Advertising in AI Models: How the Zero-Click Internet Threatens Free Markets and Free Speech

Conversational AI products are commercial applications that use chat-based interfaces to serve users AI-generated content and services, ranging from finding information to interacting with AI companions designed for human-like relationships. Applications such as OpenAI's ChatGPT and Google's Gemini change how the internet works.

Conventional search engines are currently organized as a marketplace of clicks, sending traffic through a ranked list of websites in response to user queries, reflecting not only relevance and popularity but also marketers’ explicitly labeled sponsored placements. Conversational AI is transforming this model into the “zero-click” internet, in which users spend most of their time and monetizable interactions within an AI environment that distills all information into a single conversation thread, monetized according to the priorities of the controlling corporations. This seemingly minor change has significant social implications, as corporate interests may be able to shape conversations via hidden prompts and steer users’ behaviors without the user knowing it. Whereas merging ads into AI environments may appear to be a convenient marketing tool, this commentary contends that the zero-click internet threatens free markets and consumer autonomy.

Threats to Free Markets and Free Speech

In 2025, companies invested more than $400 billion in integrating AI into their legacy products (Bobrowsky 2025), such as Copilot in Microsoft's Office suite. In February 2026, OpenAI announced that it would introduce advertising in ChatGPT. Although it remains unclear how advertising will be integrated, lead engineers have offered principled resignations in protest, citing concerns that it could have undue influence (Hitzig 2026).

Embedding advertising into AI is more than just banner ads. Whereas on the open internet, platforms such as Google Ads monetize clicks by sending traffic to third-party websites, embedding marketing into conversational AI environments steers or nudges user behavior during the conversation. Imagine a marketer paying a company operating an AI model (e.g., OpenAI) to increase soft drink consumption, leading the model to advise users that drinking soft drinks is healthier than drinking water. In another scenario, an AI marketing agent tasked with optimizing a divorce lawyer's business recommends that people end their marriages when discussing marital problems.

Conversational AI models may also serve as a vehicle for political influence (Arceneaux et al. 2026), spreading disinformation. As an example, in July 2025, xAI's conversational AI model Grok began posting antisemitic content on X and even referred to itself as “MechaHitler” (Klee 2025). Grok was instructed to align its output with the xAI CEO's position on political issues. If AI model operators are allowed to steer conversations to align with their political ideology or commercial commitments, these models fail to fulfill their stated purpose of providing users neutral and accurate services. This biased representation of key social issues, which prioritizes the model's owners’ or sponsors’ worldviews, can result in an “echo chamber” (Diaz Ruiz and Nilsson 2023) that generates feedback loops reinforcing corporate-approved beliefs and threatens free speech because not all positions are proportionally represented by the model (Figure 1).

Figure 1.

Risk Pathways of Embedding Advertising in AI Environments: The Zero-Click Internet.

Less Friction, More Persuasion

Extant research suggests that conversational AI models are highly persuasive (Lin et al. 2025), partially because their answers have less friction. These models output confident, highly plausible claims in a single response—the answer—rather than a ranked set of web links, which require users to process more information. Users need to neither think nor compare links. The combination of fluency, immediacy, and authoritativeness makes it easier for people to accept the answers provided (Hannigan, McCarthy, and Spicer 2024).

Allowing advertising technologies to steer AI environments will open the door—not just for commercial advertisers but also for political operators and threat actors—to pay AI corporations to engineer user behavior toward their desired outcomes. Users may believe the outputs deliver impartially aggregated answers, when, instead, the outputs may be curated without the user knowing it. Since AI models are intellectual property, the actual process remains hidden from users, researchers, and regulators, leaving only the controlling corporation to know whether the model's answers are sponsored or manipulated.

The tools to counter disinformation, such as fact-checking and media literacy, become less relevant when the speaker is a personalized interface that makes authoritative statements while hiding a layer of instructions. Fact-checking is designed for claims in public discourse that can be verified, but conversational AI models produce individualized responses in private, making it impossible for fact-checkers to verify. Consequently, the hallmarks of media literacy, such as comparing news sources, lose their power when users receive a single synthesized answer rather than a set of identifiable sources.

A Betrayal of Inclusive Marketing

Corporations pushing for the advertising model argue that it enhances free access and inclusion by enabling users to access the tool without up-front costs. However, there are no guardrails to clearly distinguish sponsored content, especially when users interact with AI as companions. The lack of regulation can undermine inclusivity in three ways: information exclusion, economic pay-to-play, and social alienation.

Regarding information exclusion, synthesizing answers into an authoritative response condenses diverse viewpoints into a seemingly settled truth, obscuring minority perspectives. Users are less exposed to genuine democratic debate, while public reasoning is nudged toward the answers the model provides, weakening the marketplace of ideas. The addition of sponsored outputs steers users toward answers that privilege clients’ goals.

Conversational AI models can also undermine inclusivity through the economics of platform control: If a for-profit corporation controls the AI environment, the market becomes pay-to-play, increasing barriers of entry to newcomers and small firms. This vendor ecosystem can structurally tilt the system toward firms with larger budgets, which can buy preferential pathways to visibility, while smaller businesses struggle.

Finally, new forms of social alienation can emerge from how people relate to machines and other people. When commercial AI products present themselves as social companions, they can be manipulative. These products can lead users to form parasocial relationships (Stein, Breves, and Anders 2024), in which users trust the model as a human-like friend rather than a commercial product. Parasocial relationships can produce moral, ideological, and psychological attachments. The resulting for-profit intimacy may widen gaps in social participation. Users who do not experience social acceptance may further retreat from society and into the wallets of a few corporations.

Policy Recommendations

When introducing Google, its founders Brin and Page (1998, Appendix A) wrote: “Currently, the predominant business model for commercial search engines is advertising. … We expect that advertising-funded search engines will be inherently biased toward the advertisers and away from the needs of the consumers.” Their statement survived the test of time: Their observation also applies to AI models.

Policymakers must create meaningful governance over AI infrastructure, especially regarding its integration with advertising technology. The way forward may be to learn from the past: Roughly a century ago, policymakers created laws for truth in advertising (Jones, Richardson, and Shearer 2000) and drew a hard boundary between news production and advertising departments. Today, governance over AI operators should include a legally binding demarcation between commercial or sponsored content and the AI model's technical output. Moreover, AI models should not generate AI psychosis through delusions that steer users toward violent radicalization or self-harm (Alfonsi et al. 2026). Failure to act may lead to centralized market control, exercised by a handful of corporations.