Since the release of ChatGPT, the internet seems to have hit the fast-forward button.
Sora pushing the boundaries between AI and reality, Claude3 approaching a genius IQ of 145—new "first-of-its-kind" models keep emerging, followed by claims of certain industries being "disrupted."
"Epoch-making" news fills the front pages every day. However, the foundational infrastructure connecting us to the internet—search engines—seemingly hasn't undergone any noticeable changes.
After a long evolution, search engines gradually embraced an encyclopedic approach, aiming to encompass all information on the internet. Yet, the big and comprehensive search engines struggle to meet our increasingly refined information needs.
In fact, since the first day ChatGPT was released, some predicted that AI would replace search engines. However, this "disruption" has yet to happen, and people still find themselves stuck with ad-filled search engines. Will there be AI-powered search engines in the era of AI?
Should we search on our own or ask AI?
Actually, many users have already chosen tools like ChatGPT and Gemini, replacing traditional search engines. When ChatGPT went global offline last March, many users commented, "I check every ten minutes, after all, ChatGPT being offline for us is as serious as Google going down."
Though both aim to meet our information needs, their principles differ. Traditional search engines filter through massive information based on keywords users input in the search bar, presenting numerous web links as results.
Conversely, generative AI operates through ongoing conversation, utilizing natural language processing to retrieve relevant information from its corpus in response to user queries. More importantly, generative AI can upload datasets, documents, and other materials to analyze and provide more accurate information—a capability traditional search engines lack.
Thus, rather than presenting a long list of blue links like traditional search engines, generative AI engages users in multi-turn conversations, allowing them to refine their queries based on the answers provided. The ability to connect contextually sets generative AI apart, and this is why many users are migrating from search engines to generative AI.
A research survey confirms this shift, employing the PPM (push-pull-mooring) model to investigate factors influencing user migration. Three categories of factors were identified:
The study indicates that perceived interactivity is a crucial factor leading to users' intention to switch, mainly due to the effectiveness of multi-turn conversations. For users, multi-turn conversational search can significantly shorten the path of information retrieval and reduce its complexity.Many traditional search engines have recognized this and are beginning to reform the original "one question, one answer" search paradigm, attempting to replace the traditional search box with AI-driven chat interfaces.
The search market is seeing a new trend.
In this AI-themed revolution of search engines, there are primarily two different factions.
First, traditional search engines are rejuvenating themselves with the help of AI cores. The most notable example is Microsoft, the largest shareholder of OpenAI. This search giant has launched Copilot, powered by ChatGPT, incorporating an AI chat interface within the search engine.
After a user poses a question, Copilot can aggregate information and annotate web page sources. Moreover, Copilot can also summarize and generalize the information from all web pages a user is viewing, and it supports uploading files. Domestic search giants are also actively entering the fray. Baidu has been focusing on AI since 2010, with features like Baidu Image Recognition and Wenxin Yiyan prominently featured on Baidu's homepage.
Next up, let's talk about the new forces in the search engine industry. These newcomers are coming on strong, each aiming to revolutionize traditional search engines. Take Arc search, for instance, renowned throughout Silicon Valley as "the search engine most likely to replace Google." Numerous tech bloggers have voluntarily endorsed it.
Then there's Perplexity AI, which has already made significant inroads into the existing search market. This search engine, only a bit over a year old, boasts a monthly active user count of 100 million. Even Nvidia's CEO, Jensen Huang, has given it his seal of approval, saying he uses it "almost every day."
These "newcomers" in the search engine market generally follow an "Answer First" principle, prioritizing text over web page links to respond to user queries. In their responses, they include small citation markers to indicate sources. Users can click on these to be directed to the relevant web pages.
This approach allows users to focus more on the information content than on web page links during searches. Moreover, these AI search engines strive to ensure the continuity of users' information searches to the greatest extent possible. For instance, in Perplexity AI's design logic, the UI display for "continue asking" takes precedence over that for "new questions."
The emerging players have garnered substantial support from the capital market. Arc Search completed two consecutive funding rounds of $5 million and $12 million in 2020 and 2021, respectively. Meanwhile, Perplexity AI recently closed a new funding round at $73.6 million in early 2024, raising its valuation to $52 billion.
Whether they are newcomers or traditional giants, both are heating up in the capital market and advertising, but user adoption seems to be rather slow. It's been almost a year and a half since the release of ChatGPT, yet the dominance of search boxes over dialogue boxes in the search engine market remains prominent. Why is this the case?
To answer this question, we need to understand how search engines became the cornerstone of the internet.
In the initial stages of the internet, rapid advancements in network technology led to a significant increase in information. People gradually developed a need for "search," and search engines emerged.
Early players like Yahoo designed their products based on a "directory" approach, relying on manual classification and input of webpage addresses. Limited content hindered their growth, destined to be replaced by evolving technologies.
In 1998, Google's crawler technology marked a turning point for search engines, shifting towards algorithmic recommendations and personalized search results.
Simultaneously, the market started consolidating. Small search engines were marginalized or even acquired. Large search engines accumulated market resources and development factors, and the flywheel effect began to manifest, accelerating their growth. In 2011, Baidu reached a daily search volume of 100 million, paving the way to becoming the leader in the Chinese search market.
In the growth of every search engine, three essential elements are indispensable - users, businesses, and search functionality. In the early stages of the flywheel effect, substantial effort is required to push the stagnant wheel, circling and pushing until the flywheel spins faster and faster, reaching a critical point. This triad is why a search engine can dominate the internet market.
User, the "lifeblood" of search engines, is crucial for their existence. Search engines need to design products and plan operations based on user behavior. Additionally, a vast user base attracts collaborations with advertisers.
Advertisers form the foundation of the commercial aspect of search engines. Early search engines relied on directory inclusion fees, but Google introduced the pay-per-click advertising model—displaying ads on search pages, with the ad price determined by the number of user clicks.
However, the biggest challenge for search engines in the AI era is commercialization. Traditional monetization models like pay-per-click and bid ranking cannot be directly applied to generative AI.
AI search without ads
AI differs from previous technologies with its black-box nature. The internal mechanisms and decision-making processes of AI systems are opaque. For most open-ended user queries, AI responses are randomly generated, resulting in different answers each time.
Even backend operators cannot have complete control over AI responses. Yet, commercial ads demand high click-through rates and conversions, requiring precise analysis of user characteristics to find suitable ad content and formats.
Due to the black-box nature of AI systems, AI search engines pose an unstable and insecure marketing environment for advertisers.
On the user side, learning to use AI is a process. Just as not everyone had the opportunity to drive when cars were first invented, users need time to learn and adapt to AI, which operates differently from traditional search engines. Using specific prompt language is crucial for generating accurate answers, requiring users to articulate their questions clearly.
These challenges are inevitable for AI-era search engines. However, this time, tech companies are not confined to the term "search engine." They are looking towards the future, as seen with Microsoft's Copilot, integrated not just within search engines but across the entire Windows system, advancing towards AI intelligence.
This may indeed be the new form of search engines in the AI era. In the near future, we might exclaim, "The future is already here."
