Your stockists’ listings are your product data
The lubricant carousel wasn’t built from any one brand’s website. One of the best-known lubricants in the category appeared twice, at two different prices. Another brand’s card showed its vaginal moisturiser, whilst the written answer beneath it recommended that brand’s lubricant. The answer also quoted a live offer price from Boots.
Neither brand necessarily controlled how its products were presented. The model assembled it from the listings it could find, and many of those belong to retailers and distributors rather than manufacturers.
So a retailer’s product page now does two jobs. It still serves the customer on that site, and it may also be the version of the product an AI shows to everyone who asks. Since March, Shopify has made eligible merchants’ products discoverable in ChatGPT by default through its Agentic Storefronts, without requiring a separate integration. Our UK tests showed just how much of the information presented to customers can come from third-party websites, rather than the manufacturer's own.
Complete, current and accurate product listings give AI systems useful information to draw on, whilst also helping customers make informed decisions.
The spec sheet is the sales sheet
The written answers were the most revealing. The vibrator answer gave one product a reading of 38 to 42 decibels at 30 centimetres and another 41 to 45 decibels, and named its sources: Mumsnet, a specialist review site, a lifestyle magazine and the manufacturer’s own website. The condom answer quoted thickness to the thousandth of a millimetre, and width. The lubricant answer cited pH and the absence of glycerine.
The examples demonstrate the value of publishing useful technical details, although any figures quoted by AI still need checking against their original sources. “Whisper quiet” gives a model little to repeat. “38 dB at 30 cm” gives it something specific, provided the measurement is accurate and the testing conditions are clear.
Having listed its picks, the vibrator answer offered to narrow them to toys tested for how audible they are through a closed bedroom door. That is exactly the kind of question a real customer might ask next. A brand that has run that test and published its method has something useful to offer, both to customers and to the sources an AI might draw on.
Technical specifications often sit on a distributor’s sheet, in a trade catalogue or in a PDF nobody indexes. Marketing language still does its job with the customer. A measured figure on a crawlable page also gives an AI answer something specific to reference and an independent site something to check.
One further observation, from a single question rather than a pattern: three of the five vibrators named came from the same brand. We can't say why those products were selected, but for everyone else, there was no page two in that answer. It's a reminder that reviews, forums and retailer pages deserve attention alongside your own website.