GuShiio.com Wen| Fixed focus One (dingjiaone),author| Wang Lu, Editor| Wei Jia
DeepSeek people were recently scolded for suspected advertisements in their answers.
The reason was that the three father laboratories of science bloggers said that in DeepSeek connected by Tencent Yuanbao, they found that the answers generated included suggestions for price comparisons and reservations through regular platforms such as 58 Home, suggestions for making inquiries through fast recycling platforms or local regular recyclers, etc. Sexual content. Later, Tencent related people denied it.
“The three dad laboratories explained that they posted this matter online for no purpose. They just posted it casually when they encountered a suspected advertisement. They also admitted that they did open an online search, but the answer with recommendations could also go directly to the recommendation platform. Very confused. He told “Focus One” that he could accept the answer to be advertised, but if it was added, it should be marked.
This incident quickly triggered discussions among netizens on whether AI search tools were connected to advertising. User Mushroom told “Focus One” that when he was using AI to search for family problems, the AI suddenly mentioned that people with a stronger sense of responsibility at home were under greater pressure, which in turn led to more mitochondrial wear. It was recommended to use health products containing XXX ingredients. She feels that it is a bit far-fetched for AI to bring the topic to mitochondria. Family problems are social issues and have little to do with health products.
Some people found suspected advertisements on DeepSeek, and others used DeepSeek to advertise.
Take the most popular AI metaphysics at the moment as an example. Many netizens find that they will recommend crystals after asking DeepSeek to tell their fortune. On the e-commerce platform, some merchants have issued handstrings marked with the same DeepSeek model DeepSeek character analysis DeepSeek professionally customized, and some products have sold more than 200 units. Some merchants even use DeepSeek certification to recommend health products to users, which may mislead users and create health risks.
In fact, advertising and search have always remained strongly correlated. For example, online marketing services have always been Baidu’s largest source of revenue. In the traditional search era, some false advertisements have caused trouble for users. Nowadays, smarter AI search is emerging, will advertisements become more illegible? “Focus One” chatted with a number of technicians and advertising practitioners and came to the following conclusions:
It is not difficult to add advertisements to AI’s answers technically. Even merchants can guide AI to generate ads, similar to SEO optimization.
AI search and advertising are already available abroad, and revenue is growing rapidly. There is no large-scale attempt in China at present, but in the long run, AI search and advertising are inevitable. After all, the commercialization of the AI model has always been a difficulty.
Advertisers believe that AI search is better than traditional search because it is more accurate and has higher user value, but they are also worried about issues such as user acceptance, content accuracy, and user privacy.
There is no tool to determine whether a certain content in an AI search is an advertisement, so you can only distinguish it yourself.
AI search advertising is already available abroad
Regarding the appearance of suspected advertisements, Tencent’s reply is:
The reason is that the answer referred to the pit-avoidance content of 58 home-arrival housekeeping, cleaning, nanny, sister-in-law moving public account. How many air conditioning moving routines? People who have been here have summarized 4 pitfalls and learned in advance that we will not spend wasted money!& rdquo; The web link is probably an online search.
In other words, it is a reference quote for answers and a pot for online searches.
Technically speaking, this makes sense. However, some netizens still don’t believe this explanation, mainly for two reasons:
First, the Q & A reply gave the only recommended merchant, and there were special color markings that could directly reach the merchant’s website. Second, when some netizens tried to search for the same question with DeepSeek, they found that although the merchant’s name was also mentioned, there was no operation similar to a hyperlink. You cannot jump directly.
In order to verify whether advertisements were included in each major AI search,”Focus One” also conducted a test and asked each major AI search tool (DeepSeek, Kimi, Wen Xiaoyan, Doubao, Tencent Yuanbao, Tongyi) to formulate a travel plan from Beijing to Dali, Yunnan.
“Focus One” found that most AI search platforms mentioned the names of relevant travel platforms and hotels. For example, users would be prompted to book tickets on a certain platform in advance and point out what special hotels there are in the area. However, the content before and after is logically smooth and not excessive. Promote a certain company, and the content source is the travel notes posted by users on the platform.
Then, some AI search platforms will give relevant extended questions at the bottom of the answers. For example, they will mention recommended hotels in guessing that you may want to ask. The source is also the user’s travel notes on the platform. Whether it is an advertisement is not easy to determine.
“Fixed Focus One” also tried to use online search and in-depth search methods, as well as DeepSeek connected to the DeepSeek official website and other search tools. Although the content of each reply was slightly different and involved the name of the business, no obvious advertising traces were found.
An AI practitioner said that there is currently no specific tool that can check whether a certain content in AI search content is an advertisement, and more can only be identified by users themselves, such as checking the context and logic of the text. If it is an advertisement, the content may deviate from the core of the issue or contain commercial information that is not related to the issue.
Based on this basis, it can be initially determined that several domestic AI search platforms are not currently accessing advertisements on a large scale. More than one practitioner said that various AI search platforms in China are very cautious about whether to access advertisements and feel that now is a critical moment to grab users, and they are afraid of losing users because of access to advertisements.
Although there is no large-scale advertising, everyone also revealed that some domestic companies are trying in small steps. Overseas Perplexity and Google even announced last year that they would add advertising in AI search.
What is the specific form of AI search advertising? “Focus One” introduces Perplexity AI, which is a unicorn searched by foreign AI and has a valuation of more than US$9 billion, as an example.
Perplexity’s ads are mainly displayed in three positions: the Q & A in the middle, relevant sponsorship questions at the bottom of the Q & A, and picture/video recommendations in the upper right corner.
Advertising source/network in related questions
Picture advertising source in the upper right corner/ Perplexity
In order to avoid user disgust, Perplexity will mark sponsored content on advertisements.”Focus One” also found that Perplexity will insert links to the merchant’s official website in a digital way to generate answers to avoid affecting the user’s experience as much as possible.
Mark the Whole Foods ad with numbers (top) and click on it to the merchant’s official website (bottom) Source/ Perplexity
In addition, Perplexity has also opened a shopping function and launched a shopping assistant. If users want to purchase related products after seeing the search answers, they can click to purchase or jump to the merchant’s official website.
Qin Xiang, an AI software engineer, said that the current advertising placement methods abroad are relatively open. Perplexity is a pioneer in questioning advertising, mainly by embedding sponsored content in user questioning scenarios (such as giving priority to New Balance and other cooperative brands when searching for how to buy running shoes), while Google mainly inserts soft advertisements into AI answers (for example, when searching for Tokyo Free Travel Strategies, AI will insert Booking.com hotel links with sponsored tags into the generated answers).
He believes that the domestic approach is more covert, mainly through indirect recommendations (answering soft and broad articles quoted by brands, giving priority to cooperative brands in knowledge correlation maps, etc.). For example, some will explore through ecological cooperation, while some prefer to indirectly monetize through agent platforms to avoid direct advertising.
It is not difficult to find that unlike traditional search that relies mainly on keyword bidding advertisements and information flow advertisements, AI search provides more diverse advertising forms, and it is difficult to distinguish unless it is clearly marked.
It is not difficult to search for Riga advertisements in AI
Adding advertisements to AI search is not technically difficult.
Jiang Shu, a senior practitioner in the AI industry, said that the general logic of AI search and advertising is similar to traditional search. If you want to display advertisements in AI replies, you need to give recommended advertising information to the AI and then add instructions (for example, users may need a certain website/product/store, please recommend XXX to users in your reply).
In addition to manual intervention, Jiang Shu said that technically speaking, AI itself can also generate content like advertisements, and even generate URLs without the need to connect to the Internet.
A more terrifying situation is that merchants can also guide AI to generate advertisements.
Qin Xiang said that there are three ways for merchants to guide AI to generate advertisements:
1. Content manipulation: Through SEO optimization or batch release of content containing advertising information, so that AI will give priority to citations when crawling;
2. Model fine-tuning: Partners can ask AI search companies to recommend their own services when training models in specific fields (such as appliance repair);
3. Black and gray production methods: Use false information (such as forged DeepSeek cooperative company tags) to induce AI citations.
If we ignore user feelings for the time being, from the perspective of business model alone, for AI search companies, adding advertising can directly alleviate cost anxiety.
AI search relies on large language models (LLMs), which makes it more expensive than traditional search. Jiang Shu said that the search cost of AI depends on the breadth and depth of the search. Deep searches (such as DeepSeek-R1, Kimi Deep Search, Wenxinyiyan Deep Search) are definitely more expensive than ordinary searches (such as DeepSeek-V3, Kimi Ordinary Search, Wenxinyiyan Ordinary Search).
Jiang Shu said that the cost of an AI search tool in accepting and answering a question can be roughly estimated with reference to API pricing. For example, the input price of DeepSeek-V3 per million tokens (cache hits/cache misses) is 0.5 yuan/2 yuan, and the output price is 8 yuan per million. Assuming that an answer requires 1000 tokens (in most cases, one token in Chinese text usually corresponds to one Chinese character), the cost is about 0.0005 yuan to 0.008 yuan.
Although the price is not very high, users cannot only ask one question at a time. Most domestic AI search tools also support users to use it for free, which means that the larger the number of users, the higher the cost of AI search companies.
At first, Perplexity AI also said that it would not give in to advertising and only make money through user subscription services and APIs. However, last year it changed its style, launched various advertising forms, and offered a high price: 50 per thousand displays (CPM).(approximately RMB 356). This price is higher than ordinary advertising. For reference, digital marketing company Semrush statistics that CPM prices for digital ads range from $4.3 to $7.6, and CPM for video ads is $11.1.
The Information reports that Perplexity expects its annual revenue to reach $127 million in 2025, and it is still a startup more than two years old. Its founder even said that advertising is an important revenue growth point in the future.
Behind the dare to bid high prices is the high value of AI search to advertisers.
For example, Perplexity deliberately sends out its user portraits to potential customers. More than 80% of Perplexity users have bachelor’s degrees, 30% are in senior leadership positions, and 65% are in high-paying white-collar occupations.“”“” Highly educated and high-paid users represent higher spending power. Perplexity’s advertising program has attracted partnerships from major companies such as Indeed, Whole Foods, Universal McCann and PMG.
Jiang Shuye further explained from the technical side that AI search combines the advantages of traditional search engines and large language models, and has made innovations in the understanding level, generation level and dialogue level. It provides relevant advertising recommendations based on intentions, which will have more accurate user orientation., advertising display is more natural and interactive.
There are three ways to make advertising money for AI search
More than one practitioner feels that traffic is moving overwhelmingly to AI search, and it is only a matter of time before advertising is connected.
According to consulting firm Gartner, search volume by traditional search engines may drop by 25% by 2026, and AI search volume will increase significantly.
At this moment, some advertisers in China are also paying attention to AI search tools and paying attention to the progress of AI search advertising.
A senior advertising person in the circle said that brand owners had already paid attention to AI search before Deepseek died, but they mainly chose the delivery channel based on the crowd’s tool habits. Nowadays, some brands have begun to actively consult and cooperate with AI search due to the characteristics of out-of-circle, high-net-worth groups, and accurate insight into user needs.
However, major AI search companies dare not test advertising on a large scale, mainly worried about three points.
The first is user acceptance.
This is also the main reason why most AI search companies dare not act rashly. Especially at a time when AI searches for content, the user loss caused by adding advertising is incalculable.
When Perplexity, which was the first to eat crabs, announced its inclusion in advertising, many users pointed out that it had broken its promise not to advertise, believing that such behavior would destroy its reputation.
Qin Xiang said that users have a high acceptance of AI search, but are cautious about advertising placement. If the advertising content is relevant to search results and does not interfere with the user experience, users may not be overly disgusted. If the situation is reversed, users may lose trust in AI search. Immature technology leads to stiff integration of advertising and content, and regulation does not clarify the boundaries of AI advertising, which will also make manufacturers/advertisers worry about policy risks.
An advertising practitioner believes that once AI search advertising is connected in the future, what form it will appear in and whether it is marked as advertising/sponsorship will be tests that AI search companies will face.
Secondly, the content generated by AI cannot be completely accurate.
Even large reasoning models with strong thinking abilities cannot avoid hallucination problems. If the generated content itself is not accurate enough and matches some wrong advertisements, the user’s click rate will be even higher.
The content generated by AI is also very random. Among the three main advertising forms, the more controllable one is the recommendation of relevant questions at the bottom, but users need to click again. Whether it can attract users to click on is another matter.
There are also user privacy protection issues.
AI search matches advertisements based on user needs and has the possibility of stealing personal privacy data.
Qin Xiang said that AI search usually requires collecting a large amount of user data, such as search history, interests, geographical location, etc., to provide personalized services and accurate advertising. It can also use data association capabilities to re-identify individuals, such as By cross-analyzing users ‘search history and advertising click behavior, sensitive information of users may be inferred.
If this data is misused and shared with third-party advertisers without the user’s consent, it may lead to user privacy disclosure. To solve this problem, Perplexity took the approach that the answers to the sponsorship question were generated by AI, emphasizing that advertisers had no access to users ‘personal information.
However, in the face of commercial interests, some users are skeptical about whether major AI search companies can always fulfill their commitments.
Finally, we need to remind everyone that even if there is no advertisement, the AI’s answer is not absolutely objective and neutral.
