# Opinion: Public consultations must count people, not submissions

URL: https://ai-pulse.aglaganova.com/blog/opinion-public-consultations-must-count-people-not-submissions
Published: 2026-10-29
Author: Roman Aglaganov, Full-stack developer
Category: Opinion
Region: Europe
Tags: regulation, deepfakes

> Bots wrote three in four responses to a UK health consultation, and officials only caught them because the bots were clumsy. Governments should stop treating response counts as a vote, and protect the process before the next campaign is smarter.

![Illustration: a black ballot box outline with a stream of identical grey slips pouring towards it, and one hand-drawn red slip among them](https://ai-pulse.aglaganova.com/api/media/file/opinion-consultations-1.png)

A study published this week found that 90,835 of the 118,756 responses to a 2023 UK consultation on disposable vapes came from bots, nearly all opposing a ban. The Department of Health caught them, removed them and went ahead. Officials say their controls worked. They did, this time, because the campaign was crude: duplicated text, sent fast, from a small number of internet addresses.

## The next one will be better

Today’s AI tools can write thousands of different, plausible responses, each in a slightly different voice, sent slowly from many addresses. A campaign like that would be far harder to spot. As the study’s lead author, Nason Maani, put it, “what we can’t catch, we don’t know about.” And most bodies that run consultations, from local councils to small regulators, have nothing like the screening a central government department can afford.

## Stop counting

The deeper fix is to change what consultations are for. Responses are often reported as percentages for and against, as if the exercise were a poll. It is not, and a self-selected online form can never be a reliable measure of public opinion. Its value is in the arguments and evidence people bring: the shopkeeper who explains how a rule would hit her business, the doctor who has seen the harm. Ten thousand copies of the same tick-box answer add nothing that one response does not.

If departments weighed arguments rather than counted heads, mass campaigns, human or automated, would lose most of their point. Where governments do want a measure of opinion, they should commission proper surveys with sampled, verified respondents.

## Protect what remains

Some safeguards still make sense. Departments should publish how many responses they rejected and why, so that interference becomes visible. They should share detection methods and suspicious patterns with each other and with councils. Asking organisations to identify themselves, and treating identical text as one submission, costs genuine respondents little.

The think tank Demos warns, rightly, that every barrier to bots also makes it harder for real people to take part. That is why the answer cannot only be more checks at the door. It has to be a process in which flooding the inbox simply does not work.

## Sources

- [The Guardian: AI bots flooded 2023 UK government consultation on disposable vapes ban](https://www.theguardian.com/society/2026/oct/27/ai-bots-flooded-2023-uk-government-consultation-disposable-vapes-ban)
- [BMJ Group: 3 out of 4 responses to UK smoke-free generation consultation were AI bot-generated](https://bmjgroup.com/3-out-of-4-responses-to-uk-smoke-free-generation-consultation-were-ai-bot-generated/)