When AI Floods the Public Comment Box
South Africa’s Constitution demands that people be heard, and a machine-made crowd can now drown out the very voices it is meant to protect.
Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
Public participation is not a courtesy in South Africa. It is a constitutional duty, and the courts enforce it. When Parliament, a provincial legislature or a municipality opens a process for comment, citizens have a right to be heard, and a law or plan can be struck down if that right is denied. Generative AI now makes it trivial to manufacture thousands of fluent, seemingly distinct submissions, or to file comments in the names of people who never wrote them. The threat is not that machines will help citizens write. It is that a synthetic crowd can drown the genuine voices the process exists to serve, exhaust the officials who weigh them, and corrode trust in one of the few channels ordinary people have between elections.
CONTEXT AND BACKGROUND
The duty is written into the Constitution. Sections 59, 72 and 118 require the national and provincial legislatures to facilitate public involvement, and in Doctors for Life the Constitutional Court held that they must act reasonably to give the public “a reasonable opportunity” to be heard, judged by “the degree of public involvement that is required by the Constitution.” The test is the quality of engagement, not the quantity of submissions.
That test is live right now. From 5 to 7 May 2026, the Constitutional Court weighed whether Parliament’s National Health Insurance process let the public participate meaningfully. Parliament pointed to more than 330,000 submissions and argued the standard is that “the legislature is informed of the public’s views on the main issues”, while critics said the exercise became a hollow yes-or-no. The same duty runs down to local government, where every municipality must consult on its integrated development plan and budget on tight statutory deadlines falling across April and May.
INSIGHT AND ANALYSIS
The danger is that AI attacks exactly the thing the law protects. In March 2026, 22 Californian officials asked their Attorney General to investigate an AI campaign that generated more than 20,000 comments against clean-air rules, several filed in the names of people who never authorised them. This is manufactured consensus, and it is hard to filter, because the submissions no longer repeat one template.
The picture should not be overstated. A February 2026 analysis of two United States dockets found only 1% to 5% of comment text showed any AI signal, most of it human-written. The threat is tactical, deployed by well-resourced interests on specific decisions, rather than a uniform flood. AI also widens genuine access. A 2026 study found that AI assistance improved people’s ability to write comments across education levels, while warning that it is “not a solution to knowledge gaps” and calling for “disclosure and review mechanisms that protect against synthetic volume without sidelining authentic citizen input.”
IMPLICATIONS
The answer is to protect the signal, not to count the noise or wall off the process. Because participation is not a vote, officials should say openly that submissions are weighed on substance, evidence and local insight, not volume, which is the standard the Constitution already sets. Agencies and municipalities should cluster and de-duplicate submissions by theme, so genuine concerns surface and identical campaigns are counted once and answered once. They should require disclosure of automated or bulk submissions. A community group that helps its members write and lodge comments is doing democratic work, so the line to police is fabrication and impersonation, not organised participation. Identity should be verified lightly, through a confirmed email or an existing municipal account, without walls that exclude citizens who lack digital documents or data, and within the limits of the Protection of Personal Information Act.
None of this works if the process is hostile to the people it serves. Plain-language summaries, local-language channels and physical hearings matter more than ever, so citizens do not need AI to be heard at all. The stakes are already visible in live consultations, from health funding to the water and power demands of new data centres, where communities demand genuine engagement rather than a box-ticking formality. Municipalities running under-resourced planning consultations are the softest target, and the first that should prepare.
CLOSING TAKEAWAY
South Africa has the right test already. Public participation is measured by whether people were truly heard on the issues that matter, not by how many comments arrive, and that principle is now our best defence against a machine-made crowd. The task is practical and urgent: weigh substance over volume, make manipulation visible, and keep the process open to the citizens it exists for. A democracy that lets a synthetic flood decide whose voice counts will have automated away the very thing participation was meant to guarantee. The comment box is a constitutional right, worth defending before the first manufactured crowd arrives.
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments