From "Search" to "Answer"—The Future of Next-Generation Online Search Envisioned by Perplexity

Online search has continued to evolve over many years, but with the recent spread of generative AI and Large Language Models (LLMs), the relationship between search engines and users has begun to change significantly. Now that it is becoming standard to not just "search" for information but to obtain "answers" directly, a new concept called the "Answer Engine" is attracting attention.

In this article, we will explore the current state and challenges of online search, as well as the future potential of answer engines, focusing on statements by Johnny Ho, co-founder and Chief Strategy Officer of the answer engine "Perplexity." Since high school, Johnny has competed in the International Olympiad in Informatics (IOI), a global competitive programming tournament, and has been active as a world-class engineer. How is Perplexity trying to redefine the user experience in a highly competitive search field? We will delve into its strategy, technology, and business model.

1. What is an Answer Engine?

1-1. From Traditional "Search" to "Answers"

Traditional search engines were built on the premise that users would click to read details after being presented with a list of relevant web pages based on keywords they entered. On the other hand, answer engines like Perplexity are characterized by using AI to synthesize multiple information sources and present the answers users want directly.

For example, emphasis is placed on "being able to respond to complex inquiries" and "being able to handle multi-step questions," proposing a new approach of "interpreting user input to generate the optimal solution." In fact, Johnny explains it as follows:

1-2. The "80/20" Challenge and Opportunity

What he emphasizes is the "80/20 rule." While 80% of search queries can be satisfied with simple questions, the remaining 20% of complex problems cause significant stress for users. Answer engines aim to solve these 20% of difficult problems while simultaneously providing users with additional information obtained during the resolution process, attempting to create comprehensive value that includes "the process to reach the answer and additional actions."

2. Perplexity's Technical Approach

2-1. Fusion of Traditional Search Technology and LLMs

Perplexity's internal structure is broadly divided into two stages. First, traditional search technology that determines "which information sources should be retrieved." Second, a Large Language Model (LLM) for "integrating and appropriately summarizing the retrieved information fragments."

Through this, it extracts necessary snippets from numerous pages on the web, and the LLM summarizes and organizes those snippets to return them. According to Johnny, Perplexity also utilizes a proprietary model called "Sonar," and is optimizing it to provide accurate and concise answers in a short amount of time.

2-2. Mutual Complementarity with Cutting-Edge Models

As models with higher reasoning capabilities, such as GPT-4 or GPT-3.5 (for example, models adept at continuous thinking like those called "GPT-4 01"), emerge, the role of the answer engine side may change.

However, Johnny raises the concern that "the longer a model maintains a chain-of-thought, the higher the computational cost and response time become." Depending on the user's intended use, it is necessary to distinguish between cases that should be left to high-precision LLMs and cases where simple models are sufficient.

3. Business Model and Monetization

3-1. From Subscription to Advertising

At this stage, Perplexity mainly earns revenue through a subscription model. It is a mechanism where a portion of the added value created by streamlining users' work and learning is returned in the form of a monthly fee. However, they are also considering an advertising model for the future.

True to these words, while Perplexity provides the answers users are looking for via the shortest route, it is currently in the stage of exploring 'whether we can present reliable sponsors and recommendations in a way that does not compromise the user experience.'

3-2. Giving back to publishers and content providers

With an answer engine, users do not necessarily visit the linked sites. Therefore, Perplexity has started a mechanism called the 'Publisher Program.' This distributes revenue to the sources used to generate answers based on factors such as the number of citations.

In this way, even in an era where AI summarization is becoming commonplace, they are attempting to create a model where those who produce high-quality content can be appropriately rewarded.

4. Growth Strategy and Organizational Culture

4-1. Accumulating small, Scrum-like experiments

Although Perplexity does not have the large-scale resources of a major corporation, it uses rapid iteration as a weapon to expand its features. They thoroughly follow an approach of 'creating prototypes quickly from scratch and learning even if they fail,' and it is said that each engineer sets short-term goals independently and proceeds with development while measuring progress on a weekly basis.

To borrow Johnny's words, 'Deliver value to users through small experiments, and apply the feedback gained from that to the next step.' This culture is what directly leads to their fast release pace.

4-2. User community as a growth driver

Perplexity is also focusing on collaboration with its user community. For example, in the 'Back to School' campaign, they expanded learning support features and related content for students, and actively incorporated a 'mechanism where the student community competes to share information.' Such user-participatory events generate word-of-mouth without spending money on advertising, contributing to increased service awareness and improved retention rates.

5. Future Outlook and the Future of Online Search

5-1. Evolution of multimodality and interaction

Answer engines are expected to expand into 'multimodality,' handling multiple media such as images, audio, and video, rather than just text summarization. Once voice input and voice response are implemented like in ChatGPT, humans will be able to interact with AI in a more natural way. However, Johnny points out that new issues will arise in terms of UI, such as the difficulty of verifying citation sources when using only voice.

5-2. 'Reliability' is the key to the new search engine competition

There is a vast amount of content on the web, and the amount of AI-generated text among it is increasing. While answer engines integrate a massive amount of information, if 'misinformation' or 'low-quality information' is mixed in, it will mislead users.

Therefore, 'how to guarantee the reliability of the source of information' will become a new competitive factor for search providers. Perplexity's Publisher Program can be said to be one measure to continue presenting high-quality content.

The shift from 'search' to 'answers' is not just a change in UI or UX. It also affects the revenue structure of publishers and content creators, and further requires new culture and literacy for users to obtain information efficiently.

The vision described by Johnny Ho is a 'platform where AI and users interactively shape knowledge.' In this space, users can access the necessary supporting information even for complex questions, and mechanisms are provided to encourage further action from the user. As a pioneer showing this path, Perplexity is making small updates every day, aiming for the optimal solution to derive answers.

This vision not only significantly changes the way we obtain information online but also presents a new symbiotic model with those who create content. Let us continue to watch the dawn of this new era where search transforms into 'answers'.

Related Articles