Full-Text Search Still Works. It Just Doesn’t Get You to an Answer
Originally posted by Klim Todrik on https://manticoresearch.com/blog/conversational-search/ on Aug 20, 2026 Imagine a typical online shoe store. A shopper opens search and types: I need black waterproof running shoes for daily runs on wet pavement. What would you recommend? A few years ago, almost no one expected this from a search box. The query would have been shortened to something like: black…
A shopper visits an online shoe store and types: "I need black waterproof running shoes for daily runs on wet pavement." This seemingly simple query contains multiple tasks for a search system. It must understand the color, waterproofing, activity type, and other criteria.
Traditional full-text search can handle exact terms like brand names, SKUs, and keywords. However, modern users expect more from search systems. For instance, Google reported in May 2026 that its AI Mode had more than a billion monthly users, indicating a shift towards longer, more complex queries.
In an online store setting, a query like "I need black waterproof running shoes for daily runs on wet pavement" may look like one sentence, but it involves several tasks. The system needs to extract constraints such as color and waterproofing, understand the context (running shoes), consider additional factors like price and availability, identify suitable products, and explain differences if the user asks which models are better.
Manticore's Conversational Search approach addresses these multiple tasks by combining full-text search, vector search, filters, and a language model. For example, a simple query like "Nike Pegasus 41 black" can be handled directly by full-text search. However, for queries like "light shoes for long summer walks" that contain some specific terms but also broad intents, vector search becomes useful.
After retrieving relevant products, a language model generates an answer based on the context provided. The interface then shows the products themselves. For instance, when a user types "I need black waterproof running shoes for jogging," the system first finds suitable products and then a language model generates an answer using those products as context. If the user then asks "Which of these are better for daily use?", the system can continue the conversation by using the previous results as context.
Manticore built a demo called Manticore Apparel Shop using a dataset called ConvApparel. This dataset contained 82,524 products across various categories, with each product having a description, category, images, and attributes. The system allows users to interact with it using Conversational Search commands. It combines the strengths of various search techniques to provide a more effective and conversational shopping experience.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
