Fourth on its own shelf, with zero of the sources ChatGPT was reading

The shelf, before we touched anything

We ran the category the way a shopper would. Brand-free questions — best checked luggage, carry-on that fits an overhead bin, luggage for a two-week trip — a few hundred times, and counted who got named.

The three legacy leaders took 39%, 38%, and 24% of the answers. Our brand came fourth at 4.3%. The gap wasn't about the product — it was about what ChatGPT could actually read.

The part that did: they owned 0% of the sources. Every page ChatGPT read to build its answer belonged to somebody else — retailers, roundup blogs, forum threads. The brand's own site wasn't in the conversation. It was being described by third parties working off old information, and it had no say in it.

What the audit turned up

None of this was a marketing problem. It was three things quietly broken in the plumbing between the store and everything downstream of it.

• 01

  Product descriptions never made it into the feed

  The site had good copy. The feed didn't. Descriptions were being dropped at export, so the machine-readable version of the catalog — the version that actually feeds AI shopping surfaces — was close to empty where it mattered. ChatGPT was working from titles and a price.

• 02

  The Shopify feed was mis-mapped across the catalog

  Fields were landing in the wrong slots. Material in the wrong attribute, dimensions inconsistent between parent and variant, product type set to something generic across whole collections. When the mapping is wrong, the catalog doesn't look wrong — it looks thin. And thin loses to a retailer listing that filled every field.

• 03

  Retailers were the source of truth, and their data was stale

  Because the brand's own data was unusable, the surfaces fell back on marketplace and retailer listings. Some of those were years out of date — discontinued colorways, old dimensions, features the current product doesn't have and features it does have that never got mentioned. The brand's own differentiators were invisible.

What we did

• Rebuilt the product feed to the full attribute spec and fixed the mapping across the catalog, not just the heroes.
• Restored descriptions and structured attributes end to end so the machine-readable catalog matched the site.
• Corrected the wrong and stale product data at the source, so third-party listings had something current to pull from.
• Built a prompt universe specific to how people actually shop luggage — trip length, bag type, the constraints travelers name out loud — instead of generic category queries.
• Turned on Trace attribution on Shopify and locked a day-zero baseline before any of it could move.

Four weeks later

Shelf movement usually takes 60 to 90 days. This moved faster because the problem was mechanical — the data was wrong, and once it was right there was nothing left holding the brand back.

Nothing here is a projection except the one line that says so.