Why your Shopify store is invisible to ChatGPT — and how to fix it

Most Shopify stores are invisible to ChatGPT because AI crawlers are blocked in robots.txt, the store has no llms.txt or agents.md guide, product descriptions are too thin to answer a buying question, and product pages lack structured data — four fixable problems, none of which are visible in Shopify's own admin.

A shopper opens ChatGPT and types: "best unscented shampoo for a sensitive scalp under $30." They get four specific products, three store names, and links. They click one. They buy. That entire journey happened without a search engine results page, without an ad auction, and — for the overwhelming majority of Shopify stores — without any possibility that your products were among the four.

This is not because your products are worse. It is because AI assistants cannot read your store. That is a technical condition with technical causes, and unlike ranking on Google, most of it is fixable in an afternoon by one person with admin access. This article covers exactly what those assistants read, the four specific things that block a default Shopify store, how to diagnose each one yourself, and how to tell whether the fix actually worked.

What AI assistants actually read

There is a widespread assumption that AI assistants answer shopping questions from their training data. For product recommendations, that is mostly wrong, and the distinction matters enormously for what you should do about it.

When a shopper asks for a product recommendation today, the assistant performs a live retrieval step: it issues searches, fetches pages, and reads them at the moment of the question. ChatGPT does this through its browsing and search tooling, Perplexity is built entirely around it, Gemini grounds answers with Google Search, and Claude has web search. The answer you get is assembled from pages fetched seconds earlier — which is why assistants can recommend a product launched last week.

That live fetch is performed by named crawlers, and this is where most stores fail before the race starts. Each assistant operates its own user agent, and each can be allowed or blocked independently in your robots.txt:

Note the last row carefully. Google-Extended is a separate control from Googlebot. It is entirely possible — and common — for a store to rank respectably on Google Search while being invisible to Gemini, because someone once added a directive intended to keep content out of AI training and inadvertently opted out of AI recommendations too.

The four things blocking your store

In practice, failures cluster into four causes. They compound: fixing crawler access does nothing if your product pages have nothing worth reading, and world-class product copy is worthless if the crawler is turned away at the door.

1. Blocked crawlers

This is binary and it is the first thing to check. If your robots.txt disallows GPTBot, no amount of content work matters for ChatGPT. Shopify's default robots.txt does not block AI crawlers — but many stores are not running the default. Common culprits: an SEO app that added directives, a theme customisation from a developer who read an article about AI scraping, a robots.txt.liquid template edited during a migration, or a Cloudflare or hosting-level bot rule that blocks by user agent before the request ever reaches Shopify.

2. No store guide (llms.txt / agents.md)

A crawler that reaches your store still has to work out what you sell, who you are, and which pages matter. It does that by crawling and inferring — unless you tell it directly. The emerging convention for that is a plain-text guide at your domain root: llms.txt, and increasingly agents.md, which state what the site is, what it sells, and which URLs are canonical for which purpose.

This is not a magic ranking file, and anyone selling it that way is overselling. What it does is remove ambiguity. An assistant deciding whether your store is a credible answer to "unscented shampoo for sensitive scalp" benefits from a page that says, in plain language, that you sell fragrance-free hair care for sensitive skin, here are the categories, here is the returns policy. Shopify now serves these natively via theme templates, which means publishing one is a template change rather than a hosting change.

3. Thin product content

This is the big one, and it is the one most merchants underestimate. Crawler access gets you read. Content quality gets you recommended. Those are different problems.

Consider what the assistant has to do. The shopper asked for unscented, sensitive-scalp, under $30. To recommend your product, the assistant needs to establish that it is unscented, that it suits a sensitive scalp, and that it costs under $30. If your product description reads "Amber Oud 50ml. Long lasting." — a real pattern, and a common one — the assistant has nothing to match against. It will not guess. It will recommend the competitor whose page says the thing explicitly.

The second version is not longer for the sake of length. Every clause answers a question a buyer actually asks. That is the test to apply to your own descriptions: for each sentence, what shopper question does this answer? If a description answers none, it will never be retrieved for any.

4. Missing structured data

Structured data — specifically schema.org Product markup — states your price, currency, availability, brand and review data in a machine-readable block rather than leaving it to be parsed out of page text. For a question with a hard constraint like "under $30", the difference between a stated price and an inferred one is the difference between being confidently included and being cautiously omitted.

Most Shopify themes include some Product schema. Many include it incompletely — missing availability, missing currency, or emitting it only on some template variants. It is worth verifying rather than assuming.

How to diagnose your own store in 15 minutes

1. 1

   Check your robots.txt

   Visit yourstore.com/robots.txt. Search for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Any Disallow directive against these is a hard block. Also check for a blanket Disallow under User-agent: * that would catch them.

2. 2

   Check for a store guide

   Visit yourstore.com/llms.txt and yourstore.com/agents.md. A 404 on both means assistants are inferring your catalogue rather than reading a description of it. Shopify serves a generic default for some stores — a generic file is better than nothing but describes Shopify, not you.

3. 3

   Audit ten product descriptions

   Open ten product pages at random. For each, ask: which specific buyer question does this answer? Count how many answer none. In most stores this is more than half, and those products cannot be recommended for anything.

4. 4

   Verify your structured data

   Run a product URL through Google's Rich Results Test. Confirm Product schema is present and that price, priceCurrency and availability are populated. Missing availability is the most common gap.

5. 5

   Ask the assistants directly

   Open ChatGPT, Claude, Gemini and Perplexity. Ask each a genuine buying question in your category — the phrasing a customer would use, not your brand name. Record whether your domain appears in the answer or its citations. This is your baseline.

GEO and SEO are not the same discipline

Generative engine optimisation shares vocabulary with SEO and a fair amount of technical groundwork, but the objective function is different in ways that change what you prioritise.

The last row is the one to internalise. In SEO, a weak page ranks poorly. In GEO, an unreadable store does not compete at all — it is not ranked tenth, it is absent. That is why the fixes above are worth doing before any sophisticated content strategy: they move you from ineligible to eligible.

Proving the fix worked

Here is where most AI-visibility advice stops, and where it should not. Publishing a llms.txt file and declaring victory is not a result. The honest question is whether assistants now recommend you when they did not before, and whether that produced revenue.

Two measurements, in order of usefulness. The first is citation rate: take a fixed set of buying questions in your category, ask each assistant weekly, and record how many return your domain. Because the set is fixed, the trend is meaningful even though the sample is small. The second is attributed revenue: some assistants tag outbound links, so orders can carry a signal identifying where the shopper came from. ChatGPT appends a utm_source parameter to links it produces, and referrers from assistant domains are visible in analytics.

This is the loop worth building: measure, diagnose, fix, publish, re-measure, and attribute. Nomu runs it on a schedule — probing the assistants weekly with buyer questions derived from your catalogue, diagnosing what is holding each engine back, drafting the fixes for your approval, publishing them to Shopify with the before-state captured so every change is reversible for thirty days, then re-checking after publication and reporting what happened to revenue. But the loop matters more than the tool: run it manually if you prefer. What does not work is treating AI visibility as a one-off configuration task.

Frequently asked questions

The verdict

AI visibility is currently a technical problem masquerading as a marketing one. The stores winning recommendations are rarely the ones with the cleverest content strategy — they are the ones that are readable, explicit, and structured, while their competitors are accidentally blocking crawlers and shipping two-line product descriptions.

That will not last. The advantage available right now exists because most merchants have not checked. Spend the fifteen minutes on the diagnosis above, fix what it surfaces, and record your citation rate before and after. If it moves, you have found a channel your competitors do not know they are losing.