AI writes 19,500 EU tobacco consultation responses

Roughly 30% of the 65,000 responses submitted to the European Commission's public consultation on tobacco and nicotine rules were likely written by artificial intelligence, according to an investigation by Dutch broadcasters NOS and Pointer, reported by Nieuwslens. In the Netherlands, the share was higher: about 71% of Dutch submissions were flagged as almost certainly fully AI-generated, assessed using the detection tool Pangram. The investigation traces the trail toward Philip Morris International.

The numbers matter because EU consultations are not opinion polls. They are the Commission's primary method for collecting evidence and public reaction before it drafts legislation (Commission Better Regulation). Synthetic responses don't win a consultation. They distort the record the Commission later cites when it sends a proposal to the European Parliament (the EU's directly elected legislature) and the Council (where national governments negotiate).

Why Tobacco Is the Test Case

The consultation targets the EU's biggest revision of nicotine regulation in over a decade. The current Tobacco Products Directive (Directive 2014/40/EU, the EU's main law governing cigarettes and related products) was written in 2014, before vapes, heated tobacco devices and nicotine pouches were widely sold. The revision is expected to address flavours, nicotine limits, packaging and online sales (Tobacco Intelligence).

That gap creates a clear incentive. Companies with large "smoke-free" product lines benefit if EU rules stay loose for newer nicotine products. Philip Morris International publicly urged Commission President Ursula von der Leyen to include tobacco in the EU's competitiveness agenda (EU Reporter). Organised lobbying is legal. Mass-producing thousands of individual-looking comments through a corporate tool is something different: it makes lobbying look like citizen pressure.

A watchdog report flagging rising nicotine-pouch use among young Europeans adds political charge (Euronews). The WHO's Framework Convention on Tobacco Control specifically calls for protecting public-health policy from tobacco-industry influence (WHO FCTC). Tobacco already operates under heightened suspicion, which makes synthetic astroturfing in this consultation especially damaging to the process.

What the Commission Can and Cannot Do

A counterargument deserves a fair hearing. A Brussels Times opinion piece warns against letting the Commission decide whose voice counts: AI can help ordinary citizens navigate dense EU questionnaires, translate their responses, or lower the barrier to participation (The Brussels Times). Banning all AI-assisted submissions could favour professional lobbyists who don't need the help.

But a consultation flooded by centrally generated text no longer measures what citizens think. It measures who owns the best campaign machine. That is a different problem from individual citizens using AI to polish their own views.

The Commission's most realistic option is to cluster, de-duplicate and separately label coordinated or synthetic responses when it publishes summaries. Existing tools don't cover this gap. The EU's Transparency Register tracks formal meetings between lobbyists and officials (Transparency Register), but it catches the scheduled meeting, not the manufactured crowd. The AI Act creates transparency obligations for some AI-generated content (AI Act), but the NOS/Pointer investigation found nothing in it that automatically voids AI-assisted consultation submissions.

The Vulnerability Is Structural

The weakness extends beyond nicotine. Any future EU consultation with high commercial stakes can be scaled the same way. The more technical the questionnaire, the more useful a text generator becomes and the harder it is to tell where the comments came from.

Two things remain unresolved. The Commission has not publicly explained whether it identified or discounted suspected synthetic responses in this consultation. And the NOS/Pointer investigation, while credible enough to trigger Dutch and wider European coverage, has not been independently replicated with access to raw consultation data and validated detection methods.

The open consultation was built as a democratic input channel (Have Your Say). Whether that channel still functions when the best-resourced actor can generate thousands of authentic-sounding voices is a question the Commission will need to answer before this directive reaches lawmakers.