For the last few years, winning in retail media meant winning the search results page: the right keywords, the right bid, the right placement above the fold. That's starting to change, thanks to AI.
Walmart is now embedding ads directly into Sparky, its AI shopping assistant, and testing Marty, a beta advertising assistant that handles bidding and billing for Sponsored Search campaigns. Amazon is doing the same thing with Rufus and its newly unified Alexa for Shopping experience.
The fight for shelf space is moving off the search results page and into the AI agent's conversation. Our hot take? Most advertisers aren't ready for what that means for measurement, creative, or catalog data.
Why Walmart Is Leaning Further into AI
Walmart Connect isn't a scrappy underdog anymore. It generated an estimated $4.4 billion in revenue in 2024 and grew roughly 33% year-over-year in the U.S. in Q3 2025: fast enough that several retail media analysts now describe Walmart as having reached rough parity with Amazon on capability, even if it's still well behind on scale. Sparky and Marty are Walmart's answer to a simple competitive threat: if shoppers start asking an AI agent what to buy instead of typing into a search bar, whoever owns that agent controls the shelf. Walmart doesn't intend to cede that ground to Amazon.
Marty, specifically, is worth watching closely. It's built to manage bidding and billing on Sponsored Search, meaning Walmart is starting to automate the mechanical side of campaign management well before most advertisers have fully adapted their strategy to agentic discovery. That gap between platform automation and advertiser readiness is exactly where budget gets wasted.
Amazon Already Told Us What's Coming
Advertisers don't have to guess how this plays out: Amazon effectively wrote the playbook already. In June 2026, Amazon Ads published guidance explaining how agentic shopping through Rufus and Alexa for Shopping is reshaping product discovery, and the core message was blunt: demand is shifting from search results into AI agent conversations, and those conversations don't generate the keyword impressions or PPC attribution advertisers are used to reading. A brand can look at strong ROAS in its dashboard while quietly losing impression share to an agent that recommended a competitor instead.
Amazon's guidance pointed to three concrete adjustments advertisers need to make, and there's no reason to expect Walmart's version will look meaningfully different:
Creative has to work as a conversational answer, not just a product tile. An AI agent summarizing options in a sentence or two needs assets and copy that hold up outside a grid layout.
Catalog data becomes the ranking signal. Product attributes, review depth, pricing clarity, and certifications increasingly determine whether an agent recommends a product at all: this is quietly becoming as important as bid strategy. It’s important to note that there’s still no established rulebook for how brands should show up in AI shopping agents. Much of the strategy today is rooted in testing and learning, which is where agencies can provide significant value, using their scale and breadth of experience to test across brands, identify patterns, and determine what works.
Measurement needs a new blind spot check. Watch for flat or declining product page sessions next to steady or growing unit sales. That divergence is often the fingerprint of agent-driven purchases that aren't showing up in traditional attribution.
The Window Is Shorter Than It Feels
Amazon's own guidance frames this as a two-to-three-quarter window before larger, better-resourced competitors dominate the "agent shelf." That urgency should apply equally to Walmart's rollout of Sparky and Marty. Waiting for the tools to fully mature before adjusting strategy means ceding the early advantage to whichever brands get their catalog data and creative agent-ready first.
What This Means for Brands
Three checks are worth running across your Amazon and Walmart accounts this quarter.
- Start with catalog attribute completeness on your top revenue SKUs: attributes, reviews, pricing clarity, and certifications are what an agent leans on to make a recommendation, so gaps there are gaps in discoverability.
- Next, stress-test your creative and product copy by reading it as a one- or two-sentence conversational summary rather than a product tile. If it doesn't hold up outside a grid layout, it likely won't hold up inside an agent's answer either.
- Finally, build a simple weekly view of branded versus non-branded impression share so you can catch early signs of agent-driven discovery shifts before they show up as a revenue problem.
None of this requires waiting for Walmart's tools to fully mature or for Amazon to publish more guidance. The brands that get ahead of these three checks now will be the ones an agent actually recommends once agentic shopping is the default, not the exception.
Agentic AI in retail media isn't a future trend to plan for later: it's already live on two of the largest platforms in the category. The advertisers who treat it as a "wait and see" story are the ones most likely to lose shelf space they don't even know they're losing.