llms.txt and agent-readiness for e-shops

A question we hear a lot lately: "Someone told me I need an llms.txt. Do we have one?". The short answer is that it is worth having, it costs very little, and it is not a magic switch. The bigger picture, so-called agent-readiness, matters far more than a single text file.

What llms.txt is

llms.txt is a proposal published in 2024: a Markdown file at the root of your site (/llms.txt) that gives language models a short, clean summary of what the site is and links to its most important pages. The idea is that a model with a limited context window should not have to dig through menus, banners and scripts to work out what you sell.

A typical structure:

• A # heading with the shop name.
• A short summary in a blockquote: what you sell, to whom, where you ship.
• ## sections with link lists: main categories, policies (shipping, returns, payments), FAQs, contact.
• Optionally an llms-full.txt with more content in a single file.

What it does not do

Let us be honest, because there is plenty of hype around:

• It is not an official standard. It is a proposal that sites adopt voluntarily.
• Google has said it does not use it for search or AI Overviews. There, the usual things matter: crawlability, content, structured data.
• It does not replace robots.txt. It controls nothing; it just offers a map.
• It will not lift you in any ranking on its own.

So why bother? Because some tools and agents do read it when they find it, because it takes half an hour, and because writing it forces you to answer "what are my 15 most important pages?". That alone is a useful exercise.

An example for a Greek shop

An llms.txt for a cosmetics shop might include:

• Summary: "Greek online shop for natural cosmetics. Ships across Greece with ACS and BOX NOW, free over X euros. Pay by card, IRIS, Apple Pay or cash on delivery."
• Categories: Face, Body, Hair, Sun care, one link each.
• Guides: "How to choose a sunscreen", "A routine for sensitive skin".
• Policies: Shipping, Returns, Payments, Terms.
• Contact: the contact page and support hours.

Keep it short and current. An llms.txt with the wrong shipping cost is worse than none at all.

How to set it up and keep it current

On WooCommerce you can upload a static file to the site root or use a plugin that generates it. On Shopify the root is not directly writable, so you usually need a redirect or an app. On a custom platform, a route that builds the file automatically from your categories and policies is the cleanest option.

Whichever route you take, set a reminder: whenever shipping costs, a policy or a main category changes, the llms.txt gets updated too. Also check that it returns a 200 status and that the Markdown reads cleanly, with no HTML mixed in.

The bigger picture: agents that buy

The interesting part is not models that read, but agents that act. There are already tools that open a browser, search for products, compare prices and fill in forms. Platforms such as OpenAI and Google are experimenting with purchases inside the conversation, driven by merchant feeds and protocols. In Greece this is still early days, but the direction is clear.

An agent-ready shop has the following traits.

1. Clean, semantic HTML

Agents mostly navigate through the accessibility tree, the same structure screen readers use. That means:

• Buttons that are real button elements with text, not divs with an icon.
• Form fields with proper labels.
• Variant pickers (size, colour) that expose their selected state.
• No pop-ups that cover content and are hard to dismiss.

Everything you do for accessibility, you are doing for agents too. Two wins for one piece of work.

2. Structured data everywhere

Product, Offer, Organization, BreadcrumbList and, where appropriate, FAQPage. Price and availability must match what the user sees and what your feeds say.

3. Feeds as the source of truth

A correct product feed (Google Merchant Center, Skroutz, BestPrice) is already a machine-readable version of your catalogue. If your feeds have errors, every agent that relies on them inherits those errors.

4. A checkout that does not break

An agent gets stuck in the same places a human on a bad connection does: forced account creation, CAPTCHA on every step, unlabelled fields, unexplained errors. A simple, predictable checkout helps everyone.

5. Sensible bot management

You do not want every bot on your site. You do want to tell apart scrapers that hammer your server from crawlers and agents that bring customers. Check what your firewall blocks, what robots.txt allows, and whether your CDN has specific AI bot settings switched on.

A practical list for this month

• Create /llms.txt with a summary and 10–20 key links.
• Check robots.txt for OAI-SearchBot, PerplexityBot, Googlebot and Bingbot.
• Check whether your firewall or CDN is silently blocking AI crawlers.
• Run your product pages through a structured data test.
• Do a quick accessibility test of checkout (keyboard-only navigation).
• Make sure price and availability agree across the site, schema and feeds.

If you want the context on why this matters for AI recommendations, read our guide on getting recommended by ChatGPT and Gemini.

The bottom line

llms.txt is a small, cheap step. Real agent-readiness lives in the fundamentals: clean HTML, correct data, reliable feeds and a checkout that works. If you would like your shop reviewed from this angle, see our AI visibility service or get in touch.