A consultation overwhelmed by automated responses
A UK government consultation on tobacco and vaping policy was flooded with machine-generated submissions in what researchers describe as a large-scale attempt to influence a formal public-health process.
According to research published in the journal Tobacco Control, the Department of Health and Social Care received 118,756 responses to its 2023 consultation on the tobacco and vapes bill. It identified 90,835 of them, or 76%, as having been submitted by bots.
The consultation included proposals from the Conservative government led by Rishi Sunak, including a ban on disposable vapes. Nearly all of the automated submissions addressed the same question and opposed that proposal. The Department of Health and Social Care excluded the bot responses from its decision-making.
The finding matters beyond the individual policy dispute. Public consultations are intended to give governments evidence about how proposed rules may affect citizens, businesses, health professionals and other stakeholders. Large volumes of artificial responses can make a small, coordinated campaign appear to be broad public opinion.
What the analysis found
The research was led by Dr Nason Maani of the University of Edinburgh and Professor Anna Gilmore of the University of Bath. The researchers said the episode appears to be the first documented case of large-scale automated interference in a formal public-health policy consultation.
Departmental bot-detection methods found that the submissions were often duplicated repeatedly and sent at high frequency from a small number of IP addresses. Of the 90,835 responses identified as automated, all but two answered the same consultation question.
That pattern gave officials a basis to distinguish the automated campaign from legitimate public participation. But the case also illustrates a challenge for consultation systems: detecting identical or highly repetitive submissions is more straightforward than identifying automated responses that are individually varied, more persuasive in tone or distributed across a larger number of technical sources.
Maani said the wider concern is that future campaigns could be more difficult to identify. If automated interference becomes less visible, policymakers may not know when apparent public sentiment has been manufactured.
Why the episode is significant
The volume involved was substantial. More than three quarters of all responses received by the consultation were identified as bot-generated. Had they been treated as ordinary submissions, the record could have given a sharply misleading impression of public opposition to the disposable-vape proposal.
The campaign did not change the immediate outcome. The Labour government led by Keir Starmer inherited the tobacco and vapes bill in 2024 and passed it into law, including a ban on businesses selling or supplying disposable vapes.
However, the episode raises a broader question about the resilience of democratic and administrative processes as generative AI and automated tools become easier to use. A consultation does not operate like an election, but both depend on institutions being able to distinguish real participation from coordinated manipulation.
The risk is not simply an excessive number of messages. Automated submissions can consume staff time, complicate data analysis and potentially make a policy debate look more polarized than it is. In a high-volume process, that can also make it harder for officials to identify detailed evidence from affected groups.
Detection is only part of the response
In this case, the Department of Health and Social Care used bot-detection methods and excluded the identified submissions. That response prevented the automated volume from being counted in the policy decision.
But the research points to an ongoing institutional problem rather than a one-off technical anomaly. Consultation systems need ways to accept broad public input while protecting the integrity of the resulting evidence. The balance is delicate: barriers designed to deter automation can also make it harder for legitimate participants to take part.
Caroline Cerny, deputy chief executive of Action on Smoking and Health, called the evidence concerning and said public-health authorities should be alert to the way technological advances can be used to undermine regulation. The tobacco industry has a history of seeking to weaken public-health rules, she said, though the reported research does not establish who was responsible for the automated submissions.
A warning for public institutions
The central lesson is not that consultations should be abandoned or that every coordinated response is invalid. Organized groups have long encouraged supporters to participate in public consultations. The distinction here is the scale and automated nature of the activity, and the attempt to create the appearance of mass participation through tens of thousands of machine-generated responses.
As automated systems become more capable of producing varied text, the methods used to evaluate consultation evidence will face increasing pressure. The UK case offers a clear example of why governments need to retain the ability to inspect patterns of submission, assess provenance and explain how they treat suspected manipulation.
The disposable-vape measure ultimately proceeded, but the consultation has become a test case for a much wider policy question: whether public institutions can preserve meaningful public input when the cost of manufacturing apparent public opinion falls sharply.
This article is based on reporting by The Guardian. Read the original article.
Originally published on theguardian.com
