Sam Altman has publicly stated multiple times that he feels uneasy about adding ads to AI responses, calling it a company’s "last resort."

Yet in the face of reality, Altman has chosen to back down.

In the code of ChatGPT’s latest Android beta version 1.2025.329, ad-related strings such as "ads feature," "search ad," and "bazaar content" have clearly emerged.

Although these code snippets are only a dozen lines long, they are enough to indicate that OpenAI is making technical preparations to embed ads into ChatGPT.

Admittedly, ads will bring substantial revenue to OpenAI.

However, this will completely erase the long-established labels of AI search, such as structured and ad-free outputs. Finding a balance between ads and high-quality content has become a common challenge for all large model companies in 2026.

I. Ads Are Poised to Become One of OpenAI’s Largest Revenue Streams

According to data disclosed by OpenAI, the company’s expenses in the first half of 2025 reached $2.5 billion. At this rate, by 2029, the company will have burned through $115 billion.

Even though ChatGPT currently has nearly 900 million weekly active users, only about 5% of them are paid subscribers.

Based on OpenAI’s pricing—$20 per month for ChatGPT Plus and $200 per month for ChatGPT Pro—these subscription revenues are far from covering costs, even with a 5% paid rate.

To fill this financial black hole, OpenAI has formulated a monetization plan for free users. According to internal company forecasts, starting in 2026, the average annual revenue per free user will reach $2, and this figure will grow to $15 by 2030.

The most direct monetization method is ads. OpenAI aims to generate 20% of its total revenue from ad-related sources by 2030 through its ad business.

Altman also adjusted his stance in a recent interview, stating that ads are "uncomfortable but not entirely unacceptable." This shift in attitude is more a compromise to reality than a change in philosophy.

From OpenAI’s internal discussions, its ad format is fundamentally different from that of traditional search engines. Altman refers to it as "intent-driven monetization," with the core being to transform ads into "conversational recommendations" rather than simply inserting ad links into search results.

This model works as follows: When a user asks, "Recommend the best running shoes," the AI will naturally recommend a paid brand while providing professional advice.

According to The Information, the recommendation will not be awkwardly labeled "This is an ad" but integrated into the conversation flow, such as "Based on your training intensity, Nike’s Pegasus series might be a good fit—its cushioning technology can effectively protect your knees." OpenAI may profit through revenue sharing or charging fees for "priority access to recommendation algorithms."

OpenAI has created multiple ad display prototypes internally to simulate different presentation methods. One plan is to show sponsored information in the sidebar of ChatGPT’s main response window, clearly labeled "Contains sponsored results." This design references Google’s ad display method but attempts to make ads look more like "supplementary suggestions" rather than traditional page ads.

Another plan from OpenAI is to display ads in the page sidebar when users click for further information.

For example, if a user asks for travel recommendations in Barcelona, ChatGPT will first recommend attractions like Sagrada Família—this recommendation itself is not sponsored content. But when the user clicks the link to the Sagrada Família introduction, a pop-up window containing sponsored links from multiple paid travel service providers will appear.

The purpose of this "secondary trigger" mechanism is to avoid putting users under commercial promotion pressure during the initial conversation.

OpenAI is also exploring a "generative ad" model. Instead of simply displaying copy provided by advertisers, the AI automatically generates customized recommendation content based on the user’s specific needs and conversational context.

For instance, when recommending the same running shoe, it will emphasize durability and support for marathon runners, while highlighting comfort and cost-effectiveness for beginners. OpenAI believes this dynamically generated ad content can theoretically drive higher conversion rates.

When OpenAI employees evaluate these plans internally, their core debate centers on "how to display ads without causing user backlash."

At the same time, there is a consensus within the team: ads must never interrupt or disrupt the natural flow of the conversation, and all ad exposures should be triggered only after the conversation has progressed to a certain depth.

This principle means free users will not see ads in every conversation—ads will only appear when the conversation content clearly points to consumption, travel, or product decisions.

II. Shopping Research and the Race for Ad Dominance

In November 2025, OpenAI launched the Shopping Research feature, transforming ChatGPT into a personal shopping assistant. Users can describe their shopping needs via voice or text, and the AI will ask targeted questions before providing comprehensive purchase recommendations, including product links and price comparisons.

When launching this feature, OpenAI explicitly stated that it currently does not charge any commissions or affiliate marketing fees, and merchants cannot pay to influence rankings. However, the industry generally believes this is only a temporary strategy. Shopping Research uses a specially trained GPT-5 mini model capable of reading product pages, specifications sheets, and trusted reviews—an infrastructure that can seamlessly connect to an ad system.

OpenAI has already established partnerships with retailers such as Walmart, Target, and Etsy, allowing users to search for and purchase products from these platforms within ChatGPT. While this currently constitutes free traffic referral, once the ad system goes live, these partnerships can be immediately converted into paid promotion channels.

During Black Friday 2025, e-commerce website traffic driven by ChatGPT increased by 28% year-on-year, and retail website traffic from AI surged by 670% on Cyber Monday. If OpenAI can insert ads into this traffic, even with a conversion rate only half that of traditional e-commerce, the revenue will be substantial.

However, to resist the impact on its shopping gateway and ad revenue, Amazon blocked AI crawlers from multiple companies including ChatGPT, Meta, and Google shortly before Christmas, preventing them from scraping product information.

At the same time, Amazon launched its self-developed AI shopping assistant Rufus to defend the platform’s gateway and respond to changes brought by AI search.

To further explore its ad business, OpenAI poached ad executive Shivakumar Venkataraman from Google, who previously led multiple core projects in Google’s search advertising division.

In addition, OpenAI has posted multiple ad-related positions on Linkedin, including Ad Engineer, Ad Product Manager, and Ad Strategy Analyst.

These moves all indicate that for OpenAI, ads are no longer a question of "whether to do it" but "how to do it."

Driving OpenAI’s decision are not only cost pressures but also competitive pressures. Google announced in 2025 that it will introduce ads into Gemini, with a planned official launch in 2026. This poses a direct threat to OpenAI. If competitors are using ads to reduce user costs and increase revenue, OpenAI will find it difficult to remain an exception.

Perplexity AI has already launched a "Sponsored Answers" feature, clearly labeling sponsored content in search results. While this feature has sparked some user dissatisfaction, it has also proven the feasibility of AI search ads. OpenAI clearly does not want to fall behind in this race.

In terms of market size, the global digital advertising market generates over $1 trillion in annual revenue, with Google and Meta capturing most of the share. If OpenAI can enter this market through ChatGPT, even capturing just 5% of the share would bring in $50 billion in annual revenue—enough to support the company’s long-term operations.

But OpenAI also faces the challenge of user trust. Once AI responses begin to incorporate commercial interests, how will users judge whether a recommendation is based on objective analysis or advertiser payments? When ChatGPT recommends a product, users will wonder if it’s because the product is truly good or because the brand paid for it.

There is no easy answer to this question.

OpenAI is now attempting to walk a tightrope between "intellectual neutrality" and "financial self-rescue." If this balance fails, ChatGPT may transform from an "AI assistant" into a "product-promoting chatbot."

III. AI Search Eats into Traditional Keyword Search

Undeniably, AI search is eroding traditional keyword search.

The market share of traditional search engines is declining at a visible rate. According to a 2025 search behavior research report released by HigherVisibility, Google’s share in general information search dropped from 73% in February 2025 to 66.9% in August—a 6.1% decline in six months.

Changes in the Chinese market are even more drastic. Baidu’s mobile search market share plummeted from 94.72% in 2021 to 58.6% in the third quarter of 2025.

Data from QuestMobile shows that Baidu’s media status index fell to ninth place, ranking behind Douyin, Taobao, WeChat, Kuaishou, and Xiaohongshu. Furthermore, Baidu’s online marketing revenue dropped by 18% year-on-year in the third quarter of 2025—its sixth consecutive quarterly decline.

However, these users are not abandoning search altogether; instead, they are shifting their search behavior to AI tools.

According to the "2025 China AI Terminal Ecosystem Development Research Report" released by QuestMobile, as of October 2025, the number of mobile AI users in China reached 720 million, accounting for 50% of the total number of Chinese netizens. Daily AI search requests exceeded 2.8 billion, with commercial queries accounting for 41%.

A survey of 1,000 users by the German digital industry association bitkom found that 50% of respondents sometimes use AI chatbots to find information instead of traditional search engines. Among the 16-29 age group, 5% rely entirely on AI search, 11% mainly use AI, and 20% use AI and traditional search engines equally.

In other words, more than one-third of young people already use AI as their primary or sole source of information.

Even more telling is the change in zero-click rates. A zero-click rate refers to when a user searches for a keyword but does not click on any links, instead initiating a new search or closing the page.

Data from SparkToro shows that Google’s search zero-click rate has reached 58.5%, and as high as 77.2% on mobile devices. This means that more than half of all searches no longer generate clicks—users leave after seeing search results and summaries. The core ad model of traditional search engines is failing.

AI search delivers higher conversion rates. The large model search conversion rate refers to the proportion of users whose needs are directly met from the moment they initiate a search to the end. ChatGPT’s conversion rate is six times that of Google Search, as users are more willing to trust direct answers provided by AI.

Behind this transformation is AI search addressing long-standing core pain points of traditional search.

Traditional search only provides lists of links—users need to open a dozen web pages one by one, and the quality of information varies across different sites, often contradicting each other, requiring users to verify authenticity themselves. A simple question often takes a long time to piece together a complete answer, and the cycle of "search → click → read → return → search again" frustrates users.

AI search directly provides complete answers, eliminating the need to filter links, click through them one by one, and piece together information. According to a joint survey by OpenAI and Harvard University titled "How People Use ChatGPT," nearly 35% of users cited "obtaining direct answers" as the top reason for using AI search instead of Google.

AI search supports multi-turn conversations, which further enriches search results. If the first question is unclear, users can immediately follow up, add conditions, or adjust directions, gradually clarifying their needs like conversing with a real person—instead of repeatedly modifying keywords to search again. AI can remember context to form coherent conversations, while traditional search treats users as "new users" every time.

The "AI Search User Behavior Research" shows that 72% of AI search sessions include more than two turns of conversation, as this interaction method is more in line with natural human thinking habits. In traditional keyword search, there is no concept of "continuity"—each search only relies on the current keyword.

What users value more is that AI answers are not driven by commercial interests and are free from the hassle of paid ranking. The ad model of traditional search engines has long been criticized—Baidu’s search results are dominated by paid ranking ads on the first few pages, burying real information, and users need to scroll through multiple screens to find non-ad content.

Although AI can also make mistakes, at least it is not "who pays, who ranks first."

The convenience of voice interaction is another important factor. Typing on mobile devices is slow and error-prone, and entering long questions is a poor experience. Voice input is 3-4 times faster than typing, making it particularly suitable for expressing complex questions. Research by Geokeji shows that 58% of users in AI search apps use voice input, while in traditional keyword search, even with voice input functionality, the usage rate remains less than 20%.

The average length of voice search queries is 2.3 times that of text searches—because voice input is more convenient, users can express their needs more completely, improving search accuracy. In scenarios such as driving, cooking, taking care of children, or exercising, when hands are occupied, voice search becomes the only option. These high-frequency scenarios have helped users develop a new habit of "just speaking when in need."

Gartner predicts that global search engine traffic will plummet by 25% in 2026. This is not a simple shift in market share but a revolution in interaction methods.

From "typing keywords" to "voicing complete questions," from "filtering a dozen links" to "obtaining a direct answer," from "single queries" to "multi-turn conversations," the "keyword matching + paid ranking" model that traditional search engines rely on for survival is being rapidly replaced by the "voice conversation + direct answer" model based on large language models.

Therefore, traditional search companies, like ChatGPT, are caught in a dilemma. Adapting to the new interaction method means their proven business model is destined to be disrupted; yet staying in their comfort zone means watching the size of that zone shrink visibly. In this transitional period where the new technological revolution is imminent, they can only be contradictory and hesitant—while firmly walking on the tightrope, with a similarly hesitant ChatGPT on the other side.

Sam Altman has publicly stated multiple times that he feels uneasy about adding ads to AI responses, calling it a company’s "last resort."

Yet in the face of reality, Altman has chosen to back down.

In the code of ChatGPT’s latest Android beta version 1.2025.329, ad-related strings such as "ads feature," "search ad," and "bazaar content" have clearly emerged.

Although these code snippets are only a dozen lines long, they are enough to indicate that OpenAI is making technical preparations to embed ads into ChatGPT.

Admittedly, ads will bring substantial revenue to OpenAI.

However, this will completely erase the long-established labels of AI search, such as structured and ad-free outputs. Finding a balance between ads and high-quality content has become a common challenge for all large model companies in 2026.

I. Ads Are Poised to Become One of OpenAI’s Largest Revenue Streams

According to data disclosed by OpenAI, the company’s expenses in the first half of 2025 reached $2.5 billion. At this rate, by 2029, the company will have burned through $115 billion.

Even though ChatGPT currently has nearly 900 million weekly active users, only about 5% of them are paid subscribers.

Based on OpenAI’s pricing—$20 per month for ChatGPT Plus and $200 per month for ChatGPT Pro—these subscription revenues are far from covering costs, even with a 5% paid rate.

To fill this financial black hole, OpenAI has formulated a monetization plan for free users. According to internal company forecasts, starting in 2026, the average annual revenue per free user will reach $2, and this figure will grow to $15 by 2030.

The most direct monetization method is ads. OpenAI aims to generate 20% of its total revenue from ad-related sources by 2030 through its ad business.

Altman also adjusted his stance in a recent interview, stating that ads are "uncomfortable but not entirely unacceptable." This shift in attitude is more a compromise to reality than a change in philosophy.

From OpenAI’s internal discussions, its ad format is fundamentally different from that of traditional search engines. Altman refers to it as "intent-driven monetization," with the core being to transform ads into "conversational recommendations" rather than simply inserting ad links into search results.

This model works as follows: When a user asks, "Recommend the best running shoes," the AI will naturally recommend a paid brand while providing professional advice.

According to The Information, the recommendation will not be awkwardly labeled "This is an ad" but integrated into the conversation flow, such as "Based on your training intensity, Nike’s Pegasus series might be a good fit—its cushioning technology can effectively protect your knees." OpenAI may profit through revenue sharing or charging fees for "priority access to recommendation algorithms."

OpenAI has created multiple ad display prototypes internally to simulate different presentation methods. One plan is to show sponsored information in the sidebar of ChatGPT’s main response window, clearly labeled "Contains sponsored results." This design references Google’s ad display method but attempts to make ads look more like "supplementary suggestions" rather than traditional page ads.

Another plan from OpenAI is to display ads in the page sidebar when users click for further information.

For example, if a user asks for travel recommendations in Barcelona, ChatGPT will first recommend attractions like Sagrada Família—this recommendation itself is not sponsored content. But when the user clicks the link to the Sagrada Família introduction, a pop-up window containing sponsored links from multiple paid travel service providers will appear.

The purpose of this "secondary trigger" mechanism is to avoid putting users under commercial promotion pressure during the initial conversation.

OpenAI is also exploring a "generative ad" model. Instead of simply displaying copy provided by advertisers, the AI automatically generates customized recommendation content based on the user’s specific needs and conversational context.

For instance, when recommending the same running shoe, it will emphasize durability and support for marathon runners, while highlighting comfort and cost-effectiveness for beginners. OpenAI believes this dynamically generated ad content can theoretically drive higher conversion rates.

When OpenAI employees evaluate these plans internally, their core debate centers on "how to display ads without causing user backlash."

At the same time, there is a consensus within the team: ads must never interrupt or disrupt the natural flow of the conversation, and all ad exposures should be triggered only after the conversation has progressed to a certain depth.

This principle means free users will not see ads in every conversation—ads will only appear when the conversation content clearly points to consumption, travel, or product decisions.

II. Shopping Research and the Race for Ad Dominance

In November 2025, OpenAI launched the Shopping Research feature, transforming ChatGPT into a personal shopping assistant. Users can describe their shopping needs via voice or text, and the AI will ask targeted questions before providing comprehensive purchase recommendations, including product links and price comparisons.

When launching this feature, OpenAI explicitly stated that it currently does not charge any commissions or affiliate marketing fees, and merchants cannot pay to influence rankings. However, the industry generally believes this is only a temporary strategy. Shopping Research uses a specially trained GPT-5 mini model capable of reading product pages, specifications sheets, and trusted reviews—an infrastructure that can seamlessly connect to an ad system.

OpenAI has already established partnerships with retailers such as Walmart, Target, and Etsy, allowing users to search for and purchase products from these platforms within ChatGPT. While this currently constitutes free traffic referral, once the ad system goes live, these partnerships can be immediately converted into paid promotion channels.

During Black Friday 2025, e-commerce website traffic driven by ChatGPT increased by 28% year-on-year, and retail website traffic from AI surged by 670% on Cyber Monday. If OpenAI can insert ads into this traffic, even with a conversion rate only half that of traditional e-commerce, the revenue will be substantial.

However, to resist the impact on its shopping gateway and ad revenue, Amazon blocked AI crawlers from multiple companies including ChatGPT, Meta, and Google shortly before Christmas, preventing them from scraping product information.

At the same time, Amazon launched its self-developed AI shopping assistant Rufus to defend the platform’s gateway and respond to changes brought by AI search.

To further explore its ad business, OpenAI poached ad executive Shivakumar Venkataraman from Google, who previously led multiple core projects in Google’s search advertising division.

In addition, OpenAI has posted multiple ad-related positions on Linkedin, including Ad Engineer, Ad Product Manager, and Ad Strategy Analyst.

These moves all indicate that for OpenAI, ads are no longer a question of "whether to do it" but "how to do it."

Driving OpenAI’s decision are not only cost pressures but also competitive pressures. Google announced in 2025 that it will introduce ads into Gemini, with a planned official launch in 2026. This poses a direct threat to OpenAI. If competitors are using ads to reduce user costs and increase revenue, OpenAI will find it difficult to remain an exception.

Perplexity AI has already launched a "Sponsored Answers" feature, clearly labeling sponsored content in search results. While this feature has sparked some user dissatisfaction, it has also proven the feasibility of AI search ads. OpenAI clearly does not want to fall behind in this race.

In terms of market size, the global digital advertising market generates over $1 trillion in annual revenue, with Google and Meta capturing most of the share. If OpenAI can enter this market through ChatGPT, even capturing just 5% of the share would bring in $50 billion in annual revenue—enough to support the company’s long-term operations.

But OpenAI also faces the challenge of user trust. Once AI responses begin to incorporate commercial interests, how will users judge whether a recommendation is based on objective analysis or advertiser payments? When ChatGPT recommends a product, users will wonder if it’s because the product is truly good or because the brand paid for it.

There is no easy answer to this question.

OpenAI is now attempting to walk a tightrope between "intellectual neutrality" and "financial self-rescue." If this balance fails, ChatGPT may transform from an "AI assistant" into a "product-promoting chatbot."

III. AI Search Eats into Traditional Keyword Search

Undeniably, AI search is eroding traditional keyword search.

The market share of traditional search engines is declining at a visible rate. According to a 2025 search behavior research report released by HigherVisibility, Google’s share in general information search dropped from 73% in February 2025 to 66.9% in August—a 6.1% decline in six months.

Changes in the Chinese market are even more drastic. Baidu’s mobile search market share plummeted from 94.72% in 2021 to 58.6% in the third quarter of 2025.

Data from QuestMobile shows that Baidu’s media status index fell to ninth place, ranking behind Douyin, Taobao, WeChat, Kuaishou, and Xiaohongshu. Furthermore, Baidu’s online marketing revenue dropped by 18% year-on-year in the third quarter of 2025—its sixth consecutive quarterly decline.

However, these users are not abandoning search altogether; instead, they are shifting their search behavior to AI tools.

According to the "2025 China AI Terminal Ecosystem Development Research Report" released by QuestMobile, as of October 2025, the number of mobile AI users in China reached 720 million, accounting for 50% of the total number of Chinese netizens. Daily AI search requests exceeded 2.8 billion, with commercial queries accounting for 41%.

A survey of 1,000 users by the German digital industry association bitkom found that 50% of respondents sometimes use AI chatbots to find information instead of traditional search engines. Among the 16-29 age group, 5% rely entirely on AI search, 11% mainly use AI, and 20% use AI and traditional search engines equally.

In other words, more than one-third of young people already use AI as their primary or sole source of information.

Even more telling is the change in zero-click rates. A zero-click rate refers to when a user searches for a keyword but does not click on any links, instead initiating a new search or closing the page.

Data from SparkToro shows that Google’s search zero-click rate has reached 58.5%, and as high as 77.2% on mobile devices. This means that more than half of all searches no longer generate clicks—users leave after seeing search results and summaries. The core ad model of traditional search engines is failing.

AI search delivers higher conversion rates. The large model search conversion rate refers to the proportion of users whose needs are directly met from the moment they initiate a search to the end. ChatGPT’s conversion rate is six times that of Google Search, as users are more willing to trust direct answers provided by AI.

Behind this transformation is AI search addressing long-standing core pain points of traditional search.

Traditional search only provides lists of links—users need to open a dozen web pages one by one, and the quality of information varies across different sites, often contradicting each other, requiring users to verify authenticity themselves. A simple question often takes a long time to piece together a complete answer, and the cycle of "search → click → read → return → search again" frustrates users.

AI search directly provides complete answers, eliminating the need to filter links, click through them one by one, and piece together information. According to a joint survey by OpenAI and Harvard University titled "How People Use ChatGPT," nearly 35% of users cited "obtaining direct answers" as the top reason for using AI search instead of Google.

AI search supports multi-turn conversations, which further enriches search results. If the first question is unclear, users can immediately follow up, add conditions, or adjust directions, gradually clarifying their needs like conversing with a real person—instead of repeatedly modifying keywords to search again. AI can remember context to form coherent conversations, while traditional search treats users as "new users" every time.

The "AI Search User Behavior Research" shows that 72% of AI search sessions include more than two turns of conversation, as this interaction method is more in line with natural human thinking habits. In traditional keyword search, there is no concept of "continuity"—each search only relies on the current keyword.

What users value more is that AI answers are not driven by commercial interests and are free from the hassle of paid ranking. The ad model of traditional search engines has long been criticized—Baidu’s search results are dominated by paid ranking ads on the first few pages, burying real information, and users need to scroll through multiple screens to find non-ad content.

Although AI can also make mistakes, at least it is not "who pays, who ranks first."

The convenience of voice interaction is another important factor. Typing on mobile devices is slow and error-prone, and entering long questions is a poor experience. Voice input is 3-4 times faster than typing, making it particularly suitable for expressing complex questions. Research by Geokeji shows that 58% of users in AI search apps use voice input, while in traditional keyword search, even with voice input functionality, the usage rate remains less than 20%.

The average length of voice search queries is 2.3 times that of text searches—because voice input is more convenient, users can express their needs more completely, improving search accuracy. In scenarios such as driving, cooking, taking care of children, or exercising, when hands are occupied, voice search becomes the only option. These high-frequency scenarios have helped users develop a new habit of "just speaking when in need."

Gartner predicts that global search engine traffic will plummet by 25% in 2026. This is not a simple shift in market share but a revolution in interaction methods.

From "typing keywords" to "voicing complete questions," from "filtering a dozen links" to "obtaining a direct answer," from "single queries" to "multi-turn conversations," the "keyword matching + paid ranking" model that traditional search engines rely on for survival is being rapidly replaced by the "voice conversation + direct answer" model based on large language models.

Therefore, traditional search companies, like ChatGPT, are caught in a dilemma. Adapting to the new interaction method means their proven business model is destined to be disrupted; yet staying in their comfort zone means watching the size of that zone shrink visibly. In this transitional period where the new technological revolution is imminent, they can only be contradictory and hesitant—while firmly walking on the tightrope, with a similarly hesitant ChatGPT on the other side.