Newsletters
Advertisements that have long appeared as "sponsored links" in internet search engines are now being embedded directly inside answers generated by artificial intelligence-powered search and chat systems, a shift that experts say is transforming the nature of digital advertising and the way users access information.
OpenAI announced Feb. 9 in a written statement that it had begun testing ads inside ChatGPT in the United States. The company said the ads shown during the test would be determined by factors such as the conversation topic, previous chat history, and prior ad interactions. It noted, for example, that a user searching for recipes could be shown ads for meal kits or grocery delivery services, and said users would be able to manage their ad preferences.
The company said advertisers would not be able to access users' conversation content or personal information, and would only receive anonymous, aggregate performance data such as total views and click counts. Legal and ethical debates over the practice continue, however.
In traditional search engines, a user types a query and sees sponsored links prominently placed alongside results, making it clear the content is an advertisement and allowing the user to make an informed choice. In next-generation AI systems, a user asks a question and receives a single, unified response. Advertising or sponsored content can be embedded within that unified answer, presented as a recommendation or additional information.
This shift is changing not just the format of ads, but the fundamental nature of digital advertising and the way users access information.
Koen Pauwels, a marketing professor at Northeastern University and editor-in-chief of the International Journal of Research in Marketing, evaluated the technical, social, and ethical dimensions of embedding ads in AI-generated responses for Anadolu Agency (AA).
Pauwels said that in traditional search engines, users reach information by browsing, comparing, and choosing from a list of results. With AI-generated responses, that process is reduced to a single answer.
"The user gets one synthesized answer instead of 10 links. If advertising is embedded in that answer, the user may not realize the recommendation in front of them reflects a commercial interest rather than a neutral analysis," Pauwels said. "In AI, commercial content and informational content appear in the same sentence, the same paragraph, sometimes even the same recommendation. Transparency is harder to achieve because AI's voice appears authoritative and impartial."
Unlike traditional search engines, AI-based systems can simultaneously remember previous conversations, analyze a user's interests, and build long-term context. This means ads can be shaped not just by a search keyword, but by a user's behavioral history, preference patterns, and even emotional tone.
"Conversation history is behaviorally richer than click history because it reveals not just what the user did but why they did it," Pauwels said. "When someone asks a chatbot about managing chronic pain, comparing mortgage options, or planning a family vacation, they reveal their intent, their context, and sometimes their vulnerability."
Pauwels pointed out that traditional search engines guess preferences based on actions, but AI systems can understand preferences directly from what users say. This makes ad targeting much more precise.
"Users should be skeptical of platforms that claim their AI is above commercial influences," Pauwels said. "The question that must always be asked is: Who is paying for the platform, and how does that affect what the model produces? Anthropic, the maker of Claude, for instance, is positioning itself around the absence of ads and took aim at rivals like ChatGPT and OpenAI, who plan to use advertising in Super Bowl commercials."
Pauwels emphasized that today's large language models do not clearly distinguish between "information" and "paid placement." This difference depends on how the system is built, how it finds information, and the platform's disclosure rules.
So-called "AI agents", systems capable of making decisions on behalf of users, are expected to book flights, handle grocery shopping, and automatically compare and select insurance options in the future.
"If AI agents are booking your flights, doing your grocery shopping and autonomously comparing your insurance options, the entire advertising process changes," Pauwels said. "The seller is no longer trying to persuade a consumer; it is trying to persuade an algorithm."
Pauwels said brands would start investing more in improving the training data and systems behind AI agents. "The consumer may gain in efficiency but could lose the moment of conscious choice," he said.
Pauwels pointed out that Google already dominates the search ad market. He said if AI assistants replace search engines, one or two AI platforms could become the main players. These companies would have more power over advertisers, publishers, and users, and would need new ways to measure if their AI's recommendations actually influence behavior.
"AI architecture eliminates visible alternatives," Pauwels said.
Pauwels said that placing ads in the answers from new AI search and chat systems raises several ethical issues, including transparency, the use of personal data, and the risk of manipulation.
He emphasized that users should know when commercial interests influence the information they receive.
"AI platforms collect information while providing it. Every query is a data point. Every session builds a behavioral profile," Pauwels said. "AI systems that model individual psychology can generate recommendations calibrated to a specific person's decision-making patterns, and for vulnerable population groups in particular, this can veer from persuasion toward exploitation."
Pauwels explained that the ad-supported internet is based on free access to information in return for personal data. He said AI systems are making this exchange more personal, and argued that users, regulators, and platforms need a clearer agreement about what is acceptable when trading personal data.
Set as preferred source
Set as preferred source
2 min read
