Influencers won't sell your products anymore. Shopping agents will.

Key takeaways

• AI-sourced visits to US retail sites grew 393% year over year in Q1 2026 and converted 42% better than all other traffic (Adobe, via CFOtech).
• Product pages are where AI struggles most. Adobe rates their machine readability at 66 out of 100, lower than any other page type.
• By 2030, agentic commerce could bring in up to $1 trillion in US retail revenue, according to ICSC and McKinsey (Retail Dive).

What this means for retailers: Agents recommend what they can read and trust. Clean product data, consistent prices and reliable stock levels now earn a place on the prio list.

---

For years, products took off because the right person held them up to the camera. A creator posted, a campaign followed, and sales came in. That chain is starting to break. More and more people now type their question straight into an AI assistant, something like "Yellow running shoe under $120?" or "What's a good gluten-free snack for a kids' party?", and get back three suggestions. Agentic commerce is a name now – AI shopping agents compare products, prices and reviews on a customer's behalf, and in the future they will more and more complete the purchases too.

Andy Jassy isn't impressed yet. On Amazon's earnings call on October 29, he called outside shopping agents "still not great, nor more convenient than coming to a retailer directly" (Fortune). In the same breath, Amazon reported more than 400 million customers for its own shopping assistant over the past year. So agents are far from perfect, and people in the US use them anyway: In an Adobe survey of more than 5,000 US consumers, 39% had already used AI for online shopping, and 85% of them said it made shopping better.

Europe is slower to follow. In McKinsey's ConsumerWise survey from early 2026, shoppers in Germany, France, Italy, Spain and the UK were less likely than Americans to use AI for shopping, and fewer than one in four of those who did let it handle checkout or repeat orders (McKinsey).

An agent can only recommend what it can read

Product catalogs were built for people. A shopper sees the photo and knows the shoe is yellow, skims three reviews and learns that the fit runs small. Much of what sells a product lives in images, videos and the feel of a brand, and that has worked well for decades.

An agent goes about it differently. It can look at pictures and even watch a video, but to answer "yellow running shoe under $120" it has to check thousands of products against that request within seconds. For that job it leans on what is written down in a structured way: titles, attributes, prices, stock levels and reviews, on the site and in product feeds. If "yellow" only exists in the photo, or the size sits in the description text while the size field stays empty, the shoe is easy to overlook. The same goes for a variant that isn't linked to its parent product, or a stock status that lags a few hours behind. None of this mattered while only humans were reading. Now a second reader has arrived, and it needs the same information in a different form.

Adobe puts numbers on the gap: Traffic from AI keeps climbing and converts well, yet product pages remain the weakest part of most retail sites for machines. "An average homepage score of 75% means about a quarter of homepage content across the sector is not machine-readable," Adobe writes. Product pages score lower still (CFOtech).

Retailers know this effect from Black Friday: a product with incomplete data doesn't show up in filters or promotions. A shopping agent passes over it for the same reason.

What shopping agents look at

• Attributes. Size, material, ingredients, fit and compatibility, filled in and formatted the same way across a category.
• Hierarchies and variants. They help an agent land on the right item and offer an alternative when the first choice is sold out.
• Prices. An agent lines up comparable items from several stores in seconds, so a gap on a key item can cost you the recommendation (competitor price monitoring with like-for-like matching).
• Stock. A great suggestion is worthless if the item is gone.
• Ratings and reviews. For an agent, they are the closest thing to word of mouth.
• The assortment itself. No amount of clean data helps with a product you don't carry.

Why merchandising now decides what gets recommended

Brand building still has its place of course. Someone who asks an agent for one particular brand of running shoe already knows what they want, and that wish often goes back to years of advertising or to a creator they follow. But in many of the questions people type into an agent, no brand comes up at all. A person who types "the best rain jacket for cycling" leaves the choice to the agent, and the agent picks by comparing product data, prices and reviews. Which label ran the biggest campaign hardly matters in that comparison. For retailers, part of the work therefore shifts toward category managers, merchants and buyers, the people who own the assortment, the product data and the prices.

There's a second upside. The questions people put to agents mirror the needs that surface in search, social media and reviews, often weeks before they show up in sales. We looked at how to read those signals in what to stock next.

Four checks to get your assortment ready for shopping agents

1. Ask the agents yourself. Pick your top categories and put your shoppers' typical questions to an AI assistant. If a competitor's product shows up where yours should, compare the two listings side by side.

2. Complete your product data. Fill in the attributes that are still missing, for bestsellers and the long tail alike: size, color, material, ingredients. Link variants to their parent product and keep categories consistent. Every attribute you add gives an agent one more question your product can answer. Our checklist for product data before Black Friday walks through the steps.

3. Know your price position. Focus on the handful of items that shape your price image, since those are the first ones anyone compares.

4. Find the gaps. Note what people ask for in your categories that you don't stock yet. Then decide what to add, and which slow sellers make room for it.

How AI agents get retailers ready

By hand, a team gets through a few hundred products of the assortment. AI agents for category management cover the whole assortment. They clean and structure product data, flag missing attributes and broken variants, and line up products like for like through product matching. Assortment agents spot gaps against competitor ranges and shopper demand. Pricing agents weigh prices against the market and suggest changes within your margin guardrails. Shopper insights agents pick up what people search for, ask about and review. The final call stays with your category team.