// AGENTIC COMMERCE · FOR JEWELERS
A quarter of one jeweler's
online revenue now comes from AI
People shop for jewelry by describing it — a one-carat oval lab-grown solitaire in yellow gold, a men's band that won't scratch. That is exactly the query an AI assistant answers, and exactly the structured data most jewelry catalogs don't have. We make yours readable.
// BROWNLEE JEWELERS · CHARLOTTE, NC · SEVEN STORES · NINETY YEARS
In February this channel did not exist for them
24%
of online revenue now comes from AI assistants
29×
more likely to buy than the average visitor
550%
ChatGPT growth in a single 30-day window
Zero
advertising spend behind any of it
Their first order from an AI assistant is dated 10 May 2026. The figure was verified two independent ways — Shopify's own Agentic Storefronts reporting and a first-touch trace of every order — and the two agree to within a percentage point.
+300%
orders, 30 days over prior 30
+256%
online revenue, same window
+77%
average order value since the work began
+15%
sessions — the traffic barely moved
Results from one store over one period, published with permission. They followed six months of catalog remediation and are evidence this work pays off in jewelry — not a forecast for your store. No agency controls what these platforms surface.
// WHY THIS CATEGORY, SPECIFICALLY
Jewelry is a described purchase
Nobody types “ring” and buys the first result. They specify stone, shape, carat, metal, setting, size, and a price band — then judge the answer. That is a structured-attribute query, and it maps one-to-one onto the fields these systems index.
Which is why the gap in this category is so wide. Jewelry catalogs are the least likely to carry clean identifiers — so much of the inventory is estate, bespoke, or one-of-a-kind — and the most likely to hold their real attributes inside prose a machine cannot parse. The buyer is perfectly matched to the surface. The data usually isn't.
Add the average order value of a jewelry sale and each of these pre-qualified referrals is worth several times what it would be in most categories.
// WHAT THE AUDIT FINDS ON A JEWELRY CATALOG
Six failures that don't show up in your admin
Share-of-catalog figures below are from a live jewelry store that had already been through six months of remediation. An untouched catalog scores considerably worse.
Roughly a quarter had no GTIN
Barcode data missing across about 23% of active products — on a catalog that had already been through six months of remediation. Missing identifiers are a named exclusion reason, and jewelry is the worst-affected category because so much of it is estate, bespoke, or one-of-a-kind.
6% had titles an agent can't use
Bare SKUs, numeric strings, and fragments too short to parse. A shopping assistant matching a described purchase cannot do anything with a title like "14K-R-0382" — the piece is invisible no matter how good it is.
6% had descriptions too thin to read
One block of unstructured text, or a couple of words. These systems parse attributes — stone, shape, carat, metal, setting, fit. Jeweler's voice with no structure underneath reads to a machine as an empty product.
Taxonomy category missing entirely
Products with no Shopify or Google product category assigned. Without it, a piece never enters the candidate set for a category query, regardless of how well the rest of the listing is written.
Variant and sizing data that doesn't survive
Ring sizes held as free text, inconsistent metal values, carat weights stored as prose. Jewelry has more meaningful variant attributes than almost any category, and they are the exact fields a described purchase filters on.
Products not published to the sales channel
Pieces live and selling on the website but never published to the Shop channel that feeds the agentic surfaces. Nothing errors. They simply are not eligible, and nothing in the admin tells you.
// SETUP, THEN ONGOING ENRICHMENT
Scoped on the size of your catalog
Setup fixes the catalog you have. The ongoing monthly work keeps it fixed as you add inventory — which, in this business, is constantly. Same scope at every tier; only the piece count changes the number, fixed on a call.
Single Store
Up to 750 pieces
Engagement
One-Time Setup
Then ongoing
Monthly Enrichment
6-month minimum, then month-to-month with 30 days notice. Setup and first month are billed together.
Multi-Store
750 – 3,000 pieces
Engagement
One-Time Setup
Then ongoing
Monthly Enrichment
6-month minimum, then month-to-month with 30 days notice. Setup and first month are billed together.
Group
3,000 – 10,000 pieces
Engagement
One-Time Setup
Then ongoing
Monthly Enrichment
6-month minimum, then month-to-month with 30 days notice. Setup and first month are billed together.
Over 10,000 pieces, or a group already running on a PIM? We'll quote it →
// THE SETUP
Fixing the catalog you have
- Full catalog audit against the AI shopping feed specs
- Identifier strategy for estate, bespoke, and one-of-a-kind pieces
- Structured jewelry attributes: stone, shape, carat, metal, setting, size
- Titles rewritten so an agent can parse them — no bare SKUs
- Shopify + Google taxonomy categories assigned across the catalog
- Crawler access corrected at robots.txt and the CDN/WAF layer
- Product + Offer schema shipped to your product pages
- Every agentic channel enabled per surface and verified with live queries
// THE MONTHLY
Keeping it fixed as stock turns
- Every new piece you add, enriched to spec before it goes stale
- Feed health monitoring with alerting on rejections and silent drops
- Tracked buying prompts for your categories and price bands
- Crawler access monitoring — catches CDN rule changes within days
- Surface-change briefings when a platform revises its requirements
- Monthly report plus a fix window for whatever it surfaces
// QUESTIONS WORTH ASKING
Including the skeptical ones
Are the Brownlee results typical?
No, and it would be dishonest to imply otherwise. That store had six months of catalog remediation behind it before the AI revenue appeared — categorization, structured product data, roughly 11,500 metafield entries, descriptions rewritten from single blocks of text. The AI assistants could read the catalog because there was finally something to read. The results are evidence that the work pays off in this category, not a forecast for your store.
Why does jewelry do unusually well on these surfaces?
Jewelry buying is a described purchase. People ask for a one-carat oval lab-grown solitaire in yellow gold in a given budget band, or a men's wedding band in tungsten that won't scratch. That is a structured-attribute query, and it maps directly onto the fields these systems index — stone, shape, carat, metal, size, price band. A shopper arriving from that query has already specified what they want and been shown your piece as the answer. High average order value on top of that makes each referral worth far more than in most categories.
My pieces are one-of-a-kind or estate. I can't have GTINs.
Correct, and that is exactly why jewelry catalogs need handling rather than a generic feed tool. The specs let you declare that no identifier exists rather than leaving the field empty — those are different states, and the empty one gets you filtered. Estate and bespoke pieces need that declared properly, plus enough attribute density elsewhere to stay competitive without an identifier to match on. Generic feed apps do not do this correctly.
Why is there a monthly fee at all — isn't this a one-time fix?
It would be, if your inventory stopped changing. It doesn't. Every new piece you take in arrives with whatever data your vendor or POS gave it, which is rarely enough to be eligible. Without ongoing enrichment your remediated catalog decays back toward invisible one intake at a time, and the newest stock — usually the pieces you most want moving — is the least visible. On top of that the surfaces revise their merchant requirements regularly and feeds fail silently. The setup fixes what you have; the monthly is what keeps it true.
What does the setup actually find on a jewelry catalog?
On a roughly three-thousand-piece catalog that had already been through six months of remediation, the audit still surfaced about a quarter of products missing barcode/GTIN data, 6% with titles a shopping agent cannot use (bare SKUs or numeric strings), 6% with descriptions too thin to parse, and a set of products not published to the Shop channel at all. On an untouched jewelry catalog the numbers are considerably worse.
How do you decide which tier I'm in?
Active product count, confirmed against your store before we scope the work — not an estimate you have to make on the spot. If you're near a boundary we place you in the lower tier. Over ten thousand pieces, or a group already running a PIM, we quote it rather than pretend a fixed tier fits.
Can you guarantee my store gets surfaced in ChatGPT?
No. Nobody outside OpenAI, Google, Microsoft, and Perplexity controls what gets surfaced, and their rules change regularly. Treat any agency that guarantees placement as a red flag. What we control is whether your catalog is eligible, technically reachable, and accurately represented — and then we measure what actually happens for the queries your buyers use.
Do I need to leave Shopify or add another platform?
No. Shopify's Agentic Storefronts already syndicates your catalog to these surfaces from your own admin, and the reporting we measure against is Shopify's own. This work happens inside the store you already run — the catalog data, the crawler access, the schema, and the channel configuration. If you are not on Shopify we can still audit you, and we quote implementation separately.
Your catalog is either legible or invisible
There is no third state, and nothing in your admin tells you which one you're in. Tell us roughly how many pieces you carry and we'll tell you where you stand.
Not a jeweler? The general version is here →