The way shoppers find products is splitting into two. Alongside people browsing your store, AI assistants now research, compare, and increasingly buy on a shopper's behalf. They look for a short set of plain text files at fixed addresses, and those files, not your homepage, are what an agent uses to understand what you sell.
A misconception is that most people treat llms.txt as a new SEO file to add and forget. But it is not. The honest evidence says it does little for rankings. The reason to care is ownership, and it splits sharply by platform. If you sell on Shopify, the platform already creates this file for you.
If you sell on any other platform, the file does not exist until you create it, so an agent describes you from whatever it can scrape, or from nothing at all.
This guide covers what these files are, whether they work, what Shopify did, and how to take control across the six major eCommerce platforms.
Key takeaways (TL;DR)
|
• On Shopify, your store already publishes an llms.txt and an agents.md you never wrote. The default is a generic Shopify boilerplate that routes agents to the Shop skill and Shop Pay. On every other platform, the file does not exist until you create it. • We checked eight live stores in June 2026. Every Shopify store serves files in standard and headless websites. The five non-Shopify platforms returned a 404. • llms.txt is not a ranking lever. The evidence shows little quantifiable organic lift. The reason to act is control over how an AI agent describes your store. • The files have two layers. A description layer any platform can publish today, and a transaction layer, the /.well-known/ucp profile, and an MCP endpoint, which is automatic only on Shopify. • This guide gives you the structure, a fill-in-the-blank AI prompt for each platform, and step-by-step hosting for six eCommerce platforms. |
What is an llms.txt file?
A website is built for humans to interact with. It has menus, images, scripts, and a lot of code that AI has to wade through to work out what you sell. llms.txt is a short, plain-text note you leave at the front door of your site. It says, in simple terms, here is what we do, here is who we serve, and here are the pages worth reading first.
It lives at one fixed address, yourdomain.com/llms.txt, and it is written in Markdown, the same plain format used for README files. Jeremy Howard of Answer.AI proposed the idea in September 2024. The pitch was simple. Give language models a clean map instead of making them guess from your HTML.
Think of it like the index at the front of a reference book. The book still has every chapter. The index just tells a reader in a hurry where to look. llms.txt does that for an AI.
How an llms.txt file is different from robots.txt and your sitemap
People confuse llms.txt with robots.txt. They are not the same thing. A robots.txt tells crawlers which pages they may or may not enter, crawl, and index. Your sitemap is a phone book. It lists every page so search engines can find them all.
An llms.txt is neither a directive nor a full list. It is a curated recommendation. It highlights your best pages and explains, in one sentence, what each one is for. A robots.txt file restricts, a phone book enumerates, and an llms.txt file recommends. Three different jobs, three different files.
What agents.md adds
The agents.md file is the more powerful cousin of the llms.txt. Whereas llms.txt tells AI what your store is about, agents.md tells AI how to act inside it. It describes how to search your catalog, how to build a cart, how to start a checkout, and the rule that a human has to approve the payment.
If llms.txt is the recommendation, agents.md is the operating manual for an AI tool that wants to buy something. On Shopify, the two are now wired together, which is the part we get to shortly.
Does llms.txt actually work? The honest consensus
According to Adobe Analytics, AI traffic to US retail sites rose 393 percent year over year in the first quarter of 2026, and the same data shows that traffic now converts 42 percent better than non-AI traffic, a full reversal from a year earlier. 
Short version. For getting cited in AI answers, the evidence says no, or at least not yet. A SERanking study of roughly 300,000 domains in late 2025 found no measurable improvement in AI citations from publishing llms.txt. The major AI crawlers from OpenAI, ChatGPT, Google, and Anthropic do not fetch the file in any real volume.
But wait, it gets even more ambiguous. In May 2026, Google published guidance on optimizing for AI features that is similar to plain traditional search advice: good content, structured data, crawlable pages, and it never mentions llms.txt.
Additionally, BuiltWith counts more than 7.3 million live sites serving an llms.txt file, almost exactly matching its count of live Shopify stores, which suggests most of that adoption is Shopify turning the file on by default, not merchants choosing to. The issue is that no major AI platform has confirmed it reads them. So if an SEO agency sells you llms.txt as a ranking lever, push back.
An llms.txt does not improve rankings, so most people write it off. That question is framed wrong. What's important is whether an AI agent can read what you sell and buy it for a customer. That is a different job, with different files, and on most platforms, it does not work until you set it up.
What Shopify is doing right now
In May 2026, over a few days, with no email and no changelog, Shopify started serving llms.txt, agents.md, and a set of agent discovery files on every store. Your store almost certainly has them now.
To summarize, the sequence this year has been as follows. Google announced the Universal Commerce Protocol at NRF on January 11, 2026, built with Shopify. Shopify turned on its Agentic Storefronts feature by default for eligible US merchants on March 24. The llms.txt files then appeared in early May.
On a default Shopify store, llms.txt and agents.md serve the same document at two URLs. You can override either one. However, the file that needs to be right is agents.md, since that is what an AI agent reads to transact. If you want llms.txt to serve as a plain description rather than a copy of the agent instructions, you can override it separately, but do not leave agents.md as the default if you care what an AI agent tells a buyer about you.
What the default file actually says
For example, I reviewed the much-cited Allbirds Shopify store, where the route at /llms.txt returns a full set of agent instructions. Two things stand out.
First, it ends by inviting the reader to start their own Shopify store, a link sitting on your domain and paid for by your traffic.
Second, and bigger, it calls the Shop skill the recommended way to “transact across Shopify stores,” routing purchases through Shop Pay.
The wording was identical on Skullcandy, a second Shopify store I checked the same day.
Read that again. The description your store hands to an AI is nudging it toward a cross-store shopping layer that Shopify controls, not toward you. It is good business for Shopify. It is not automatically good business for you. This is the ownership question, and it is the real reason to act, in my opinion.
What it means for you
You can override these files. Drop a file at templates/llms.txt.liquid in your theme, just as robots.txt.liquid has worked for years, and your version replaces the default llms.txt file autogenerated by Shopify. The agents.md works the same way through templates/agents.md.liquid.
As of this writing, Shopify has not officially documented this override, and the behavior still reads as in testing. It works on live stores today, but it can change. The other trap is bigger. Your custom file replaces the default completely, it does not merge.
If you strip out the agentic commerce discovery links to /.well-known/ucp and the MCP endpoint, you cut your store off from the checkout orchestration Shopify has already wired in for you. Keep those links. Change the description only in the llms.txt file.
What live stores are actually serving right now?
Furthermore, I tested some popular stores on other eCommerce platforms to see how they are handling llms.txt files. The result is a clean split. On Shopify's storefronts, the file is there and rich. On the sites I tested, there was no llms.txt file.
The case I expected to be a miss was Gymshark, which runs headless on Shopify, and it turned out not to be one. Both gymshark.com/llms.txt and gymshark.com/agents.md return the Shopify default, because the apex redirects those paths to Shopify's own checkout domain.
The www host, served from their CDN, returns a 404, which is why you check every host and not just one. So with headless, the file does not always vanish, and you serve it yourself only when your front end responds to that path without proxying to Shopify.
Helly Hansen on Magento, Columbia on Salesforce, Berlin Packaging on BigCommerce, Daelmans on WooCommerce, and Bang and Olufsen on commercetools all returned nothing. That blank is the opening.
On Shopify, the platform already writes your agent file and points it to Shop. On every other platform, the page is empty, which means whoever writes it first decides how an agent describes the store. Right now, there is almost no one from the brands I looked up.
The agentic commerce stack, in plain terms
Let's step back from Shopify for a minute, because this is bigger than one platform. llms.txt and agents.md are the front door. Behind that door is a number of protocols that decide whether an AI agent can find your products and buy them. Most teams have never heard the names. They are about to become real for every merchant.
The layer that handles discovery and checkout is where the fight is. Google's Universal Commerce Protocol covers the full journey from finding a product to managing the order, and a March 2026 update added cart and live catalog access. OpenAI's Agentic Commerce Protocol, built with Stripe, handles checkout inside ChatGPT.
Underneath sits AP2, Google's payment authorization protocol backed by Visa, Mastercard, and American Express, and MCP, the open standard from Anthropic, now run by the Linux Foundation, which is the pipe through which the other layers travel.
The practical point. ChatGPT transacts through ACP, while Google's AI Mode and Gemini use UCP, so most retailers will need to support both. You can't pick a side. You have to make sure your store can speak to both.
|
Protocol |
Who owns it |
What it does |
Where it shows up |
|
UCP |
Google, built with Shopify |
Full journey: discover, cart, checkout, order management |
Google AI Mode, Gemini |
|
ACP |
OpenAI with Stripe |
Agentic checkout against an indexed catalog |
ChatGPT |
|
AP2 |
Google, with Visa, Mastercard, Amex |
Proves a human approved the spend |
The payment layer under both |
|
MCP |
Anthropic, Linux Foundation |
Connects agents to your catalog and tools |
The transport under the stack |
How to structure the agentic commerce files, modeled on what Shopify built
I pulled apart what a live Shopify store actually serves, so you can copy the parts that make sense and skip the parts you cannot run. A standard storefront publishes four things that fall into two layers.
First is the description layer, which is comprised of /llms.txt and /agents.md at the root.
Second is the transaction layer, a /.well-known/ucp profile that declares the store supports the Universal Commerce Protocol, plus an MCP endpoint that an AI agent calls to search, build a cart, and check out. Shopify adds a fourth file, a small dedicated sitemap that points agents at the agents.md.
Any website, not just those on Shopify, can build a similar description layer today. The transaction layer is the part that is automatic only on Shopify. Let's now look at the two layers in more detail.
The description layer, which every website/platform can publish now
An llms.txt is a Markdown index. A title, a one-line summary, then short linked lists of your best collections/categories, your flagship products, and your policy pages.
Whereas agents.md is the instruction file. It states what you sell, points to the same key pages, lists the rules an agent should follow, confirms with the human before buying, uses live pricing, does not invent claims, and summarizes your delivery, returns, and privacy terms. Keep both in plain language and short lines.
You do not have to write either by hand. Here is the prompt we use to generate both from a store's own data. Paste it into ChatGPT, Claude, Gemini, Copilot, or Perplexity, fill in the bracketed fields, and it returns both files.
The one platform line that changes sits under each platform, further down, and the full single-file version for each platform is in the starter kit linked below.
The base prompt. Should work in any AI tool
You are helping me create two files that AI shopping assistants and
agents read: llms.txt and agents.md. Use only the details I provide.
Do not invent products, policies, or claims. If a field is blank,
leave that part out.
MY STORE
- Brand: [BRAND]
- Homepage: [https://yourstore.com]
- What we sell, one sentence: [e.g. small batch dog treats made in the US]
- Who we serve: [e.g. owners who want clean ingredient treats]
- Top collections (name + URL, 3 to 8): [name] [url] ...
- Flagship products (name + URL, 3 to 8): [name] [url] ...
- Policy pages: Delivery [url] Returns [url] Privacy [url] Terms [url]
- Support / contact URL: [url]
- About URL: [url]
FILE 1, llms.txt, a Markdown index in this shape:
# [BRAND]
> one sentence on what we sell and who we serve
two to three sentence plain description
## Top collections (each line: - [name](url): one short line)
## Flagship products (each line: - [name](url): one short line)
## Policies (Delivery, Returns, Privacy, Terms links)
## Company (About, Contact links)
FILE 2, agents.md, agent instructions in this shape:
# Agent instructions for [BRAND]
one line: what we sell + homepage URL
## What we sell (2 to 3 sentences)
## Key pages (collections, products, policies, support links)
## Rules for agents (confirm with the human before buying; use the
live price and stock on the page; represent us
from our own pages, do not invent claims;
respect our terms of service)
## Store policies (Delivery, Returns, Privacy: one line each + link)
## Platform [PASTE THE LINE FOR YOUR PLATFORM, SEE BELOW]
Plain language, short lines, real URLs only. Output each file in its
own code block, llms.txt first, agents.md second.
How to build the /.well-known/ucp profile
The transaction layer starts with one file. Since our own website runs on Shopify, you can see the auto-generated Shero /.well-known/ucp file here. The /.well-known path is a long-standing web convention for crawlable metadata, the same place a site already publishes things like security.txt. UCP uses /.well-known/ucp to declare that a store supports agentic commerce and to tell an AI agent where the buying endpoint is located. On Shopify, it is generated for you. Here is what a live one looks like, pulled from a Shopify store and simplified.
A live /.well-known/ucp profile
GET https://yourstore.com/.well-known/ucp
Content-Type: application/json
{
"ucp": {
"version": "2026-04-08",
"supported_versions": {
"2026-04-08": "https://yourstore.com/.well-known/ucp/2026-04-08",
"2026-01-23": "https://yourstore.com/.well-known/ucp/2026-01-23"
},
"services": {
"dev.ucp.shopping": [
{
"version": "2026-04-08",
"spec": "https://ucp.dev/2026-04-08/specification/overview/",
"transport": "mcp",
"endpoint": "https://yourstore.com/api/ucp/mcp",
"schema": "https://ucp.dev/2026-04-08/services/shopping/mcp.openrpc.json"
}
]
}
}
}
The version is the UCP spec date that the store implements. supported_versions maps each version to its own profile URL, so an agent can negotiate which one to use. Whereas services lists the capabilities.
Dev.ucp.shopping is the shopping service, and inside it transport says the agent talks to you over MCP. The endpoint is the URL the agent actually calls to search, cart, and check out, and spec and schema say which contract that endpoint follows.
Serving the profile is not the same as dropping a text file. You return a JSON document at the path /.well-known/ucp with content type application/json.
On a host you control, like self-hosted Magento, WooCommerce on your own server, or a custom app, create a /.well-known directory in your web root and serve the JSON there, or map a route to it. On SaaS without root access, BigCommerce, Salesforce, or headless Shopify, you need an app, a controller, or an edge function to answer that path.
Remember, the profile is a signpost. The endpoint it points to must be a working MCP server that can run search, cart, and checkout against your live catalog. That is a real service, built against the UCP and MCP specs or bought from a vendor, not a file you upload.
A signpost to a shop that does not exist is worse than no signpost, because an agent will try the endpoint, get nothing, and may drop you from its results.
So, be honest about where you are. For almost every store not on Shopify in 2026, the right move is to publish the description layer now and leave the UCP profile alone until you actually run an endpoint, or until your platform runs one for you, as Shopify does.
The agentic discovery sitemap, copied from Shopify
Shopify does one more thing worth copying. It publishes /sitemap_agentic_discovery.xml, a standard sitemap whose only job is to list the agents.md with a weekly change frequency. That is how the file gets found without cluttering your main sitemap.
Worth noting, Shopify does not reference these files in robots.txt at all. It relies on the well-known path and this dedicated sitemap. Here is the same file, ready to edit.
sitemap_agentic_discovery.xml
<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
<url>
<loc>https://yourstore.com/agents.md</loc>
<changefreq>weekly</changefreq>
</url>
<url>
<loc>https://yourstore.com/llms.txt</loc>
<changefreq>weekly</changefreq>
</url>
</urlset>
Swap in your domain, host it at the root, then submit it in Google Search Console under Sitemaps, next to your main sitemap.
On a platform where you cannot place a file at the root, serve this the same way you serve the other files: via a redirect or route. It is a five-minute job that hands agents a clean, declared path to your files instead of making them guess.
How to put your own agentic files in place by platform
The description layer is the same job everywhere. Get clean llms.txt and agents.md to load at the root. What changes is how you host them, and whether you can add the transaction layer at all.
Here is the detail for each of the six major platforms, with the one platform line to drop into the base prompt above. The audit earlier in this guide shows which of these serve a file today and which return a 404.
Shopify
Shopify already writes both files for you on a standard storefront, so your job is to replace the default llms.txt, not build from scratch. I am suggesting replacing the default one because it points shoppers to the Shop App and a signup link (as the screenshot earlier in this guide showed).
You replace it with a version that describes your store on your terms while keeping the agentic configs intact.
The theme template override
Your custom file fully replaces the default. It does not merge. So anything you leave out is gone, including the UCP and agent discovery links Shopify injected. Copy those forward.
One, in Online Store --> Themes --> duplicate your live theme so you work on a copy.
Two, edit the copy and add two files, templates/llms.txt.liquid and templates/agents.md.liquid.
Three, paste your generated content, and keep the UCP discovery block from the current default agents.md.
Four, preview the copy, load /agents.md and /llms.txt on the preview to confirm, then publish.
The transaction layer keeps running on its own. The /.well-known/ucp profile and the MCP endpoint are served by the platform, not by your theme, so you do not rebuild them. You only stop deleting their discovery links when you override the file.
The Shopify Markets trap
If you run more than one market on a single domain, the default does not localize currency or locale. Use Liquid localization objects so each market gets the right context, or an agent will quote US pricing to a UK shopper.
Add to the base prompt, Shopify
## Platform Shopify. Agentic checkout over UCP is available automatically through our platform endpoints. Keep the UCP discovery links Shopify already publishes. Do not invent new endpoints.
WooCommerce
In Woo, you own the web root, which makes this the easiest platform to host your agentic commerce files on. There are two paths that trade convenience for control.
The plugin path. AIOSEO, Yoast, and Rank Math each generate llms.txt and refresh it from your sitemap on a schedule. Turn it on, then edit the output, because the plugin cannot know which collections actually earn the revenue.
The manual path. Upload both files via SFTP to the WordPress root, the folder that contains wp-config.php, with permissions 644. A real file in the root is served before WordPress routing ever sees the request.
The common failure is a 404 after upload. That is almost always a security or caching plugin, or a CDN, intercepting the path. Add an exception for the two files.
You have no UCP endpoint unless you build one, which is out of scope for most Woo stores, so publish the description layer and stop there.
Add to the base prompt, WooCommerce
## Platform WooCommerce. We do not run an agentic checkout endpoint. Do not include any UCP, MCP, /.well-known/ucp, or /api endpoint in the file. Describe the store, its key pages, and its policies only.
Magento and Adobe Commerce
Magento HTML is among the hardest outputs for a model to parse, which makes a clean, hand-built file worth more here than on most platforms. The web root is the pub directory. When self-hosted, place llms.txt and agents.md in pub with permissions 644 and confirm they load.
Adobe Commerce Cloud is trickier. The runtime filesystem is read-only and resets on every deploy, so a file you copy over SSH disappears at the next release.
Commit the files into your git repo so they deploy with the build, as shown below. Add a writable mount only if you need to regenerate the file at runtime.
Adobe Commerce Cloud, deploy safe hosting
# Adobe Commerce Cloud: the runtime filesystem is read-only and resets
# on every deploy. Commit the files into your git repo so they deploy with
# the build and are served as static files from the web root:
# pub/llms.txt
# pub/agents.md
#
# Only if you need to regenerate the file at runtime, add a writable
# mount in .magento.app.yaml:
mounts:
"pub/agentic":
source: local
source_path: "agentic"
When you list pages, exclude faceted navigation URLs. Magento notoriously generates thousands of filtered links with no unique content, which makes your catalog look bloated and confuses an agent about what you actually sell.
You have no UCP endpoint unless you build one.
Add to the base prompt, Adobe Commerce or Magento
## Platform Adobe Commerce or Magento. We do not run an agentic checkout endpoint. Do not include any UCP, MCP, /.well-known/ucp, or /api endpoint in the file. Describe the store, its key pages, and its policies only.
BigCommerce
BigCommerce does not let you write to the true root, so you host the files on a path you can write to and redirect the root path to it.
First, upload llms.txt and agents.md via WebDAV or the storefront file manager to a web-accessible folder.
Second, in Server Settings, 301 Redirects, add a redirect from /llms.txt to the hosted URL and from /agents.md to its URL.
Third, load yourstore.com/llms.txt and confirm it resolves to plain text.
The redirect is the whole trick. AI Agents request the root path first, and the redirect is what makes a file living in another folder answer at /llms.txt. If you would rather not touch WebDAV, apps such as Searchanise automate the upload and the redirect.
Same as Woo and Magento, you have no UCP endpoint unless you build one.
Add to the base prompt, BigCommerce
## Platform BigCommerce. We do not run an agentic checkout endpoint. Do not include any UCP, MCP, /.well-known/ucp, or /api endpoint in the file. Describe the store, its key pages, and its policies only.
Salesforce Commerce Cloud
SFCC has no public root filesystem. Every URL routes through a cartridge controller, so this is a developer job, not a no-code one. The cleanest method is a small custom cartridge with a controller that returns the file as plain text.
A minimal SFRA controller
// cartridge/controllers/Llms.js (in a custom cartridge)
var server = require('server');
server.get('Txt', function (req, res, next) {
res.setContentType('text/plain; charset=utf-8');
res.print('# Your Brand\n> what you sell\n... your llms.txt content ...');
next();
});
module.exports = server.exports();
// Controller URLs are not at the root by default. Map /llms.txt to this
// controller with a 301 redirect or an SEO URL rule so agents reach it
// at the path they expect.
The alternative is to host the files as static content in the library and add a 301 redirect from /llms.txt and /agents.md. Either way, run the content past whoever owns disclosure before launch, because the file is public to anyone who asks, including competitors. You have no UCP endpoint unless you build one.
Add to the base prompt, Salesforce Commerce Cloud
## Platform Salesforce Commerce Cloud. We do not run an agentic checkout endpoint. Do not include any UCP, MCP, /.well-known/ucp, or /api endpoint in the file. Describe the store, its key pages, and its policies only.
commercetools
Unlike the other platforms here, commercetools is headless. It has no storefront of its own, so the files live in your front-end framework. The static path is fastest, drop llms.txt and agents.md in the public folder, and they go live in minutes. For a catalog that changes often, generate the file from your data so it never goes stale.
A Next.js route that generates llms.txt from your catalog
// app/llms.txt/route.ts (Next.js app router, generated daily)
import { getTopCategories } from "@/lib/catalog"; // your commercetools call
export async function GET() {
const cats = await getTopCategories();
const body =
"# Your Brand\n> what you sell and who you serve\n\n## Top collections\n" +
cats.map(c => "- [" + c.name + "](" + c.url + "): " + c.blurb).join("\n");
return new Response(body, {
headers: {
"Content-Type": "text/plain; charset=utf-8",
"Cache-Control": "public, max-age=86400" // refresh once a day
}
});
}
Because you own the front end and the API, commercetools is the platform where building a real UCP endpoint is most feasible. It is still a project, not a toggle, but the possibility is open if agentic checkout becomes worth it for you.
Add to the base prompt, commercetools
## Platform commercetools, headless. We do not run an agentic checkout endpoint unless we build one in our storefront. Do not include a UCP or MCP endpoint unless I tell you we have one live.
|
Platform |
Native file today |
Where it hosts |
Transaction layer |
|
Shopify |
Yes, auto-generated |
Theme Liquid override |
Automatic |
|
WooCommerce |
Via plugin |
Plugin or SFTP to root |
Build it yourself |
|
Magento and Adobe |
No |
pub directory or repo commit |
Build it yourself |
|
BigCommerce |
No |
WebDAV plus a 301 redirect |
Build it yourself |
|
Salesforce |
No |
Cartridge controller or redirect |
Build it yourself |
|
commercetools |
No |
Frontend static file or route |
Build it yourself |
What to put in the files, and what to leave out
Getting the file to load is the easy half. Writing content that actually helps an AI represent you is the trick.
Put these in
Lead with one plain sentence naming what you sell and who buys it, written for a model that has never heard of you. Then your top five to ten collections by revenue, not by menu order, each with a sentence of context.
Add three to ten flagship products with enough detail to answer what is this and who is it for. Then your returns, delivery, and privacy pages, and your contact details.
Leave these out
No faceted or filtered navigation URLs. No cart, checkout, or account pages. No expired sale pages. And no marketing language. Industry-leading tells an AI tool nothing. Waterproof trail running shoe, 200 grams, built for technical terrain tells it all it needs to recommend you.
Tools, directories, and templates
You do not have to write these from scratch. Here is what is worth using as of mid 2026, plus one starter kit I've built so you can avoid serving a blank page.
See real files in the wild
Three directories let you read live files and copy good structure: directory.llmstxt.cloud, llmstxt.site, and llms-text.com. One caveat. Adoption skews toward developer tools and SaaS, so read them for format, not for eCommerce examples.
Generators and validators
As mentioned above, on WordPress and WooCommerce, AIOSEO, Yoast, and Rank Math generate the file and keep it in sync with your sitemap. For other platforms, standalone tools like Firecrawl and LLMTEXT by Parallel.ai produce a draft, and Parallel also validates the output and converts it for MCP.
Documentation platforms such as Mintlify, Fern, and GitBook generate it automatically. One warning applies to all of them. A generator dumps URLs without judgment, so a high-value product page lands next to a stale tag archive. Always edit the output.
Our starter kit and AI prompts
We built a starter kit so you do not start from a blank page. It includes an annotated example llms.txt and agents.md, the Shopify Liquid override with the discovery links left in place so you do not break agentic checkout, the agentic discovery sitemap XML, and one fill-in-the-blank AI prompt per platform.
Paste the prompt for your platform into ChatGPT, Claude, Gemini, Copilot, or Perplexity, add your store details, and it writes both files in the structure above. Download the agentic files starter kit, edit the placeholders, and host it.
What to do next
Five steps, in order.
- One, confirm your AI crawlers are not blocked in robots.txt.
- Two, read your own llms.txt and agents.md and see what your platform is saying about you.
- Three, clean your product data, because clean titles and accurate inventory are what agents actually use to recommend you.
- Four, write a custom file if the default Shopify is misrepresenting you.
- Five, publish the agentic discovery sitemap and submit it to Search Console so agents have a declared path to the file.
Notice the order. The file is the last step, not the first. A perfect llms.txt sitting on top of thin product data helps no one. Fix the data, then point the agents at it.
What does the future look like?
Here is my prediction, and it is dateable and debatable. By the end of 2026, serving llms.txt and agents.md will be standard on every major platform, the way robots.txt and sitemaps are now. Winning will not be the biggest ad budgets, but the websites whose data is clean and whose brand can be described correctly by an AI that has never spoken to a human at the company.
The question is whether the description AI reads about your business is one you wrote, or one a platform wrote for its own reasons. That choice is still yours. For now.
