Once upon a time, traditional keyword search was a technological breakthrough. It was the primary way people found information online, thanks largely to the way Google improved relevancy. This type of search remained consistent for two decades. But over the past several years, the rise of GenAI has caused a shift in how users search and how shoppers discover products.
It’s AI and GenAI technology that’s changing the online shopping experience — specifically via chatbots. Yet, simply replacing search with chat doesn’t solve the problem of helping shoppers find what they want quickly. The search box is just too ingrained in how people navigate digital spaces.
Early experiments, including high-profile launches like Amazon’s Rufus, have shown that shoppers don’t naturally gravitate toward chat interfaces for product discovery. Shopping chatbots often create friction, forcing users to wade through lengthy, jargon-filled responses.
The real promise of AI is it’s potential to transform the search box into the “intent box,” a new manifestation of intelligent search which we’ve summarized in five important trends that are transforming how people interact with ecommerce websites and the search box itself.
Relevant reading: From Search to Intent Box: The Future of Ecommerce Product Discovery
Trend One: From Keywords to Intent
Search results were once entirely focused on the keyword or phrase a user typed (or spoke) into the search box. But keywords alone don’t communicate intent, so the results can be imprecise, broad, or completely blank (e.g., a “zero results” search). Things can really go awry when a shopper isn’t sure what they want.
Imagine a shopper visits an online furniture store and types in, “comfortable sofa for small apartment.” With traditional keyword search, the following issues tend to occur:
- Results are imprecise or too broad: The search engine might return hundreds of sofas, but many are too large, or the “comfort” aspect isn’t easily filtered. The user must scroll through pages of results, clicking on sofas to check dimensions and read reviews for comfort. This gets frustrating quickly and can also result in choice paralysis, killing the potential sale.
- The results are nonexistent (zero results): If the site’s product descriptions don’t explicitly use the exact phrase “comfortable sofa for small apartment,” or if “small apartment” isn’t a defined attribute, the user might get a “no results found” message, even if the store does have suitable options
In the above scenarios, the search engine doesn’t understand the intent behind “small apartment” (e.g., looking for compact, loveseat, apartment-sized) or “comfortable” (e.g., plush, deep-seated, soft fabric).
Search engines need a better way to understand intent, particularly since the way shoppers search is different now than it used to be. It’s much more common for people to use natural language and ask questions or describe the qualities of the product they want instead of typing in an exact term or brand name. A primitive search experience — one that relies on keywords to match the query to the product —isn’t up to the task of helping shoppers find what they need.
Trend Two: RAG Is the New GenAI Standard
Raw GenAI is powerful, but it’s also unpredictable. Anyone who’s seen a chatbot confidently invent facts knows the risks. These include AI hallucinations (when your bot friend outright makes stuff up), outdated information, and other inaccuracies.
Retrieval-Augmented Generation (RAG) is important when using AI in an enterprise setting because it prevents unpredictable and inaccurate GenAI responses. RAG works by grounding AI in real, up-to-date information from your own trusted sources like product catalogs and knowledge bases. It moves the LLM model away from a reliance on training data and retrieves relevant content in real time, then uses that to generate an answer. This is what allows enterprise AI systems to deliver responses that are accurate and current.
Coveo’s unique approach to RAG retrieves only the most relevant content using a four-step process:
When grounding is missing, chatbots like Amazon’s Rufus may produce less-than ideal results (e.g., recommending Pepsi as a healthier alternative to Coke). This answer is obviously incorrect. It’s also potentially damaging to users’ trust in Amazon. In contrast, when Dell implemented grounded GenAI, the system only recommended products and advice that matched Dell’s actual catalog and brand priorities. It gave shoppers relevant, on-brand answers every time.
Coveo’s RAG approach combines advanced retrieval techniques that include vector search, semantic matching, and keyword filters to pull the most relevant information for each query. The AI then generates a response using only this verified data, with clear citations so users can check the source for themselves.
This is how an intent box gives enterprise retailers the functionality and usability of GenAI, without the chaos. You get the creativity and fluency of large language models, but every answer is grounded in your own data, policies, and catalog. This grounding is key because it translates to a much better search experience for shoppers without the risk that things will go awry.
Trend Three: Goodbye Chatbots, Hello Embedded Intelligence
Most shoppers don’t want to strike up a conversation with a robot in the corner of the screen. We just want to get to the product, not get stuck in endless back-and-forth with Chirpy the overly upbeat shopperbot. When chatbots are bolted onto every page, they tend to distract more than they help.
The intent box is designed to keep shoppers in the flow of the buying journey rather than sidelining that journey altogether. Unlike a separate chat window, it’s integrated with the search box. This amounts to a smarter, more relevant user experience, one that feels natural. Shoppers get:
- Relevant answers right in the search box, not a separate chat
- Guidance and education when needed, without leaving the page
- A seamless path from question to product, with no unnecessary detours
An intent box takes search seriously without replacing the search experience or navigation. It’s embedded into search, making AI technology part of the search process. Chat-first assistants tend to underperform unless they’re placed in what’s essentially a “common sense place” like a product detail page, where a shopper might genuinely need extra information or guidance. Forcing a conversation elsewhere just slows down discovery and adds friction to the conversion path.
Trend Four: Dual-Speed Discovery for a Multimodal Shopper
Not every shopper is looking for the same thing or shopping at the same pace. Some know exactly what they want and expect instant results. Others are still exploring, comparing, considering, and seeking advice. The point of the intent box is to meet every shoppers’ needs, in real time, with a single, adaptive experience.
Grocery shopping is the perfect use case for this level of adaptability. Here are two scenarios representing different shopper intent:
- The knows-what-she wants shopper: Clarice enters a clear, transactional query, “barbecue sauce” and the system uses semantic search, lexical matching, and intent detection to return mostly barbecue sauce. In this case, Clarice gets in and out fast, with personalized results that she can refine to the exact product she wants via filters and sorting options.
- The he’s-not-sure-and-needs-guidance shopper: Bill’s query is more open-ended and exploratory. He types, “I host a summer barbecue in a park, and I need to cook for six people. What are your recommendations?” The intent box shifts gears, providing a richer, GenAI-powered response. It goes beyond a list of products, to incorporate recipes which may appear alongside relevant products. In this case, the system detects that Bill isn’t ready to convert, so the answer can be more nuanced and even take a bit more time to render. The tradeoff is that it will be much richer. It will guide Bill to the ingredients and recipes he needs.
This dynamic approach means the intent box can be both fast and relevant. It’s a guided experience either way, providing fast answers (and the right products) or a more nuanced journey. The same search box adapts to Clarice or Bill’s intent and context. It’s one search box that supports infinite contexts.
Trend Five: From Utility to Revenue Driver
Search should be considered a strategic growth lever, particularly for ecommerce websites. Amazon’s Rufus made headlines for generating $700 million in revenue, but most of that gain is ad-based—not conversion-based meaning the revenue came from sponsored placements vs. actually helping shoppers find and buy what they want.
Brands using the intent box model for search see higher relevance, richer experiences, and a faster path to purchase because relevant search experiences equal good business. This directly translates to revenue by:
- Connecting shoppers to products quickly: An intent box interprets queries quickly by producing the right items quickly, particularly helpful for shoppers who know what they want.
- Inspiring bigger baskets: Personalized and contextually rich suggestions are part of the intent box process. They’re meaningful in the moment, inspiring shoppers to add more to the cart.
- Providing real sales guidance: An intent box combines the best of search with the best of chatbot interaction and guidance, answering natural language questions and providing contextually aware suggestions the way a good sales assistant does.
- Facilitating a great experience: All of the above work together to make shopping easier for customers like Clarice and Bill. That drives loyalty in the form of repeat business, positive reviews, and peer recommendations.
By interpreting queries accurately and providing tailored and relevant results, an intent box creates better business outcomes. It’s supremely functional, transcending the novelty of chatbots like Rufus with a grounded and AI-driven approach to product discovery. This is how you turn a simple utility into a revenue driver.
Search Isn’t Going Away — It’s Evolving
The intent box shifts the focus from endless conversation to actual conversion. It’s a tangible example of how AI can move beyond novelty and deliver real business outcomes. Like search, success for the intent box is measured by sales metrics. It’s built to help shoppers discover and buy what they need, quickly and confidently.
This approach works because it brings together multiple AI techniques — semantics, similarity matching, and named entity recognition — to understand shopper context and deliver the right results in real time. Great technology should work behind the scenes, helping users without getting in their way, and that’s exactly what the intent box does.
For enterprise brands, the intent box matters because it grounds generative experiences in trusted data, respects merchandising strategies, and adapts to each shopper’s journey.
Product discovery has been moving toward intent-driven search for years. The intent box works alongside the traditional search box, layering in AI and GenAI so that online experiences feel much more like offline ones. That is, easy, relevant, and shaped by what each shopper needs in the moment. This is search, evolved. Retailers who embrace the intent box are setting the stage for meaningful digital transformation.