TL;DR

Automated submissions made up 76% of the responses to the government’s 2023 consultation on its tobacco and vapes plans, according to research published on Tuesday 27 October and reported by the Guardian. Almost all of them targeted a single question and opposed the ban on disposable vapes. The health department (DHSC) caught and removed them, and the ban went ahead.

What the study found

The paper, published in Tobacco Control, comes from Edinburgh’s Dr Nason Maani and Bath’s Professor Anna Gilmore. It rests on the department’s own screening, which flagged 90,835 out of 118,756 submissions as bot-generated.

Every one of those automated replies except two answered one question in the “smoke free generation” consultation, which Rishi Sunak’s Conservative government ran. DHSC said the submissions were “often duplicated many times and sent at high frequency from a small number of IP addresses”. The authors say it “appears to be the first documented instance of large-scale automated interference in a formal public health policy consultation”.

The department left the flagged responses out of its analysis. After taking office in 2024, Labour carried the bill into law, disposable vapes ban included.

Why detection is not the end of it

Maani warned that later campaigns may be harder to spot, and that “what we can’t catch, we don’t know about”. His wider concern is that a few powerful interests could pass themselves off as a mass movement.

Azzurra Moores of the thinktank Demos made the trade-off plain. Any barrier raised against bots, she said, puts off genuine respondents too, while AI makes fake replies far harder to tell apart. Action on Smoking and Health called the findings “deeply concerning” given the tobacco industry’s record of opposing health regulation.

DHSC said its “robust controls” worked and that it will stay alert in future consultations.

Looking Forward

Here the bots were crude: duplicated text from a handful of IP addresses. In our view, the harder test is a campaign built with current language models that writes varied, plausible replies from many sources. Any UK department or regulator running an open consultation, including on AI policy itself, now has a documented case to plan against. That may mean identity checks, weighting by evidence rather than volume, or publishing how responses were screened.