that renders the default inside your template. You can’t append to it. You can’t extend it. If you want to add your own context, you have to paste Shopify’s entire protocol block into your template by hand and keep it in sync from then on. The moment you create a template, you own the full output of that path, forever.

Which means you also stop inheriting whatever Shopify improves in the default later. They’re actively building out agentic commerce. The default will get better on its own. Your hand-written replacement won’t, unless you keep maintaining it.

The real choice: leave it, or own it

Because there’s no way to extend the default, customizing isn’t “adding your spin.” It’s a binary. Either you keep Shopify’s default by doing nothing, or you take full ownership of the file and everything Shopify would have put in it.

For most stores, doing nothing is the right call. The default is valid, it’s maintained for you, and it improves as Shopify builds. You don’t need to touch it to be correct.

Customizing only makes sense if you can answer yes to both: do you know something concrete the default can’t express, and are you willing to maintain the whole file going forward. Concrete means functional, not brand-speak:

How specific products should be recommended, in terms an agent can act on
Special handling an agent would otherwise get wrong: age verification, subscription rules, B2B pricing tiers
Which products are perennial versus seasonal, or which variants actually matter

If you can’t answer yes to both, leave the default in place. A brand paragraph is not worth giving up Shopify’s maintained file.

The part most coverage is missing: you’re a builder, not just a merchant

If you build Shopify apps, this isn’t just a merchant story.

Your own app’s site needs an agents.md.

AI agents that field questions like “what’s the best subscription app for Shopify” are crawling discovery files, not just app store listings. If your app’s marketing site doesn’t have an agents.md, you’re not in the conversation.

As AI-assisted commerce matures, the apps that get surfaced will be the ones that made themselves readable. Apps that didn’t will get there eventually, asking the right questions a year too late.

Your agents.md should tell an AI what your app does, who it’s for, what problem it solves, and what makes it worth recommending over alternatives. Not in marketing language. In terms a recommendation system can act on.

Your app can also write into merchant stores.

This is the bigger opportunity most coverage has skipped.

Think about what your app knows. An inventory management app knows which SKUs are high-velocity and which are aging. A subscription app knows the difference between subscription customers and one-time buyers. A generic AI agent crawling a product catalog has no way to infer any of this.

All of that context can live in a merchant’s /agents.md. And your app is the only thing in a position to put it there, because your app is the only thing that has that data.

The stores with agents.md files rich in operational context will get more accurate recommendations than the ones that pasted in brand copy. Apps that surface this for their merchants have a concrete answer to the question every founder is quietly sitting with: how do we stay relevant when AI handles the front end of commerce.

That’s a real question. And this is one real answer.

What to do this week

Step 1: Read your own files first. yourstore.myshopify.com/agents.md (and /llms.txt). See what Shopify generated before you touch anything.

Step 2: If you’re a merchant, default to leaving it alone. Customizing means owning the whole file forever and giving up Shopify’s future improvements. Only override if you know something concrete the default can’t express, like subscription rules or B2B pricing, and you’ll commit to maintaining it. Never override just to add a brand paragraph.

Step 3: If you’re an app founder, audit your own site. yourdomain.com/agents.md. Does it exist? Does it give an AI enough to recommend you? If not, build it. This is an afternoon of work and it compounds over time.

Step 4: Think about what your app knows that merchants’ files don’t. Talk to a few of your active merchants. Ask how they’d describe their store to an AI assistant. The gap between their answer and what a generic file contains is your product opportunity.

Key takeaways

Customizing your /agents.md fully replaces the Shopify default for that path. There is no merge, no append, no way to extend it. It’s all or nothing.
For most merchants, leaving the default in place is the right call. The default is maintained by Shopify and improves over time. A hand-written replacement is yours to maintain forever.
Customize only if you have something functional the default can’t express, like subscription rules, B2B pricing tiers, or age verification requirements, and you’re willing to own the file from that point forward.
If you build Shopify apps, your app’s own marketing site needs an agents.md. AI agents fielding “what’s the best app for X” queries are reading discovery files, not just App Store listings.
The deeper product opportunity for app builders: your app has operational context a generic crawl can’t surface. That context belongs in merchant files, and your app is the only thing positioned to put it there.

FAQ

Should I customize my Shopify /agents.md? For most merchants, no. The default is maintained by Shopify and improves as their agentic commerce infrastructure develops. Customizing means replacing the default entirely and maintaining the full file yourself. Unless you have specific functional information the default can’t express, leave it alone.

What’s the difference between /agents.md, /llms.txt, and /llms-full.txt? On eligible Shopify stores, /llms.txt and /llms-full.txt both point to the same content by default. There is no “full version” with additional data. The /agents.md file is a separate Shopify-specific convention. All three were auto-generated when Shopify rolled out AI file support. Customization of all three became possible on May 28, 2026.

What happens if I accidentally replaced my default file? Delete your custom template. The fallback hierarchy means Shopify’s default will take over again for that path once your custom template is gone.

When does customizing actually make sense? When you have functional context the default can’t infer: subscription and recurring order rules, B2B pricing tiers, age verification requirements, or product attributes that matter for recommendation accuracy. Brand copy and mission statements are not a good reason. If you’re not sure, check what the default already contains before deciding.

If I build Shopify apps, what should my agents.md say? What your app does, who it’s for, what problem it solves, and what makes it worth recommending over alternatives. In terms a recommendation system can act on, not marketing language. Think of it as a structured brief for an AI that’s never heard of your app.

Right now, Shopify has built the understanding layer for you. The only way to lose that is to replace it without knowing what you’re replacing.

So go read it first.