Decoding Kimi: How did the big model application become so popular?

If you live in China, you may not know ChatGPT, but you must know Kimi. From elementary school to university, from students to workers, you can't avoid Kimi's advertisements. Liu, who just entered junior high school, searched for teaching videos on Bilibili, and a soft advertisement popped up saying: "If you have any questions about your homework, ask Kimi directly."

Xiao Wang, who surfs the Internet on 5G every day and is always on the front line of gossip, saw an app recommendation in his spare time: "Kimi Smart Assistant, helps you understand the hot memes on the Internet in real time."

Xia Qian (pseudonym), who was scratching her head and working on her thesis for fear that she would fail the plagiarism check, was just about to watch a video to complain when a line of words below caught her attention: "A magical tool for writing thesis that you must not let your supervisor find out! Kimi smart assistant, generates thesis with one click."

▲ Image source: Screenshot from Station B

There are also office workers who get the AI ​​pie of "Since using Kimi, I can get off work on time every day" with just one click; a retired aunt in her 50s joins in the fun, and as soon as she hears that it's free, she immediately downloads an app to check the recipe.

▲ Image source: Screenshot from Station B

The crazier the advertising traffic, the more Kimi users will increase.

Relevant data shows that Kimi's monthly active users in March increased by 321% month-on-month to 12.61 million. Kimi's popularity has obviously attracted the attention of other large model manufacturers, and then just like all car companies began to learn from Lei Jun, an online marketing war started by Dark Side of the Moon, followed by ByteDance, Zhipu, Kunlun Wanwei, and MiniMax.

01 Decoding Kimi’s advertising myth

After Kimi became popular, his "advertising myth" spread in the market. Keywords have always been the mantra to unlock the traffic password. In order to decipher it, "Zi Quadrant" conducted search tests on Kimi one by one on the current mainstream platforms and compiled the following table.

▲ The picture is compiled and produced by Quadrant. Please indicate the source when reprinting

By comparing Kimi’s spell settings on different platforms, we can clearly see his marketing ideas and promotional focus.

Kimi has both hard and soft advertising on Bilibili. Technology bloggers are the soil for Kimi's success. He mainly uses video explanations, targeting scenes such as workers, professionals, work efficiency, and ChatGPT-related hot words to obtain traffic. Judging from the frequency and type of advertisements, it is also Kimi's main battlefield.

Chen Rui, CEO of Bilibili, once revealed that technology is one of the most popular contents on Bilibili, and AI is the fastest growing among the technological contents. In the past year, 200 million users have watched technology videos on Bilibili, and the number of views of popular science content has increased by nearly 200%. UP hosts have produced more than 1 million AI-related articles throughout the year, and 60% of the consumers of AI-related content are born after 2000.

Taking into account the user attributes of Bilibili and Kimi's current user portraits, which are roughly students, college students, interns, workers, and professionals, this also explains why Liu and Xiao Wang can accurately receive Kimi advertisements.

On Bilibili, Kimi has associated almost all the terms that users may search for "AI" on Bilibili, such as ChatGPT series, AI series, efficiency improvement series, assistant series, working people series, etc.

On Bilibili, Tiangong also has the same idea of ​​investing in traffic. One of the benefits of doing so is to dilute the labels of competitors to a certain extent. For example, Zhipu, which claims to be a benchmark against OpenAI, failed to win entries such as ChatGPT, domestic ChatGPT, and ChatGPT substitutes. Xingye, which sells AI chat and role-playing, did not put any effort into the two entries of AI role and AI role-playing.

The "all" in the associated terms implies two operational logics. One is that it doesn't matter whether the cat is black or white, as long as it catches the mouse, it is a good cat, and you should go all in if others invest in it; the other logic is to cover them as comprehensively as possible after knowing which are your potential user groups. Kimi is obviously the latter.

As for Douyin, according to "Zi Quadrant", Yuean once tried to copy the key hot words of Bilibili, but the effect was not satisfactory.

Therefore, Yuean changed its marketing strategy and focused on Yang Zhilin’s personal label “Tsinghua genius young entrepreneur”. So we can see that Kimi’s appearance on Douyin is always associated with Tsinghua University.

▲ Source: Screenshot from TikTok‍‍‍‍

If you search for words like "Tsinghua AI", "Tsinghua AI Artificial Intelligence", "Tsinghua AI Artificial Intelligence Company", "Tsinghua AI Software", "Tsinghua AI Smart Assistant" on Douyin, you can find a large number of soft advertisements placed by Kimi, and the unified wording is: "Developed by Tsinghua University team, completely free to use, unlimited times".

This strategy failed among the so-called highly educated people on Zhihu. After all, Zhihu has always been Zhipu's main battlefield, so the "Tsinghua" trick failed. Instead, it exaggerated the emotions, took "domestic substitution" as the core strategy, and conducted a series of tool comparisons and evaluations directly against its opponents.

Obviously, in order to fire the first shot in the marketing war, Yuean has made a lot of preparations.

We found that in addition to front-end advertising, the product side also made corresponding changes to welcome users.

Focusing on user experience, Kimi has adapted interactions between different devices.

An independent developer gave [Zi Quadrant] an example. On the surface, the web version, APP version and mini program version of Kimi are not much different from other similar products, but when it is used as a plug-in in a web page, Kimi's presentation effect is better. The above developer emphasized, "At least it shows that Dark Side of the Moon took more user scenarios into consideration when developing products." [Zi Quadrant] also noticed that Kimi's App version recently ushered in a UI reconstruction. While optimizing the experience issues, it launched the "Dark Side of the Moon" light theme mode.

▲ Image source: Screenshot of Kimi Smart Assistant

Last year, when other large model companies were still struggling to open up the 2B market, Kimi had already announced his "debut declaration" to abandon 2B and turn to 2C applications.

However, according to what "Zi Quadrant" learned from an investor close to Kimi in March: "It is not Yuean's own rhythm to invest heavily in the market for Kimi, but a 'last resort' to speed up the pace of cashing out under the urging of investors."

02 Are the “craftsmen of nature” imitating others?

With Kimi as an example, other big models are gradually increasing their marketing efforts. Following the same method, "Zi Quadrant" also sorted out the marketing strategies of Doubao, Zhipu Qingyan, Kunlun Wanwei Tiangong and MiniMax Xingye, which are currently active on social platforms besides Kimi.

▲ The picture is compiled and produced by Quadrant. Please indicate the source when reprinting

By sorting out and making a table of "Recommended vocabulary for large-scale application products on different platforms", we found that Kimi made his fortune on Bilibili, Doubao became the king of Tik Tok, and Others started to go with the flow.

It is actually easy to understand that Doubao corresponds to Douyin. As the two biological sons of ByteDance, Douyin’s traffic is Doubao’s own backyard.

Doubao gave almost no breathing space to others. All the tags snatched by Kimi and other companies on Bilibili were snatched up by Doubao. Any keywords related to AI were filled with Doubao’s advertisements.

According to incomplete statistics from [Zi Quadrant], when the most exaggerated time came for an entry, the number of hard and soft advertisements for Doubao was nearly ten, not including the 24-hour live broadcast on Doubao’s official account, the promotion of Doubao by top influencers, and the collaboration between Doubao and ByteDance’s Xinghui and Maoxiang (Hualu).

However, according to the new regulations on massive advertising, the use of AIGC content production-related tool software will be banned from April 2. However, when we tested it after April 20, we were still able to search for a large amount of related content. I don’t know if this rule also applies to Doubao.

▲ Source: Screenshot of the massive advertising background‍

Obviously, continuing the marketing war in this way is not a solution.

The somewhat rough category divisions such as "AI + scene words", "domestic substitution (ChatGPT alternative)", "efficiency tools for workplace people", and "Tsinghua Entrepreneurship" make it easy to imitate.

From the table, we can see that the traffic labels of the five large model applications have a high degree of overlap, which further strengthens the feeling of homogeneity from the user's perspective.

When competitors joined the traffic-investing team one after another, Kimi was the first to react to this problem and posted a topic on Xiaohongshu titled "Kimi can actually do this", hoping to explore more usage scenarios by collecting users' prompts, and by filtering the tags that come with Xiaohongshu users' posts, Kimi can further refine the user portrait.

For example, the label "college student" can be further refined into papers, literature search, games, electronic pets, classes, etc.

This is a typical product manager's approach, which is to improve product capabilities to make up for temporary technical deficiencies.

▲ Source: Screenshot of Kimi’s official account on Xiaohongshu

Faced with this kind of pressure, Zhipu Qingyan, Tiangong and Xingye all “lay flat” to some extent. Zhipu Qingyan specializes in offline advertising, with “Kimi-flavored” posters scattered all over airport billboards, subway stations and office buildings; MiniMax Xingye focuses on overseas markets, with domestic advertising only a small part.

In comparison, Tiangong seems to be just "following the trend". It spent money, but its presence is still very low.

03 Putting “new wine” in “old bottles”, is it necessary to fight a marketing war?

Many people only saw the large model applications of Kimi, Zhipu Qingyan, Tiangong and others in this fierce marketing war, but in fact this is already the second wave of marketing activities of large model companies.

The first wave was pure large-model marketing.

There was a lot of hype before the New Year last year, but there were not many contestants at that time, and it died down not long after.

Although the influx of large models into Internet platforms is a new thing, it is essentially a business that loses money but gains publicity.

On short video platforms that are more entertainment-oriented, users have little awareness of big models, and it is difficult to tell others what big models are used for in a simple and clear sentence.

Even if you gain some users through marketing, it is difficult to bring about direct conversions.

So, after only lasting for more than a month, the first wave of pure large model marketing finally disappeared.

The second wave is the marketing of large model applications.

Starting from Kimi's outage, the whole big model circle was involved by means of online traffic and offline territory grabbing.

Although the first wave of large-model marketing failed, there is a hidden logic in it: only with products can users be attracted; if you want to have a product, you must first have a large model of the base.

Behind Kimi is the self-developed model of Dark Side of the Moon, behind Zhipu Qingyan is ChatGLM, and behind Tiangong is the Tiangong big model. This indirectly shows that the threshold of this big model application marketing war is not low.

Returning to the two marketing behaviors in the big model circle, it is essentially "new wine" in "old bottles", using the old tricks of Internet marketing to wrap the new soul of big models. The essence of marketing is customer acquisition and conversion.

A typical example is the once glorious online education, which spent tens of millions of yuan on advertising in just one summer vacation to attract students. This does not apply to big model companies, because the download of big model applications is free from the beginning.

Online education companies can spread their investment costs by expanding the number of students enrolled, but the opposite is true for large-scale model applications. The larger the user base, the higher the maintenance and token costs they bear.

If the big model company only completes the task of attracting traffic and the first wave of initial users, it will fall into a vicious circle of more marketing and more losses.

The only solution is to increase the proportion of paying users or to increase revenue through value-added services. However, it is questionable whether users are willing to pay in a short period of time.

There is another problem here. Whether it is from the investment platform or the potential users, the current wave of large model companies are competing for the same user group.

The reality is that the functions of each company are basically the same, and the number of users to be attracted is limited. In the second half of the marketing war, if the competition continues, there may not be enough users.

So far, we have been discussing the premise that Kimi and others’ marketing behavior is necessary.

But don’t forget, what is the key to Kimi’s success?

It is the user tap water at the beginning, and it is the product thinking that is superior to that of competitors.

The aggressive marketing strategy of Dark Side of the Moon can easily make people overlook one fact: Kimi’s investment follows the product logic, while his peers are only focused on fighting business wars.

Drawing on practices abroad, large model companies are each the king, and the idea that users have to choose one may only be temporary.

Poe, an AI Chatbot platform produced by Quora, the "American version of Zhihu", is a reference case. The platform integrates mainstream AI large models such as ChatGPT, Claude-insant (similar to GPT-3.5), GPT-4, Claude2, Google-PaLM, Llama2, etc., allowing users to use multiple large models with one account, allowing users to jump freely between large models while avoiding the tediousness of opening multiple memberships.

▲ Image source: Screenshot from Poe’s official website

Poe is already showing signs of growing in size.

Currently, Poe is one of the five largest generative AI-related platforms, and creators have built more than 1 million robots on the Poe platform. Although Poe has been criticized for being a "shell", the value it provides to users is real. For example, users can clearly understand the characteristics of each AI application during use and pay on demand.

Poe is not yet mature, but its incentive mechanism for creators and GPTs-like ecosystem are ahead of OpenAI.

Putting aside the marketing war, perhaps China can also explore new ways of aggregating large-scale model applications, supplying them, and distributing commercial profits. This will be a new topic for large-scale model companies.