Skip to main content

Blog Post

Retrieval and Search

The Great Search Disruption

Rethinking the human-computer interface in the era of GenAI

For decades, we’ve searched for digital information under a simple premise: you type in a search query and you get back a list of links that contain the words you used. This interface was a simple and effective way to find websites and other resources that contained the information you sought.

But how many links do you have to click and how many websites do you have to scan before you find the answer you’re looking for? Somehow, Google turned search into… research. It was time-consuming and sometimes frustrating, but it beat going to the library so we got used to it.

Now all of that is changing.

ChatGPT has opened our eyes to what’s possible when software is capable of understanding and answering questions. Microsoft, Google, and others recognize that this interaction paradigm is the successor to legacy search, and they are trying to capitalize on it.

Any organization that’s in the business of providing answers to customers, internal users, or anyone with a question can do the same by using Vectara’s Grounded Generation, which we announced last week. Your users ask a question, and Vectara will provide an accurate, dependable answer based on your own data. It provides the API backbone to your conversational search experience.

Today, we’re releasing the UI counterpart to this API. The open-source vectara-answer project is our laboratory for experimenting with the new UX paradigm of searching for answers, not just links. With this project we’ll build and share sample applications that you can play with, build upon, copy, and reference as you integrate conversational search into your own products.

Those ten blue links have taken us far, and now there’s a new way to get what we really care about when we search: answers. With your help and contribution, we’ll evolve our understanding of what the future of search looks like and share it in vectara-answer.

AskNews: News source answer aggregator

To demonstrate what we’ve learned so far, let’s look at the AskNews sample application.

Figure 1: AskNews initial state

AskNews enables the user to ask a question about recent events. It will retrieve the most relevant results from BBC, NPR, FOX, CNBC, and CNN news articles, and summarize them into an answer, backed up by citations.

Figure 2: AskNews answer and results state

We built the AskNews application using our vectara-ingest project to crawl news sources and index them into Vectara corpora, and we used vectara-answer to provide the front-end you interact with. This UI demonstrates a few of our current beliefs about what makes a conversational search experience effective: comprehension, explainability, and question recall


Legacy search often becomes a game of “guess the magic word.” If you try enough combinations and variations of keywords, eventually you’ll find one which yields some helpful links. Conversational search takes this responsibility off the user by taking into account their question’s underlying meaning – not just the specific words they used.

The idea is to enable the user to interact with the software as naturally as possible, even if that means asking a question in one’s native language. That’s why AskNews will provide the same response to someone asking “Should AI be regulated?” in English or “人工智能应该被监管吗?”, which is the same question in Simplified Chinese.

Figure 3: AskNews answering a question in Simplified Chinese


Sometimes we learn the most from the answers we don’t expect. But those are also the moments when fact-checks are most vital.

When a user submits a question, AskNews uses Vectara’s Query API to retrieve the most relevant articles, which Vectara stores as documents in a vector database. It treats these results as the source of truth when answering the user’s question. AskNews then naturally enables the user to validate its answers by cross-referencing the results it used to generate its answer. 

This way you can find an explanation for any part of an answer you’re not sure about. You can even click on a citation to jump directly to the highlighted reference. If that’s not enough, you can click the result link to go directly to its source.

Figure 4: Highlighted citation in the AskNews user interface

Question recall

Here’s the thing about the news: there’s always something new. So you might ask a question on Sunday night, and then get a different answer when you ask the same question on Monday morning.

For a user who depends on answers that change over time, recalling the specific answer to a question is less valuable than being able to ask that question again and just get an up-to-date answer. This is why AskNews securely stores the questions you ask in local storage on the user’s computer. You can see the questions you’ve asked and re-ask any questions that might have new answers since the last time you asked them.

Figure 5: The history of the questions a user has asked AskNews

Explore conversational search UX with vectara-answer

Like any new technology, LLMs require new ways of thinking about the human-computer interface for us to fully utilize their potential. We expect to discover new methods of interaction that will be significant departures from what legacy search has taught us to expect.

The vectara-answer project is our first step in this journey. We’re thrilled to invite the community to use this project to build modern conversational search applications, use it to deeply understand how users want to interact with information, and continuously improve their users’ search experience.

To get started with vectara-answer see the Quickstart guide.

Recommended Content


Vectara: Hybrid Search and Beyond [PDF]

In the AI era, how people interact with information has changed. Users expect relevant answers to questions in natural language, not a shopping list of hit or miss search results. They expect the best semantic or exact matches regardless of typos, colloquialisms, or context. Additionally, it is Vectara's mission to remove language as a barrier by allowing cross-language hybrid search that delivers summarized answers in the language of your choice. The Internet, mobile, and AI have made information accessible, now Vectara helps you find meaning quickly through the most relevant answers. Get to know Vectara, and if you have a question, just ask.

Get Introduced to Vectara
Resource Image
Close Menu