The Science Papers Behind Sparky: How Sellers can get recomended by Walmart's AI Search

On June 6, 2025, Walmart launched Sparky, an AI shopping assistant that's now live in their mobile app. This isn't just another search improvement—Sparky represents Walmart's entry into agentic AI. In this article, we'll explain what this means for Walmart sellers, what this means for e-commerce, how technology like Sparky works, and how to optimize your listings for it.

What Is Walmart Sparky?

Walmart Sparky is an AI assistant accessible through a smiley face button in the Walmart app. Ask Sparky a question, and it will derive your intent and context to provide a carousel of items along with a description on why these items fit your specific situation.

For example, ask "what's the best laptop for an art student" and Sparky recognizes you need graphics capabilities, adequate storage, and student-friendly pricing. It can handle complex requests like "what sports teams are playing tonight and help me find the right jersey" or provide weather-based outfit recommendations—all in one conversation.

Why Walmart Built Sparky

Walmart's decision to build Sparky was a response to concrete market signals from their own customer research. Walmart's 2025 Retail Rewired Report revealed that 27% of their shoppers now trust AI recommendations more than influencer endorsements, and 47% would trust AI to purchase household essentials within a budget. The data also showed that 69% of Walmart customers consider shopping speed a key factor in retailer choice, while traditional search methods were creating friction. Additionally, external research shows: Over 58% of consumers have replaced traditional search engines with Gen AI tools for product/service recommendations. Amazon have already seen significant success with the launch of their own AI Chatbot, Rufus, which accounts for ~14% of Amazon searches.

Understanding the Science behind Walmart Sparky

Two groundbreaking research papers from Walmart's Global Technology team reveal the sophisticated systems behind their AI search capabilities. By understanding this research, sellers can optimize their product listings to be more discoverable and recommended by Sparky.

Paper 1: Large Language Models for Relevance Judgment in Product Search

The first paper, presented at SIGIR 2024, demonstrates how Walmart uses Large Language Models (LLMs) to understand the nuanced relationship between what customers search for and which products actually meet their needs. The researchers trained models on over 6 million query-item pairs to achieve near-human understanding of relevance.

Key finding: The models achieved "agreement with human evaluators between 85% and 89% of the time" when evaluating product relevance. This means Walmart's AI understands product relevance and customer intent almost on a par with human scorers.

Paper 2: Semantic Retrieval at Walmart

The second paper, from KDD 2022, reveals Walmart's semantic search architecture that goes beyond simple word matching to understand meaning and context. This system particularly excels at understanding the intent behind unusual or specific customer queries.

Key insight: "Semantic retrieval helps bridge the vocabulary gap especially for tail queries; it helps with synonyms, misspellings, and other query variants that users type."

How can Walmart Sellers optimize for Sparky?

Based on insights from these research papers, here are six ways to align your listings with how Walmart's AI understands customer intent:

1. Align with Customer Problem-Solving Intent

The papers note: "E-commerce queries often have multiple interpretations. For example, the query 'apple' could refer to the fruit or to the electronics brand."

Strategy:

• Anticipate the various problems your product solves

• Address different customer intents in your descriptions

• Think about the "job to be done" rather than just features

• Consider seasonal or situational uses

2. Understand Your Audience's Language

The semantic retrieval paper states: "Semantic retrieval also helps to better understand the semantics of longer queries that might contain a nuanced intent from the user."

Strategy:

• Study how your target customers describe their needs

• Include the emotional and functional language they use

• Address the underlying concerns behind searches

• Think about the journey that brings customers to your product

3. Optimize for Specific, Contextual Queries

The research found: "The hybrid system used in production at Walmart.com... demonstrates the benefit of such a system for tail queries" - those specific, detailed searches that reveal clear intent.

Strategy:

• Create content for specific use scenarios

• Address niche applications thoroughly

• Help the AI understand when your product is the perfect match

• Think about the "long tail" of specific customer needs

4. Create Context Through Product Attributes

The research reveals: "We experimented with different number of attributes... When adding the color red to a product, the input looks like '[title tokens] [color token] red'. This technique allows the model to determine which attributes have been concatenated."

Strategy: Don't just list attributes - create meaningful context by:

• Showing how attributes relate to use cases

• Connecting features to customer needs

• Building a complete picture of your product's purpose

• Helping the AI understand not just what your product is, but what it's for

5. Utilize Azoma, to get your brand mentioned by Walmart Sparky

Utilize Azoma in order to research how customers are talking to Sparky, and publish optimized content in a few clicks. Book a demo.

The Semantic, AI Based Future of E-commerce Serach

Walmart's research shows that Sparky represents a fundamental shift from keyword matching to true semantic understanding. As noted in the papers: "LLMs are prime candidates for relevance classifiers: they are capable of teasing out long-memory dependencies and contextual relations."

For sellers, this means success comes not from keyword stuffing, but from creating listings that genuinely communicate:

• Who your product is for

• What problems it solves

• When and how it should be used

• Why it's the right choice

The research concludes that models "can perform on par, at 89% parity" with human evaluators. This near-human understanding means sellers must think like humans - focusing on meaning, context, and genuine value rather than algorithmic optimization.

As the 'LLMs for Relavance Judgement in Product Search;' states in it's opening sentence emember: "High relevance of retrieved and re-ranked items to the search query is the cornerstone of successful product search." In the age of AI search, relevance means truly understanding and serving customer intent.