Tobacco giant uses AI to sway regulation: thousands of responses auto-generated

Philip Morris International is using artificial intelligence to sway new European tobacco laws. An investigation by NOS and Pointer found that a large share of responses to the European Commission’s public consultation were auto-generated by the tobacco giant’s AI tool. The tactic has triggered sharp criticism from Brussels, the Dutch Parliament, and the medical community.

The European Commission is working on the biggest overhaul of EU tobacco rules in fifteen years, with a focus on vapes, e-cigarettes, heated tobacco, and nicotine pouches. Those products now account for roughly 40 percent of Philip Morris International’s revenue—giving the company a major financial stake in how the rules land.

Joint reporting by NOS and Pointer shows Philip Morris built a dedicated AI tool that lets citizens easily submit responses during the Commission’s public consultation. Researchers estimate that about 30 percent of more than 65,000 responses from fifteen EU countries were AI-written. In the Netherlands, that figure jumps to 71 percent.

The campaign runs under the slogan “Your Voice. Your Choice.” Posters in tobacco shops direct customers via QR code to an online questionnaire. After a few multiple-choice answers, an AI system automatically generates a first-person response tailored to the user.

Users can then submit that text almost instantly to the European Commission as an official comment. From the Netherlands alone, this happened more than seven hundred times, according to the investigation. Philip Morris says users fully own their submissions and can edit them before sending. The company calls the campaign a legitimate way to help citizens take part in the democratic process.

Typically, a European Commission consultation draws a few hundred to at most a thousand responses. This one has already racked up around 80,000.

Experts say that’s no coincidence. They call it a modern form of astroturfing: organized lobbying dressed up as spontaneous grassroots support. AI, they argue, makes these efforts far faster and vastly more scalable.

The Commission reacted unusually bluntly. A spokesperson said attempts to artificially influence public consultations are condemned. Brussels insists these processes must reflect real citizens’ views—not be steered by commercial campaigns.

The Commission adds that its analysis does account for coordinated campaigns and AI-generated submissions.

Outrage is also mounting in The Hague. Party leader Mirjam Bikker calls the tactic an assault on democracy. In her view, participation becomes “corrupt” when an industry channels citizens toward prepackaged positions.

Henri Bontenbal calls it “cheating,” saying the campaign speaks volumes about the company’s mindset.

The criticism taps into a broader debate over generative AI in lobbying and policymaking. While AI is often used to summarize documents or analyze proposals, this case shows the same tech can inflate the appearance of public support.

Dutch pulmonologists and health organizations have filed a complaint with the Dutch Advertising Code Committee. They argue the campaign amounts to illegal tobacco advertising and misleads consumers.

According to the doctors, the AI tool steers users toward arguments against stricter rules. Even users who indicate they support tougher regulations are, they say, still shown texts that highlight the benefits of alternative nicotine products.

The case highlights a wider shift: generative AI is rapidly becoming standard in political influence operations. Where lobby groups once relied on painstakingly organized endorsements, AI can now churn out thousands of unique, human-sounding messages in seconds.

That makes public consultations more vulnerable to large-scale influence campaigns. Although the European Commission says it can spot these patterns, experts increasingly question whether current consultation procedures can withstand the era of generative AI.

Policymakers now face a new test: how to keep civic participation representative when AI can mass-produce persuasive, personal-sounding responses?