In the new round of the AI development wave, people often discuss how to monetize AI large models. Of course, charging membership fees is the most direct option. Another option, which has been very popular in recent years, is to apply AI capabilities in the marketing field, simplifying the cumbersome advertising production process through various intelligent agents and further improving efficiency to obtain profits.

For example, a Reuters report shows that recently, Mondelez has been trying to invest more than $40 million in developing a tool for generating advertising ideas, which is expected to save 30% to 50% of the brand’s marketing production costs.

Obviously, at this stage, the combination of AI and marketing is quite close. However, OpenAI is attempting to complete monetization in another way that is closely combined with marketing. According to relevant reports from overseas media The Information, OpenAI is trying to build its newly launched text-to-video application Sora into a social media application. This means that from now on, OpenAI is no longer just a pure AI tool provider, but has transformed into an Internet company.

In addition to the social media software, which is more like that of an Internet company, OpenAI is also making active changes in terms of advertising. According to reports from overseas media The Information, in order to better build its own advertising monetization model.

Loosening Red Lines and Performance Pressure

Although OpenAI has always had a relatively resistant attitude towards advertising since its establishment, its attitude towards advertising seems to be changing rapidly starting from July this year. Its CEO, Sam Altman, said in a podcast in July that he believes certain advertisements (such as those on Instagram) can add value to users. This statement is in sharp contrast to his position last year when he called advertising a “last resort.”

The fundamental reason for this significant shift in attitude towards advertising is performance pressure. As a company still in a stage of rapid growth, OpenAI’s employee scale has nearly quadrupled in the past two years, from 800 people to about 3,000 people, and its revenue in the first half of this year has soared to $4.3 billion.

This means that in the face of investors, OpenAI needs to find a sustainable and scalable source of income. And the fact that the basic functions of ChatGPT are free makes it not easy for OpenAI to increase its income ceiling. Therefore, even if the appearance of advertisements may arouse users’ dissatisfaction, OpenAI must consider opening up new revenue channels through advertisements to further reduce its dependence on external financing.

This change in attitude towards advertising was quickly reflected in specific implementation.

A former OpenAI employee said that Kevin Weil, a former Meta executive who served as the chief product officer, after joining in 2024, proposed in an internal memorandum to increase ChatGPT’s weekly active users to 1 billion and achieve the goal of the “toothbrush test” (referring to a product that users use every day) proposed by former Google CEO Larry Page. This strategy of blatantly emphasizing the number of users rather than product quality aroused the dissatisfaction of some executives, including the then chief technology officer Mira Murati.

And this growth concept is being implemented. In various meetings in the past year, Altman has been constantly emphasizing the importance of increasing ChatGPT’s user usage rate. To this end, the product team has launched some functions aimed at increasing user stickiness, such as allowing ChatGPT to actively suggest follow-up tasks after answering. According to a former employee, in order to encourage daily use, the company is also committed to improving ChatGPT’s accuracy in tasks such as querying stock prices, sports scores, and weather.

In contrast, from the user’s perspective, according to OpenAI’s interviews with some users, similar to GEO mentioned in the article, at present, many AI search users have begun to assume that ChatGPT’s answers are based on sponsored rankings.

At the specific implementation level, a team named “Strategic Initiatives” is exploring the advertising business. According to an on-duty employee, the team is led by Irina Kofman, who once helped establish Meta’s responsible AI team, and most of its members are also from Meta. One of the key research areas of this team is to explore whether ChatGPT can display advertisements based on its “memory” (that is, the information it remembers about users). This is highly similar to Meta’s core business model of precise advertising delivery based on user data.

Is Advertising Monetization Inevitable? But Internal Contradictions Have Just Emerged

First of all, we must be clear that OpenAI’s hope of monetizing through advertisements is not just a simple change in business strategy, but that OpenAI is accelerating its transformation from a pure research laboratory to a mature technology giant. And the first problem brought about by this transformation is the “cultural difference” between traditional laboratory employees and corporate employees of technology giants.

This cultural difference is extremely obvious and even full of contradictions. If we carefully observe the various implementation strategies mentioned above, we can find a very interesting phenomenon: these projects are almost all led by former Meta employees.

Obviously, in order to accelerate the implementation of advertising, OpenAI is recruiting a large number of “former Meta employees” to join. According to the analysis of LinkedIn data by overseas media The Information, among OpenAI’s 3,000 employees, 20% (about 630 people) have previously worked for Meta. This “Meta alumni” force is so powerful that there is a special channel in the company’s Slack for them to communicate.

The large number of Meta employees joining has made many old OpenAI employees worry about the company’s future. They believe that with the increasing influence of these “Meta alumni”, the company’s culture and strategy will be directly reshaped. Although the new application department CEO Fidji Simo has tried to appease employees, saying that he does not want to replicate Meta’s experience, internal doubts still exist.

Some employees are vigilant about Meta’s long-term struggles in user privacy and content moderation, worrying that OpenAI will follow the same old path. And although at present, OpenAI’s advertising strategy is still centered around the newly launched Sora, many people are still worried that OpenAI will also complete monetization through similar sponsored advertising rankings for ChatGPT.

After all, with the emergence of GEO (Generative Engine Optimization), advertising marketers can targetedly publish relevant articles and content, thus directly influencing AI assistants with Internet search capabilities such as ChatGPT, DeepSeek, and Doubao, and turning advertising content into AI-generated content to convey to users.

Then, a brand-new question arises: Does a brand paying to use GEO to optimize AI-generated content count as advertising?

At this stage, there is obviously no answer to this question, but for GEO practitioners and OpenAI, the significance of GEO’s existence is completely different, and they are even naturally hostile.

From the perspective of GEO practitioners, they have insight into the fact that the channel through which users obtain information is shifting from search engines to AI large models, so GEO came into being. Moreover, there is nothing wrong with GEO’s approach in terms of rules. Practitioners will also deliberately output content that conforms to the preferences of AI large models, such as breaking down content in the form of questions and answers, allowing AI to quickly identify the content, and citing industry whitepapers and certified data to enhance authority, and then using XML Schema to remember product parameters and certification information for AI to crawl.

But for ChatGPT, it is completely different. On the one hand, its user growth has almost reached the limit. According to the calculations in Mary Meeker’s AI trend report, ChatGPT’s weekly active users have exceeded 800 million. Even excluding the Chinese market and underdeveloped regions with weak Internet infrastructure such as Africa, now ChatGPT has almost covered all the people it can reach. This means that continuing to optimize the model’s capabilities to attract users is no longer the company’s top priority at this stage. But it is undeniable that with the increasing use rate of GEO, users will inevitably realize that AI is being manipulated and even “lying”, which will inevitably lead to a continuous decline in users’ trust in AI. There is no doubt that this is digging at the foundation of OpenAI.

And precisely because of this, some employees are worried that the relationship between GEO and GPT will make OpenAI directly step in once it cannot solve the problem of GEO stuffing advertisements into AI, so as to quickly make a profit while the public’s trust in AI still exists.

Conclusion

Although the “Meta-ization” trend is obvious, OpenAI is not monolithic inside. Some employees welcome the business discipline and attention to business models brought by former Meta employees, believing that this is necessary for the company’s rapid development. At the same time, the company is also trying to set up a “firewall” to protect its research culture. After Simo took over most of the business, the core research department led by the chief research officer Mark Chen still reports directly to Altman and will move into a separate San Francisco office.

Even Altman, who pursues growth, is promoting some functions aimed at preventing users from becoming overly addicted, such as prompting users to take a break after long-term use. This reflects the company’s complex mentality in the pursuit of business success and maintaining a healthy product ecosystem. However, from Sora’s social experiment, to the open attitude towards the advertising business, and then to the increasing emphasis on user engagement indicators, OpenAI is undoubtedly stepping out more and more similar footprints to those of early Facebook on its commercialization path.