Someone in Silver Lake opens ChatGPT and types: “warm vanilla eau de parfum under $80, not too sweet, something I won’t smell on everyone else.”
She is not going to visit your store to answer that. The agent answers it. If your product comes back in that answer, you just made a sale from a conversation you were never part of.
And if it doesn’t come back, you will never know she asked.
Most LA brands have not registered that this is already running. You did not opt in. If you are on Shopify and your products are eligible, you are in there right now.
You are in it whether you looked or not
Shopify switched this on by default.
No app to install. No extra fee. Agentic Storefronts is automatically activated for eligible merchants, and your products sit in Shopify Catalog, which standardises them and distributes them to connected AI platforms. ChatGPT is live today. Microsoft Copilot and Google’s AI Mode are rolling out behind it.
Shopify’s own Q1 2026 figures, for a sense of slope:
- AI-driven traffic to Shopify stores grew 8x year over year
- Orders from AI-powered searches grew nearly 13x
- New buyers are ordering through AI channels at nearly twice the rate of other channels
Small base. Very steep line. That is the part worth your attention.
We ran Shopify’s own audit tool. It said this:
Good news. This store is on Shopify. All Shopify merchants’ eligible products are listed in Shopify Catalog.
That was the whole report.
Shopify publishes a free readiness scanner at shopify.com/agentic-readiness. We pointed it at a fragrance catalogue we built on Shopify Plus, expecting a field-level breakdown. No gaps, no score, no list of what to fix. Just confirmation that we exist.
The detailed scan appears to be built for merchants who are not on Shopify yet, which tracks, because the tool is also a sales funnel. Fine. But it means the official way to check your readiness tells a Shopify merchant nothing they can act on.
So we went and read the data ourselves.
Being included is not the same as being recommended
Catalog takes the product data you give it, standardises the structure, and fills gaps by inference from machine learning and transaction signals. Shopify is explicit that you do not need to change anything to take part.
Taking part was never the question.
The question is whether an agent can match you against a specific request. Beauty and fragrance are decided almost entirely on attributes. Concentration. Size. Notes. Family. Gender. Skin type. Finish. Shade.
Go back to Silver Lake. Warm, vanilla, EDP, under $80, not cloying. That is five constraints, and the agent resolves them against fields. A field that does not exist cannot be matched. You do not get a partial score for being close.
Nothing about your store looks broken while this is happening.
What we actually found
The commerce fundamentals were solid. Name, SKU, description, brand, images, price, currency, availability, return policy, shipping details, aggregate rating. All present on every one of the five products we read by hand. That is better than most catalogues we inherit across 500+ Shopify builds, and it clears Shopify’s stated baseline of a title, an image, a price and shipping to the US or Canada.
The additionalProperty field was empty on all five. This is where concentration, size in millilitres, fragrance family, notes and gender would live. Nothing. The phrase “Eau De Parfum” does appear in the product titles as free text, so inference has something to work with for concentration. Notes and family have nowhere to come from at all.
The gtin field was missing on all five. On four of them a 13-digit EAN barcode was sitting in the mpn field instead. The identifier exists. It is filed in the wrong drawer.
Nothing was blocked. The robots.txt does not shut out GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended or ClaudeBot.
This is not an access problem. It is a description problem. The store is entirely readable and genuinely thin on the handful of attributes its category actually competes on.
We would expect some version of this on most beauty and fragrance catalogues we open. Not carelessness. Until this year there was no reason on earth to care whether a machine could parse your concentration field. Product pages were built for people, and they were right to be.
Six mistakes worth twenty minutes
Not a checklist of things to add. These are the specific ways we see it go wrong on the beauty catalogues that come through our door.
Treating the product title as branding. “Patchouli Wood” is a name. “Ajmal Patchouli Wood Eau De Parfum” is a name plus two matchable attributes. Shopify’s own guidance is to put materials, sizes and use cases in the title rather than trusting the description to carry them.
Filing the barcode in mpn. An EAN or UPC belongs in gtin13 or gtin12. Both fields hold identifiers, which is exactly why this gets missed, and they are not interchangeable to anything reading them.
Sizes living in the description. If 50ml and 100ml are a sentence rather than variants with their own price and stock, there is nothing for a price constraint to bite on. How we structure product data usually starts here.
Writing descriptions for the brand deck. Evocative copy about a windswept coastline tells an agent nothing. Notes, longevity, when to wear it. Boring works.
Assuming notes are on the page somewhere. They usually are, as an image or a paragraph. Neither is structured. Metafields are where that belongs, and almost nobody has done it.
Never opening the Agentic Storefronts screen. Know which channels are on before you form a view on whether any of this is working.
We’ll read your product data the way an agent reads it
Send us a product URL. We’ll tell you which attributes are structured, which are sitting in free text, and which are missing entirely, and what that costs you when a shopper asks for exactly what you sell.
The part that should worry you
You cannot see any of this in your analytics.
A shopper asks. Your product does not surface. Nothing happens. No bounce, no abandoned cart, no session, no entry in any report you look at on Monday morning. The sale goes elsewhere and your dashboard is identical to yesterday’s.
Every other channel you have ever optimised gave you a failure signal. This one does not.
LA is usually early on things like this, because the category rewards being early. But the brands that win the next two years in beauty will not be the ones with the most beautiful product photography. They will be the ones whose data was boring enough for a machine to read, filed by someone who checked.
