Are you tired of overwhelming your shoppers with endless product listings? Do you want to boost conversion rates and create a more personalized shopping experience? The answer lies in leveraging Related Prompts used as reductive questions, a powerful technique for transforming your commerce store into an intelligent shopping assistant capable of understanding your shoppers’ intent and tailoring their journey with precision.

Related prompts are synthetically-generated search suggestions that revolutionize the way shoppers find what they're looking for with more conversational, human-like, and intuitive interactions with your commerce store. The approach of GenAI for e-commerce search brings within a specific application of the related prompts innovation, which are reductive questions.

Also known as narrow or filter questions, reductive questions are strategic follow-up questions designed to refine a shopper’s search to better understand their context and act as a shopping guide for them towards the perfect product. Instead of simply displaying a massive list of results for a broad query like "haute couture dresses", the system, acting as a digital shop assistant, asks targeted questions like "Which category are you interested in?" or "Which color are you looking for?" to narrow down the options based on the shopper’s answers and your products catalog. These simple, yet strategic questions help narrow down the options and create a more personalized experience.

How reductive questions work

To picture the user flow where reductive questions are integrated, let's use an example to identify the different steps followed. Imagine that a shopper enters a query like "gym shorts". The system breaks down the context of the search into smaller pieces through entity detection (also known as query understanding). These entities are the key attributes of this initial query—for example, type (clothing), category (sportswear), fabric (cotton), sport (training), etc. Then, the system retrieves relevant reductive questions based on the current context, such as "Are you looking for sportswear for women, men, or children?" or "Do you need shorts to practice a specific sport? Tennis, football, weight lifting, etc.".

Then, the shopper answers a question, updating the context. The results that could have been presented to the shopper are grouped, and discarded or shown depending on their relevance. Finally, the system retrieves a new page of products and updated questions based on the refined context. The steps are repeated several times to fine-tune the initial query as much as possible and offer shoppers the best and most reliable search results.

To create extraordinary context and generate appropriate reductive questions, Catalog introspection is paramount. Since reductive questions run on Deepseek and operate on self-hosted GPUs, the Empathy privacy-compliant LLMs leverage chain-of-thought reasoning capabilities to classify and contextualize the initial query from your catalog. When creating reductive questions, the most important thing to avoid is narrowing the shopper’s search to a product you don’t have in stock. Knowing the products available in your catalog and their attributes enhances the shopping experience and increases conversion rates.

The power of context

The magic behind using Related Prompts as reductive questions lies in understanding the context of the shoppers' search. At Empathy, we use advanced techniques like entity detection to automatically analyze queries and extract key attributes. For example, for the query “gown for a party”, we can identify different entities such as category (dress), occasion (party), and section (woman). Combined with user data such as browsing history and stated preferences, this creates a deep understanding of shoppers’ intent that makes every interaction feel personal and meaningful.

For instance, after a shopper searches for "laptop", the system might know the category is "Electronics" and propose questions such as "What is the purpose of the laptop?" suggesting options such as "Gaming", "Work", "Personal". This isn’t guesswork, it’s intelligent reasoning powered by data from your catalog and insights into shopper behavior.

Reductive questions as a game-changer for commerce stores

Reductive questions' new approach doesn’t just make shopping more joyful and satisfactory, and surely less frustrating, it drives results. When shoppers are presented with options that truly match their desires, the likelihood of a purchase skyrockets. People are far more likely to buy when they quickly discover products they genuinely love, and that sense of discovery translates directly into higher conversion rates.

Reductive questions also transform every journey through your commerce store into a personalized adventure. Each interaction feels carefully customized, as the system understands and adapts to individual preferences and needs, offering a unique, tailored product selection.

Behind the scenes, this intelligence comes from a deep understanding of your product catalog. The system analyzes your actual inventory to ensure that every question it asks is relevant to what’s truly available. For instance, if someone is searching for a t-shirt and your store currently stocks only blue and white options, the system will suggest those colors, never leading shoppers toward dead ends or unavailable products.

As the conversation evolves, the system intuitively ranks and presents the most helpful questions first, always prioritizing those that will bring the shopper closer to their perfect match. Every interaction feels smooth, purposeful, and, above all, helpful.

Building a contextual graph: The technical backbone

At the heart of this seamless experience lies a sophisticated technical foundation: the contextual offline graph of shoppers' choices.

It all begins with a thorough entity analysis of your product catalog. The system dives deep, identifying specific fields, such as categories like "t-shirt", "dress", or "trousers", attributes like "color", "size", or "price", and even special flags such as "organic" or "vegan". By aggregating these fields, it can instantly determine, for example, exactly how many “red midi dresses” are in stock or which sizes are available for a best-selling sneaker.

Using this data, the system constructs a dynamic contextual graph, which is a living map of your catalog’s possibilities. Each node in this graph represents a specific context, such as “red dress, size L,” and keeps track of how many products fit that context, also known as count. As shoppers interact with reductive questions, their choices, called links, create new pathways through the graph. When the shopper selects “size M", the context shifts; when they choose “blue”, it narrows further. Each link between nodes reflects a shopper’s decision, weighted by how effectively the choice narrows down the results. Consequently, a path is a series of consecutive links that lead from an initial context to a leaf node.

The real genius emerges at the crossroads, which are points in the graph where multiple links originate from the same node and share the same field—multiple shopper choices branch out from a single context. These crossroads are mapped into reductive questions for the system to identify which prompts will have the greatest impact on the shopper’s journey, ensuring every prompt is both timely and meaningful.

By continually analyzing this contextual graph, the system anticipates the most valuable questions to ask next, guiding every customer toward their ideal product with empathy. This is the AI governance in retail driving a new era of GenAI conversational search.

Crafting effective reductive questions

Reductive questions must be effective to reduce shopper's frustration, enhance the shopping experience, and boost conversion rates to achieve business goals. Different methods can be used to craft reductive questions that will be shown in your commerce store:

• Suggestion text: Use of generative privacy-minded LLM to create natural-sounding suggestion text (prompts) for the questions, considering language nuances, genre, and plurals. For example, if you are looking for "Jeans," the system might ask, "What wash are you looking for?"
• Option selection: Extract predefined answers (options) directly from the target nodes of the crossroad previously mentioned.
  • Number of options: Carefully consider the number of options to present. For flag fields, a single option can activate a filter. For other attributes, aim for 2-5 options for optimal user experience, with a "show others" option for cases with more possibilities. The system should have ranking criteria to present the most relevant options.

The future is conversational

Reductive questions represent a significant step toward more conversational and intuitive online commerce experiences. By understanding shoppers' intent and guiding them with intelligent questions, commerce stores can create engaging and satisfying shopping journeys, leading to increased sales, improved shopper loyalty, and a competitive edge.

With reductive questions at the heart of your business strategy, this future is more than possible; it's immediately achievable.

So why wait? Start transforming your store into an empathetic shopping assistant now—and watch as your customers fall in love with how smart shopping can truly be!

Try it out!

Are you curious about what Related Prompts can do to elevate your commerce store's search and discovery experiences? Do not hesitate and ask for a demo today. Give your shoppers a human-like conversational interaction they would never want to end!

Keep reading!

Discover more about how we implement the latest discoveries in artificial intelligence and bring them to life in our Empathy’s Open Innovation projects.

Learn more on our Open Innovation page (opens new window)