news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

·Vertical search with content has a bigger market than general search tools. "Even with Google search and other search engines now, vertical search engines are still alive and well."

OpenAI's entry into the AI ​​search war has made the already crowded search track even more lively.

Overseas, emerging AI search companies such as Microsoft Bing, Google AI Overviews, and Perplexity have increased their investment in AI search. In China, Mita Technology and listed company Kunlun Wanwei (300418) have entered the search market.

But in the eyes of industry insiders, it is difficult to change the search landscape. According to a research report by CITIC Securities, the core competitiveness of search engines covers user usage habits, the breadth of front-end index content, and back-end advertising systems. AI search has obvious advantages in answering complex questions, but in the short term, there are still obvious defects such as high call costs and insufficient index content. The impact of OpenAI's SearchGPT on the market landscape is relatively limited, but it will most likely accelerate the evolution of traditional search engines, forming a search product form that is mainly based on traditional search and supplemented by GPT models in the medium term, and will correspondingly affect the cost structure and commercial realization model of the search industry.

"The era of using technology to make search tools has passed. It is difficult to use AI to transform an industry that is already in the sunset." An Internet company manager told The Paper that users may gradually abandon the use of search tools to search for information and may shift to content platforms for search. For this reason, vertical search with content will have a larger market than general search tools in the future. "We still need to pay attention to user needs and not search for the sake of searching. General search now still uses public content, but platform content like Douyin and Xiaohongshu cannot be searched from outside."

AI disrupts traditional search

It is understood that OpenAI's recently launched SearchGPT was developed in collaboration with several news publishers and is powered by the GPT-4 series of models. It can aggregate online information, including news, and allow users to ask follow-up questions, just like using ChatGPT currently. The source link is placed in brackets at the end of each answer, and the sidebar displays more results and contains the source of relevant information. The product was only available to 10,000 test users when it was released.

Long before OpenAI, Perplexity, an AI search startup backed by Amazon founder Bezos and chip giant Nvidia, had already launched a similar product. Perplexity's AI chatbot can summarize search results, list references to answers, and help users optimize queries to get the best answers. As a star AI search company, Perplexity has won fans like Nvidia CEO Jensen Huang, who once said he used the product "almost every day." In April of this year, Perplexity, a team of more than 50 people, received $63 million in financing, and its valuation doubled to $1 billion in three months.

Kingsley Crane, an analyst at financial services firm Canaccord Genuity, said the AI ​​search tools launched by OpenAI and Perplexity put pressure on Google to do better on its home turf.

The release of ChatGPT in 2022 caught Google off guard. Chatbots can give quick and complete answers to questions, which may make traditional search engine link lists and the ads that accompany these links redundant. In May of this year, Google launched AI Overviews, which can solve more complex problems at one time instead of breaking them down into multiple searches. "Sometimes you want a quick answer, but you don't have time to piece together all the information you need, and AI Overviews can do it for you." Google expects that more than 1 billion people will use AI Overviews by the end of this year.

Prior to this, Microsoft also added generative search to its Bing, which is similar to Google's AI Overviews, which builds summaries based on search queries rather than just a simple list of results.

In China, startup Mita Technology launched an AI search that provides "no ads, direct results", providing direct and accurate search answers, including reference sources, automatically generating outlines, mind maps, related events and people, and screening useful information from hundreds of millions of documents. Kunlun Wanwei launched Tiangong AI search, which can provide multi-modal search experience such as text content, graphic interweaving and chart generation. It extends related issues around a simple instruction of the user, automatically generates research outlines, practice summaries, and mind maps, helping users grasp the core content and complete their complex research needs.

According to a research report by CITIC Securities, AI search accounts for 24.2% of AI product visits in March 2024, making it the second largest usage scenario after large models. Microsoft Bing, Perplexity and other products have made leading progress in implementation. However, despite the large number of AI search competitors, the number of users is still an order of magnitude lower than that of Google. The emergence of SearchGPT is expected to have limited impact on the search industry in the short term, and attention should be paid to possible changes in the product form and business model of the industry in the medium term.

Exploring commercialization paths

"The search engines made by traditional giants, especially those in the Chinese context, have not done a good job of providing everyone with effective information. Competitive rankings make it difficult for people to find effective information, and artificial intelligence technology may indeed bring everyone a new search experience. Everyone is thinking about what the next step is to increase the penetration rate of AI, and everyone thinks of AI search." Wang Yiwei, a lawyer by training and the chief operating officer of Shanghai Mita Network Technology Co., Ltd., previously told The Paper that the benefit of AI search is "first of all, there are no advertisements," "redundant information is gone, and the big model will solve the problem of inaccurate information and poor relevance because it can understand questions more accurately."

But how can AI search without advertising support enterprises to go further? Take Perplexity as an example. Perplexity's services include free and paid versions. According to Bloomberg, Perplexity's annual recurring revenue is $20 million. Perplexity tries to increase revenue by selling artificial intelligence services to enterprises, such as launching an enterprise version of chatbots for $40 per month and adding features such as security and data protection measures. In order to expand its user base, it has signed distribution partnerships with two major operators, Japan's SoftBank and Deutsche Telekom AG, to promote AI search services to more than 300 million users worldwide.

But Google is still retaining advertising services. In May, Google said that the links included in AI Overviews received more clicks than traditional web page listings. Ads will continue to appear in dedicated locations throughout the page with labels that distinguish between organic and sponsored results.

In April, the Financial Times reported that Google said it was "not committed to or considering" an ad-free search experience, but would "continue to build new premium features and services to enhance Google's subscription services." Google's search business is considering charging for new advanced features driven by generative artificial intelligence. For many years, Google has provided free consumer services funded entirely by advertising, and the proposed reform of its search engine would mark the first time that Google has placed all of its core products behind a "paywall."

Because generative AI requires more computing resources, this type of search is more expensive than traditional search. "The computational cost of AI search is higher than Google's traditional search process. Therefore, by charging for AI search, Google will seek to at least recover these costs," said Heather Dawe, chief data scientist at UST, a digital transformation consulting firm.

Wang Yiwei said that AI search is a red ocean business, and the overall domestic environment is more urgent. Chinese companies usually consider commercialization earlier. However, if Perplexity charges a fee, it cannot become a search engine with a wider range of application scenarios. "Perplexity started to consider commercialization after reaching one million daily active users, so we will not consider charging for the moment. Next, we may try to cooperate with the B-side." As for the product, "Let's not talk about commercialization, marketization first, so that more people can use it is also a core."

Kunlun Wanwei told The Paper that the big model can better understand user intentions and preferences and provide highly personalized search results. This personalized experience can enhance user stickiness and create more opportunities for cross-selling and value-added services for companies. With the big model's deep insight into user behavior, companies can achieve more accurate advertising positioning and recommendation systems, and improve advertising conversion rates.

Search giants' position is hard to shake

AI has disrupted traditional search and supplemented the existing search ecosystem, and the catfish effect in the online search market continues to emerge. However, it is not easy for Microsoft, Perplexity or OpenAI to shake Google's current search position.

Microsoft's search engine market share did not increase after adding artificial intelligence technology earlier. According to the search engine market share data of Statcounter, a US website traffic monitoring agency, Google's market share is 91.05%, while Bing's market share is 3.74%. Google's market share has not been significantly affected.

On the other hand, in the second quarter of this year, Google's advertising business grew from $58.143 billion last year to $64.616 billion, and the revenue of its largest business unit, Google Search, increased to $48.509 billion, a year-on-year increase of 13.80%. Bing still has a long way to go to compete with Google's dominance.

Perplexity was founded by former OpenAI employees and is much smaller and less funded than OpenAI, let alone Google. Perplexity co-founder and CEO Aravind Srinivas once said his company is competing with "big sharks."

But Srinivas also said that Perplexity's advantage is flexibility and designing technology from scratch with a focus on accuracy. The team frequently updates the application's dataset, collects information from users when the chatbot answers incorrectly, and uses OpenAI's GPT-4, Anthropic's Claude, and Meta's Llama-3 large models to provide the best answers.

Kunlun Wanwei believes that in the future, AI search may not eventually replace the "big brother" in the search market, but it is possible to squeeze out the "number two and number three" in the market. "Even though there are Google Search and other search engines now, vertical search engines are also doing well, such as DuckDuckGo abroad. If your intelligent agent can meet the needs of some vertical users, your AI search can exist for a long time, and these specific fields may not be covered by search giants. For example, Google has an academic search to search for papers, but China has CNKI and overseas has Arxiv. In the vertical field, there is still a lot of room for AI search to empower through intelligent agents."

The stickiness of search products comes from user habits, content breadth, and advertising product capabilities. CITIC Securities Research believes that at this stage, AI search is more likely to serve as a supplement to the existing search ecosystem. Google's search industry layout is leading, and its advantages are still very significant under the condition that the AI ​​search experience is not significantly lagging behind. For Google, as product functions continue to improve and search costs gradually fall back to a reasonable range, Google's relative share advantage is expected to be maintained. SearchGPT had relevant market news before the May 2024 press conference, and it is expected that large-scale public testing may be around the fourth quarter of 2024, when Google's products are also expected to expand their user reach.

Liu Cong, vice president of iFLYTEK and director of the research institute, told The Paper that AI search will definitely bring about changes in the search market share. Traditional search has shown signs of being diverted after the emergence of various large models, but it is still a long way to go to replace traditional search. "The extent of replacement depends not only on the product itself, but also on the audience."

Unique content is the advantage of search

In China, the landscape of the search engine world has been changing, from portal websites such as Sohu and NetEase in the PC Internet era and traditional search engines such as Baidu, to Douyin and Xiaohongshu becoming new search portals.

An Internet company manager engaged in large-scale model development told The Paper that as early as 2014 and 2015, domestic search engine companies had already made arrangements for AI+search. "You can enter keywords and get answers directly. The starting point at the time was to be able to quickly give users results. If you give users web pages, they still have to spend time looking for answers. If you give you answers directly, the search efficiency will be much higher. But the technology at that time was actually not capable of doing this. Many of the answers given were wrong, the penetration rate was relatively low, and the product failed."

The above-mentioned person said that from the perspective of user needs, AI+search has market demand and can improve the existing search experience, but the search itself is weakening. "Even if there are no large model companies coming out, users' use of pure search tools to search for information has been weakening, and the search penetration rate of large companies has been declining. This represents a trend. Users may slowly abandon the use of search tools to search for information and may move to platforms with content." He said that vertical searches with content have a bigger market than general search tools. Searching on content platforms will see a higher density of results, improve search efficiency, and increase user trust. "The era of using technology to make search tools is over, and it is difficult to use AI to transform an industry that is already a sunset industry. We still have to pay attention to user needs and not search for the sake of searching. General searches now still use public content, and a lot of content like Douyin and Xiaohongshu cannot be searched."

Wang Yiwei also believes that unique content is still a major advantage. "Food, clothing, housing, transportation, what to buy, all this information is in Xiaohongshu and Douyin. But this data is not open to the public because it is too high-quality and is contributed by everyone. We can't get it, and neither can traditional search engines." He said that the AI ​​search engine has gone from 50 points to 90 points, covering a lot of details such as product design. The key lies in how deep the understanding of search is and how thorough it is, which is a competition of cognition.

"We will not let the MiTa AI search engine solve math problems or program. This is not the direction we want to optimize first. What we need to do now is to find game guides, useful resources and information, and accurately understand the needs of working people. When my boss gives me a task, I will immediately search it with MiTa and get a preliminary idea." Wang Yiwei said.

From the perspective of AI search technology, CITIC Securities Research Report stated that current AI search products can improve efficiency to a certain extent, but cannot fully meet user needs, and problems such as model illusions further increase user conversion costs. Wang Yiwei said that response speed, real-time performance, accuracy, and eliminating machine illusions are all issues that need to be solved next.