A category manager at a major beauty brand did something in front of me recently that I keep thinking about. We were chatting when she picked up her phone and asked it to find a gift for her sister. Her own products didn’t show up. A smaller competitor’s products did—twice. She laughed at first. Then she stopped laughing. She had just watched her category get shopped without her.
During the coming holiday season, a growing share of shoppers will start the hunt inside an artificial intelligence assistant—ChatGPT, Gemini, Microsoft Copilot—rather than on a search engine or a brand’s own site. Google and Shopify put out a shared open standard this year, the Universal Commerce Protocol, so that AI assistants can read a merchant’s live catalog—size, shade, price, stock—instead of scraping a product page. OpenAI rebuilt its shopping experience inside ChatGPT: A shopper describes what they want or uploads a photo, compares options, then finishes the purchase on the retailer’s site. The card networks have built the rails, too. Visa Intelligent Commerce and Mastercard Agent Pay issue tokenized credentials scoped to an AI agent, with limits the shopper sets in advance.
None of this AI choosing will ever be visible to shoppers. They will just ask for something under $50 for a friend with sensitive skin, and an answer will appear.
This doesn’t replace how people buy clothes and makeup. It adds one more way in. And it works differently from every other customer entry point you have built.
WORDS ARE DOING THE JOB IMAGES USED TO DO
Fashion and beauty are great at persuading through pictures. It’s all about the campaign, the packaging, and the way a product reads under studio light. When an AI assistant makes the first pass—reading product data, not looking at photographs—none of your visual marketing matters. What matters is written product information. For most companies, that’s the saddest document in the building. Somebody wrote the product descriptions fast on a deadline. They got copied between three systems, and nobody has looked at them since.
That neglected text is now a storefront. If the description of a coat doesn’t say what it’s made of or how it fits, there is nothing for an AI assistant to go on. If a foundation listing describes the shade in language only your team understands, it gets passed over for a brand that wrote it plainly. Fame—a well-known brand name, a celebrity face—helps you less here than plain description does, which I realize is an annoying thing to hear after years of building it.
PEOPLE ASK FOR THINGS IN A MESS
In-person requests never sound like the keyword searches brands optimized for, and neither do the questions people ask an AI assistant. Someone wants a product that won’t irritate their skin, a replacement for a discontinued lipstick, or a gift for a mother-in-law. Brands that read well to an AI assistant tend to have product copy that sounds like it was written by somebody on the shop floor. Delivery timing, stock on hand, and returns and exchanges get pulled into this, too. They used to live in the fine print. Now they are part of the answer, because the shopper asked about them.
WHAT WOULD I DO IF I RAN A CATEGORY RIGHT NOW
Most of this is editing, not technology. It can and should be done before the season starts. The steps run in order: Each one tells you where to point the next.
- Shop for yourself first. Ask an assistant for a gift in your category at your price point. Then ask for something suited to a particular skin concern or occasion. See if your products show up, whether the details are right, and what brands or products are standing next to yours.
- Don’t try to fix the whole catalog. Start with the gifting price points and the hero shades and sizes—those few that carry most of a line’s volume. That is where being clear pays back inside a quarter.
- Use AI as a blunt editor on your own listings. Paste one in and ask what a customer still wouldn’t know after reading it. The answer comes back in seconds: It is usually fit, fabric, finish, skin type, or how long the thing lasts.
- Ask it to summarize your product and a competitor’s. Putting those two paragraphs side by side ends internal arguments faster than any deck I have ever built.
- Let AI pull existing information into your descriptions. You already hold the missing information in tech packs, laboratory sheets, and supplier documents. Have a person check the result. While AI’s speed is worth having, the errors are not.
- Find what is wrong in what AI surfaces. Shades that you discontinued are still listed. Sizes showing as available that aren’t. Returns terms nobody can locate.
- Add one question to your range reviews, next to margin and sell-through: Can this product be described in one plain sentence a stranger would understand?
THE PART WORTH PROTECTING
It would be easy to overcorrect. Nobody has ever fallen in love with a fragrance because a machine described it accurately. Desire still comes from images, from story, and from someone in a store paying attention to you. The AI assistant only handles a narrow slice—the practical middle where a person is comparing options and wants to stop thinking about it.
You have two consumers now. One is a human being you are trying to move. The other is an AI system you must feed with descriptions tailored to how consumers query. The failure to watch for is the second one bleeding into everything and flattening how your brand sounds and resonates. The emotional part is still where the margin lives.
The shelf that matters most this holiday season is one you can’t see. To get onto it, you must describe your products honestly and plainly enough that an AI tool can tell what they are and who they are for. That used to be back-office tidying, handed to whoever happened to have time. I would give it to someone senior this year instead.
Charisma Glassman is a senior partner and global head of retail, consumer, and e-commerce at Genpact.