news

The change came later than expected.

Since ChatGPT emerged at the end of 2022, search was considered the first market to be reshaped, but it was not until 2024 that search ushered in the expected changes.

Recently, OpenAI officially announced its entry into the AI ​​search field, making the already crowded AI search field even more lively. Abroad, emerging AI search companies such as Microsoft Bing, Google AI Overviews, and Perplexity continue to increase their investment in AI search. In China, Mita Technology, Kunlun Wanwei(300418.SZ, share price 26.55 yuan, market value 32.696 billion yuan)After entering the search market, traditional search giants such as Baidu and "search newcomer" Quark have also added AI functions to their respective products.

According to industry insiders, it is still difficult for AI search to truly change the traditional search landscape. The biggest obstacle to AI search may be "data" - data determines the quality of search, which in turn affects the user experience, and the user experience is ultimately related to user migration and retention.

Recently, AI product consultant Huang Chaoqiang said in an interview with the "Daily Economic News" reporter that AI search products are still in a very early stage. On the one hand, the products are still more text-based, with very few multimodal features; on the commercialization level, the investment and financing ecosystem is far less prosperous than overseas.

Whether the sluggish market has really ushered in new variables may still remain a question mark.

From "going to a restaurant" to "ordering takeout", AI search has both "long" and "short"

Will ChatGPT replace search engines?

This question, which sounds a bit "outdated", was the first hotly discussed topic after ChatGPT exploded the market at the end of 2022. The search engine market, which had been silent for many years, collided with a very impactful new technology. Before the change occurred, those who wanted to listen to the story had already entered the market and waited.

But it was more than a year later that the search track became really crowded.

After more than a year of anticipation for reshaping the search landscape, OpenAI finally announced the launch of its AI search tool SearchGPT recently. Prior to this, traditional search engine giant Google had launched AI Overviews and announced that it would cover 1 billion users by the end of the year. AI search startup Perplexity is gaining momentum, with reports saying that the company's monthly revenue and usage have both increased sevenfold this year. Jensen Huang even said that he uses the product "almost every day."

In the domestic battlefield, Kunlun Wanwei released "Tiangong AI Search" in August 2023, becoming the first company in China to officially launch an AI search application; Mita Technology launched Mita AI Search "with great momentum". In March 2024, the monthly visits of "Mita AI Search" increased by 500% month-on-month; MainFunc, co-founded by Jing Kun, former CEO of Xiaodu Technology, and Zhu Kaihua, former CTO of Xiaodu, launched the AI ​​search product Genspark; recently, Quark announced the upgrade of the "Super Search Box" and launched a one-stop AI service centered on AI search. Baidu, the "big brother" of domestic search engines, chose to build in AI assistants and intelligent agents for users to call at any time. In the first quarter earnings call this year, Robin Li said that about 11% of search results were calculated using generative artificial intelligence technology.

In February this year, global research and advisory firm Gartner predicted that by 2026, search volume on traditional search engines could drop by 25%.

How big is the difference between AI search and traditional search engines?

The reporter of Daily Economic News searched the most common health scenarios on Baidu, Quark, Mita, and Tiangong AI search at the same time: "What are the symptoms of lumbar spondylolisthesis?" From the results, Baidu and Quark's AI search results are deeply integrated with their respective traditional searches, and the AI ​​search results are linked to the content products of Baidu Health and Quark Health respectively. In addition, from the search results, Quark does not make a clear distinction between AI-generated content and traditional search results. The results of the original search engine are naturally supplemented after the content generated by the AI ​​search summary.

As new players in AI search, Tiangong AI and Mita Search have some differences in the way they display information compared to Quark, Baidu, etc. Mita Search and Tiangong AI Search have added several modes, "simple", "enhanced" and "research", when presenting search results, corresponding to results of different levels of detail. Mita Search has also added functions such as outline and presentation document generation. At the level of result presentation, Mita also lists more extended results such as "latest treatment methods" and "quality of life improvement strategies".

It is not difficult to see that in order to make up for the relative weakness of insufficient proprietary content products, new players have introduced more AI functions and richer result displays.

Feng Lei, former product manager of MiniMax Conch AI and owner of AI technology self-media "Orange Soda Shop", described the difference between traditional search and AI search as "going to a restaurant to eat" and "ordering takeout at home" in an interview with the Daily Economic News. He said that traditional search provides a list of URLs through which users find content, while AI search provides direct answers, similar to takeout services, and users can wait for AI to deliver the answers to their doorsteps without leaving their homes.

Feng Lei believes that the search market has been stable for a long time, and AI search will be a new variable and a more futuristic form, because it omits the intermediate steps and meets the needs quickly and directly. However, to change the market structure, AI search needs to gradually do a good job in the previous search scenarios.

In an interview with the Daily Economic News, Quark also said that search is not a new thing, and AI search is not a brand new thing either. In Quark's view, the need for information retrieval, creation, and summarization has always been the core need of users. The new experience brought by AI search can better meet the needs of users and solve practical problems in all aspects. Search is the entrance for users to obtain information, which meets the needs of users for information retrieval. After obtaining information, users also have the needs of information processing and information generation.

The required model capabilities are not much different, and the barrier is still the data scale

Contrary to the outside world’s impression, AI search may be something with low barriers.

On January 25, Jia Yangqing, a former Alibaba executive, released a demo on his social platform, using less than 500 lines of Python code to implement an AI conversational search engine. AI product consultant Huang Chaoqiang gave an example, saying that using Bing or Google API to get search results, using Prompt (prompt word) engineering or SFT (Supervised Fine-Tun-ing) to do fine-tuning in the middle, and then letting the big model output the corresponding content based on this, this process actually has no big barriers in essence.

An insider of a search company also admitted in an interview with the Daily Economic News that when comparing various AI search products, they are not actually comparing model capabilities, because the technical model capabilities required for AI search are not very different. The key to ultimately determining the quality of AI search products lies in the product capabilities themselves. This is like everyone can buy nails and hammers in the market, but whether they can make high-quality products shows the gap between companies, and the reason for this gap lies in product capabilities.

This means that in the competition for AI search, at least for now, the construction of barriers still follows the rules of the traditional search engine market - scale barriers.

In the past, search engines, as information portals, were able to continuously optimize algorithms and improve the relevance and accuracy of search results by collecting and analyzing massive amounts of user search data over a long period of time. Data accumulation is an important competitive barrier, because it is difficult for new entrants to obtain the same amount and quality of data in a short period of time. Over time, users will form usage habits and loyalty to a particular search engine, and this stickiness makes it difficult for users to switch to other platforms. This user loyalty has become a barrier that competitors find difficult to break through.

"Good data is the first principle," Feng Lei still believes that whether it is search or AI search, mastering the best data can form a barrier. New entrants need to find alternatives to these relatively private data. And whoever can provide higher quality and higher value data will have a greater chance.

AI search is still some distance away from becoming a "disruptor" in the existing search market and reshaping the search industry, and the biggest constraint may be "data". Data determines the quality of search results, which in turn affects user experience, and user experience is related to the migration and retention of end users.

Integration is the trend, but AI search is unlikely to shake up the market for now

But some people also believe that the birth of AI search is not to subvert or replace traditional search engines.

Practitioners from the above-mentioned search companies believe that the current industry structure remains very solid. The data accumulated from long-term search work is a moat and cannot be replaced by companies that start from scratch in AI search.

Is the threshold for AI search high?

He believes that, on the one hand, it seems that AI search can be built by just borrowing information indexes and adding a model. But in fact, once the information index or information source is blocked, the product will be immediately affected. On the other hand, search is not as simple as having information indexing capabilities. Real search requires a deep understanding of user needs through long-term accumulated user search habits. At the same time, for the search function, users have formed fixed usage habits, and the possibility of new products leveraging users through search functions is almost zero.

It is becoming increasingly difficult to convince users to download a new app. For new entrants, AI search is currently unable to replace traditional search, and new products have not yet achieved the user's real daily search experience and habits. "It is unlikely to expect users to download a new app just for the AI ​​search function."

AI search and traditional search are not in a life-and-death relationship. In the opinion of the above-mentioned search industry practitioners, AI search can solve problems that are difficult to efficiently display and solve in traditional search. Only by combining the two can a perfect solution be built.

The integration of traditional search and AI search is an inevitable trend. Huang Chaoqiang believes that judging from Google's promise to cover 1 billion users by the end of the year, this is irreversible. The reason for this is how to better meet the needs of users - if one answer can satisfy the user, why give the user ten links?

For new entrants, the key to changing the current market structure is how to meet user needs more deeply.

At present, for specific knowledge content, AI search can directly integrate and display relevant knowledge, and even present it to users from multiple dimensions. The richness of its content far exceeds that of traditional search, which is a clear development direction. In order to meet user needs at a deeper level, it is necessary to deeply understand the user's true intentions. It is difficult to make accurate judgments by relying solely on the questions asked by users in AI search. However, if AI search is combined with terminal technologies such as mobile devices or browsers to obtain richer user data, the specific needs of users can be grasped more accurately. In addition, in addition to information integration, it is also necessary to consider whether services can be integrated into it, or how to better meet user expectations in terms of information presentation.

For new players, another breakthrough direction comes from vertical search attempts. In addition to general AI search attempts, some AI searches focused on vertical fields have also begun to attract market attention. For example, in the e-commerce field, Amazon and Alibaba International have successively launched e-commerce search tools.

The old forces are unwilling to be overthrown and begin to integrate new forms, and the new players will decisively develop along the new path. In the new arena of AI search, everything is still full of possibilities and unknowns.

Daily Economic News