Retailers are building AI shopping agents. Most CPG brands aren’t ready

Two reports landed this month that say the same thing from different angles, and a Walmart release on Friday showed why it matters.

On October 30, BCG published a survey finding that 59% of U.S. CPG companies report no meaningful agentic AI capabilities. Nine days earlier, Kearney benchmarked 106 consumer companies and found that only 2 of 53 consumer products makers run anything live in agentic commerce. Five of six digital retail platforms do.

The same day BCG published, Walmart gave its Sparky assistant the ability to turn a photo of a handwritten shopping list into a cart. Retailers are building the agents. Brands are mostly still working out how to get mentioned by them.

What did BCG find about CPG readiness for agentic commerce?#

BCG’s report pulls from several of its own surveys. On the consumer side, its December 2025 GenAI Consumer Survey found 80% of U.S. consumers already use AI, and nearly half of those use it to shop. Among those AI shoppers, 96% use it for product discovery, 80% for consideration and 24% for purchasing. Only 4% of consumers said they would let an agent complete a purchase alone.

On the company side, BCG’s GenAI CPG Survey 2026 found 41% of U.S. CPG companies have begun building some agentic AI capabilities. The other 59% report none that matter.

BCG also says 32% of retailers give direct data access only to selected agents, and 87% don’t share customer data with agents at all. A brand can’t assume a retailer will feed its product story into the retailer’s own assistant.

How far behind retailers are consumer brands?#

Kearney’s benchmark, published October 21 and summarized by eMarketer on October 30, rated 12 months of each company’s own communications (September 2025 to August 2026) on a five-level scale, from no mention of AI to a live agentic commerce offering. Quiet work won’t show up, but the gradient is steep.

The gap widens when you count live offerings instead of strategy decks.

Kearney’s category read is useful for anyone in CPG. Beauty leads and holds one of the two live offerings in consumer products. Home care is concentrating on retailer agents such as Rufus and Sparky. Pet care shows the broadest activity. Packaged food, sitting on decades of recipe content, “has barely started.”

The other finding is about seniority. In retail, agentic commerce shows up on earnings calls and in risk disclosures. In consumer products, Kearney says the topic usually sits one to two levels below the executive committee. The work is real and often funded, but it’s run by digital and ecommerce leaders, not the board.

Why do retailer agents matter so much for CPG brands?#

Because the retailer owns the moment the brand gets picked. Look at what Walmart added to Sparky on October 30: turn a handwritten or Notes app list into a cart, describe a meal and get recipes with ingredients ready to buy, and ask for healthier swaps like “Find me a lower-sugar alternative to these marshmallows.”

Think about what a shopping list actually says. “Paper plates.” “Broth.” “Marshmallows.” Those are categories, not brands. Every one of those features takes a generic line and turns it into a specific product, and Sparky makes that call. If you sell a household or pantry staple, that’s your shelf now. Walmart’s release doesn’t say how Sparky ranks those picks.

So AI visibility isn’t only a ChatGPT problem. ChatGPT, Gemini and Perplexity shape the shortlist (we covered how ChatGPT picks brands before it searches). The retailer agents, from Sparky to Amazon’s Alexa for Shopping, make the swap at the point of purchase. Brands need to show up in both.

Where are CPG companies putting money next year?#

Mostly into getting found. BCG asked CPG companies where they plan to invest over the coming year:

BCG calls GEO the “no-regrets move” but warns it won’t differentiate on its own once everyone does it. We agree with half of that. Visibility is table stakes. But a brand that can’t say which of its SKUs an assistant recommends for the questions that matter is nowhere near the point where everyone does it.

Does AI visibility work actually move the numbers?#

The best public example is Oreo. Mondelēz had blocked AI crawlers to protect its content. Andrew Lederman, its VP of global digital commerce, told Digiday in April that Oreo was cited in only about 10% of AI chatbot cookie recommendations. After unblocking the necessary crawlers and rebuilding its brand sites to be machine readable, Oreo came up about 70% of the time for specific cookie queries, and closer to 30% for broader snack topics.

BCG reports CPG companies piloting GEO see 1% to 5% incremental revenue on average, and those testing owned agents and ecosystem integration see 3% to 10%. It also describes a luxury fashion group that prioritized 500 consumer prompts, ran about 10 pilots starting with structured data and metadata fixes, and more than doubled its brand site’s citation rate in AI search within eight weeks.

And the traffic is worth chasing. Adobe found AI-referred visits to U.S. retail sites grew 393% year over year in Q1 2026 and converted 42% better than non-AI traffic in March. The same report scored the average U.S. retail product page at 66% machine readable, so a third of the content on a typical product page is invisible to the models.

To make BCG’s ranges concrete, plug in your own numbers. This is arithmetic on BCG’s reported ranges, not a forecast for your brand.

What should a CPG brand do in the next 90 days?#

Unless you’re in beauty or pet care with real advisory data, skip the agent build for now. Do this first:

• Check that AI crawlers can reach you. Mondelēz says its own crawler blocks were part of why Oreo sat at 10%. Look at robots.txt and your bot management settings on the brand site this week.
• Write the prompt list. Pull 50 to 100 real shopper questions per category, including generic list lines like “paper towels” and “lower-sugar cereal.” BCG’s luxury example used 500.
• Measure by SKU, not by brand. Track which product each assistant names, in ChatGPT, Gemini, Perplexity, Rufus and Sparky. A brand-level score hides the SKU that’s losing.
• Make promotions and pack sizes machine readable. Kearney’s point: agents compare price, promotion, pack size and value claims across retailers in seconds. Count, size and cost per use belong in titles and structured data.
• Give it an owner with an executive sponsor. If it lives two levels down, it gets the leftover budget.

The brands that close that gap first get to set what the agent says about their category, and everyone else will be reacting to it. For the restock side of this, see our post on AI agents and the household reorder. For more CPG and ecommerce charts, see Xenalytics.

If you want to see which of your SKUs ChatGPT, Gemini and Perplexity recommend today, run a free SKU audit or book a demo. How we score each product is on the SearchPanel homepage.