The Authorization Gap: What Amazon’s Perplexity Injunction Reveals About the Commerce Layer Walmart Is Building Differently

When U.S. District Judge Maxine Chesney granted Amazon a preliminary injunction against Perplexity on March 10, blocking the startup’s Comet browser from accessing password-protected Amazon accounts, the ruling produced an immediate wave of coverage focused on the legal drama: a plucky AI startup versus the world’s largest e-commerce platform, cease-and-desist letters ignored, technical blocks circumvented within 24 hours, a federal judge concluding that Amazon demonstrated “strong evidence” of unauthorized access. That framing is accurate but misses the more consequential point for anyone selling through Walmart.

The case turned on a single finding. Chesney ruled that Comet accessed Amazon accounts “with the Amazon user’s permission, but without authorization by Amazon.” User consent and platform authorization are not the same thing. A shopper who hands Perplexity’s agent their Amazon login credentials has not granted Perplexity the right to operate on Amazon’s platform. Those are separate questions, and the court treated them separately. That distinction, unremarkable in abstract contract law but genuinely novel as applied to AI agents completing retail transactions, establishes something that will shape how every authorized agentic commerce system is structured from here forward. Walmart is building its agentic infrastructure on the authorized side of that line, and the architecture it is assembling has specific implications for how suppliers compete for discovery, advertising placement, and conversion in the next phase of retail.

The Architectural Fork in Agentic Commerce

Amazon’s posture toward third-party AI agents has been consistent: block them, litigate if necessary, and build proprietary alternatives. In November 2025, the company separately blocked ChatGPT from shopping on its platform. Decrypt reported that Amazon’s updated Business Solutions Agreement, effective March 4, 2026, formally requires all AI agents to identify themselves when accessing its services. Amazon CEO Andy Jassy acknowledged the potential of the category on an earnings call, describing agentic commerce as having “a chance to be really good for e-commerce,” but qualified that agents aren’t yet accurate enough on personalization and pricing to justify uncontrolled third-party access.

Walmart’s approach is structurally different, and suppliers should understand why that difference matters before it shows up in their Connect budget conversations.

In October 2025, Walmart announced a partnership with OpenAI enabling customers to purchase Walmart products within ChatGPT using Instant Checkout. Two months later, at NRF in January 2026, Walmart and Google announced a new shopping experience built on the Universal Commerce Protocol (UCP), an open standard co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, and endorsed by more than 20 companies including Mastercard, Visa, Home Depot, and Best Buy. UCP is designed to establish a common language between AI agents, merchant systems, and payment providers, replacing the bespoke, platform-by-platform integration model that creates exactly the kind of friction Perplexity tried to work around.

The UCP approach inverts the dynamic that produced Amazon’s lawsuit. Rather than forcing external agents to scrape or circumvent platform protections, it gives authorized agents a standardized way to announce themselves, negotiate what they can handle, and complete transactions through infrastructure the retailer controls. Walmart CEO John Furner described the company’s orientation explicitly in the NRF announcement: “The transition from traditional web or app search to agent-led commerce represents the next great evolution in retail. We aren’t just watching the shift, we are driving it.”

That is not a branding statement. It is an architectural choice with specific implications for how product discovery, advertising, and conversion will function on Walmart’s platform in the near term, and for which suppliers are positioned to capture what gets built on top of it.

What the Sparky Data Is Already Telling You

Before the legal frameworks for agentic commerce fully solidify, Walmart’s own consumer data is providing early signals about how shopper behavior is changing, and the numbers suggest the shift has already moved past the experimental stage.

On the Q4 FY26 earnings call on February 19, Walmart CEO John Furner reported that customers who use Sparky have an average order value approximately 35% higher than those who do not. Walmart U.S. President and CEO David Guggina added that roughly half of the company’s app users have now engaged with the assistant. According to a Walmart Connect survey conducted in August 2025 through the Walmart Customer Spark Community using Walmart Scintilla data, 81% of Walmart customers reported using Sparky to check product availability or review product specifications before making a purchase.

Those figures deserve operational translation. A shopper arriving at a product detail page via traditional search has already made a category decision and is in evaluation mode. A shopper arriving via Sparky may be in an entirely different part of the journey. Sparky is designed to handle occasion-based planning. Khurrum Malik, VP of Business and Product Marketing at Walmart Connect, described the typical use pattern at CES 2026: customers asking the agent to help them prepare for a birthday party or solve a specific problem, with the agent reasoning across categories and assembling a basket rather than answering a single product query.

This matters for how suppliers think about their item setup and content quality in Supplier One. Sparky’s ability to accurately recommend a product in a multi-item, occasion-based context depends on the completeness and specificity of the underlying item content. If product attributes, use cases, and compatibility information are thin or generic, the agent’s ability to surface that item in a high-intent, basket-building session is compromised. The Automated Brand Shop Builder and the broader push toward richer content attributes are not peripheral concerns for teams already running strong on OTIF and SQEP compliance. As Sparky’s influence over pre-purchase consideration grows, content completeness becomes a prerequisite for the new discovery surface, not an optimization deferred until after other priorities are met.

The New Advertising Equation Inside Agent Interfaces

The Amazon injunction is partly an advertising story. Amazon’s complaint noted that Perplexity’s automated traffic posed risks to customer account data but also added strain to ad systems where automated traffic must be identified before advertisers are billed. When an AI agent operates on a platform without identifying itself, the platform cannot reliably distinguish that traffic from organic human behavior. The economic exposure is direct: ad impressions served to an unidentified bot session generate billing without delivering genuine reach.

Walmart is addressing this from the design layer rather than the enforcement layer. Rather than blocking third-party agents and managing exposure after the fact, it is building the agent-to-ad relationship into the platform architecture from the start.

Walmart Connect began testing advertising formats within Sparky in fall 2025, with Walmart Connect’s January 2026 blog post confirming continued testing through this year. The formats are still developing. Based on Walmart Connect’s description, ads within Sparky can appear as sponsored prompts when shoppers ask the agent for product recommendations. That framing is significant: in traditional sponsored search, a shopper enters a keyword and sees a sponsored placement alongside results. In an agent interaction, the shopper states an intent and the agent reasons toward a recommendation. Sponsored placements inside that reasoning process function differently from keyword-match bidding, and the auction mechanics are still being defined.

What is not ambiguous is the scale of the advertising business Sparky sits inside. Walmart reported global advertising revenue of nearly $6.4 billion for fiscal year 2026 (ended January 31, 2026), a 46% increase year over year, with Walmart Connect U.S. growing 41% in Q4. eMarketer projects U.S. retail media ad spend will reach $69.33 billion in 2026, up from $58.79 billion in 2025, with the category potentially approaching $98 billion by 2028.

The Marty advertising assistant, in beta for Sponsored Search campaigns and expected to roll out to all sponsored search advertisers in the first half of 2026, operates alongside Sparky rather than inside it. According to Walmart Connect, Marty allows advertisers to build, optimize, and troubleshoot campaigns through a conversational interface in Ad Center, with access to four research reports: change analysis, account impression share of voice, account keyword impression share of voice, and account category benchmark opportunity. Walmart reports that 97% of user queries in the beta have been unique, which signals that advertisers are applying the tool to highly specific questions about their own accounts rather than running standard queries.

For suppliers running Sponsored Search, the near-term action is to engage with the Marty beta before broad rollout. The four reports available through conversational chat, particularly the category benchmark opportunity report, provide a faster path to identifying underbid positions and share-of-voice gaps than pulling the same data manually through Ad Center. The advertisers who are fluent with the assistant’s capabilities when full rollout happens in the first half of 2026 will have a structural advantage in bid optimization during the adjustment period.

The Attribution Problem That Agents Create, and What Walmart Is Testing

One complexity the Amazon-Perplexity case surfaces, and that Walmart suppliers will need to monitor, is what agent-driven traffic does to standard attribution models.

When a shopper uses Sparky to build a basket and complete a purchase, the conversion path looks different from a shopper who clicked a Sponsored Products ad and landed on a product detail page. The agent may have considered the product alongside competitors, weighted review data, checked availability, and included the item in a multi-product recommendation before the shopper confirmed. Traditional last-click and view-through attribution models don’t cleanly map to that journey.

Walmart Connect is addressing this through planned end-of-campaign reports and path-to-conversion reports announced for the Walmart Connect Ad Center in FY27. These are intended to translate omnichannel retail data into clearer performance intelligence. Walmart’s closed-loop measurement capability, connecting ad spend to both online and in-store sales through its first-party data, is already the foundation of how Walmart Connect differentiates its attribution from non-retail media alternatives. The path-to-conversion reports will extend that logic into agentic interfaces.

For suppliers managing Connect budgets now, the practical implication is to begin treating Sparky interaction data as a leading indicator of demand that may not yet fully appear in sponsored search conversions. Scintilla’s Shopper Behavior module is the relevant starting point. If a specific item’s shopper journey shows increasing consideration from non-search surfaces, that is an early signal of traction in agent-assisted discovery, and Connect investment allocation should reflect that rather than waiting for conversion data to catch up with behavior data.

What the Authorization Gap Means for Marketplace Sellers

The injunction’s core finding, that user consent and platform authorization are separate requirements, has a direct parallel in how Walmart structures access for Marketplace sellers and their technology partners.

Sellers using third-party tools that interface with Seller Center, pull data from Scintilla, or automate listing management need to ensure those tools are operating through authorized integrations rather than scraping or mimicking browser sessions. The Amazon case makes clear that a seller granting a third-party tool access to their own account does not automatically give that tool the right to operate on the platform. As Walmart continues to formalize its agentic commerce infrastructure through UCP and its partner super agent ecosystem, the expectation that all third-party agents identify themselves and operate through sanctioned channels will only tighten.

This is worth raising with any solution provider whose tools touch Supplier One, Seller Center, or Scintilla. The question is not whether the tool works. It is whether it operates through documented, authorized API connections rather than methods that replicate the identity-masking behavior Amazon documented in its complaint against Perplexity. The Amazon ruling gives you a concrete reason to ask that question now rather than after a problem surfaces.

The Strategic Picture at This Stage

Walmart’s dual commitment to authorized third-party agents (ChatGPT, Gemini via UCP) and its own native agentic infrastructure (Sparky, Marty) reflects a bet that the future of discovery is platform-authorized openness rather than either walled gardens or ungoverned scraping. Furner acknowledged on the February 19 earnings call that the interaction between agentic commerce and advertising monetization is still being worked out: “How this will work with agentic commerce, I think we’re all learning as we go.” That candor is worth taking at face value. The platform is moving quickly, but the measurement and monetization frameworks for agent-driven commerce are still in early definition.

Consumer adoption is also running ahead of consumer comfort with full automation. Walmart’s own Retail Rewired Report 2025 found that 46% of respondents were either somewhat or very unlikely to use an AI agent to handle an entire shopping trip. ChannelEngine’s Marketplace Shopping Behavior Report 2026, drawing on a survey of 4,500 shoppers across the U.S. and four European markets (ChannelEngine is a marketplace management software provider), found that only 17% feel comfortable letting AI complete a purchase outright. Sparky is handling the early stages of the journey, discovery, specification review, and occasion planning, more than autonomous checkout. That is consistent with Walmart CTO Hari Vasudev’s May 2025 description of the company’s surgical approach: agents deployed for specific tasks, with human thresholds carefully evaluated before expanding autonomous execution.

The category that closes that loop, moving from agent-assisted consideration to agent-completed purchase, is exactly where the UCP investment is pointing. When those mechanics mature, the suppliers whose item content is complete in Supplier One, whose Connect presence extends to Sparky-integrated ad formats, and whose attribution models are calibrated for multi-surface journeys will be capturing inventory that is, right now, still being defined. The Amazon injunction is a reminder of what the adversarial path produces. Walmart’s authorized architecture is designed to produce something more durable, and the suppliers building fluency with current tools, from Marty’s beta reports to Scintilla’s Shopper Behavior data, are establishing the operating competency that will matter most when the next phase of agentic commerce moves from testing into standard practice.

Winning With Walmart

Winning With Walmart is an independent platform for suppliers, sellers, solution providers, and industry experts.

Articles are developed by staff editors, contributing experts, and trusted partners.

Some are researched and drafted with the assistance of AI tools and are reviewed, fact-checked, and edited by Winning With Walmart editors before publication.

Editorial judgment, sourcing decisions, and final approval rest with the publication's human editors in every case.

We are committed to accuracy and fairness. If you believe this article contains an error, we welcome your feedback.