“AI clog”: British public bodies swamped by chatbot complaints as submissions grow tenfold

Writing a complaint to a council or objecting to a development used to be difficult. You had to understand the issue, find the right form and strike a formal tone. That acted as a natural filter. ChatGPT, Claude, Gemini and Copilot have removed it. Now everyone has a free legal assistant, and British public bodies are calling the consequences “AI clog”: a blockage caused by AI.

The numbers: seconds to write a letter, hours to answer it

Holly Wickens, a staff member in a Labour MP’s office, says half of incoming letters are written with AI in one way or another. According to Deborah Evans, head of Lawyers in Local Government, the average complaint to a local authority used to run to about two pages; now most are ten times as long. The most burdensome example she knows of runs to around 200 pages.

Workloads in MPs’ offices have doubled or more, according to the GMB union. In offices that keep their own figures, staff are working overtime. Transport for London received a record 4,769 freedom of information requests in 2025–26 and reports an increase in appeals that appear to be AI-generated: they contain numerous technical assertions, often incorrect. Claims filed with employment tribunals in a quarter rose 28% year on year, while the backlog of open cases had reached 70,000 by June.

Researchers at the Hertie School, Cooperative AI Foundation and GovAI call this “agentic flooding”. The asymmetry is simple: creating a submission takes seconds; dealing with it takes hours.

How AI complaints differ from human-written ones

Council lawyers describe some telltale features. The letters are verbose, use bulleted lists and cite legislation. The model tends to turn any complaint into a legal claim. It brings in precedents, but not always from the right jurisdiction: it might cite an excellent court ruling from Canada. Sometimes it simply invents references.

Employment relations expert Jim Moore adds an interesting detail. These complaints sound threatening but often omit the most important point: what the complainant actually wants. Another problem is that assistants tend to agree with users, giving people inflated confidence in their chances of winning. One startup founder expects a challenge to a dismissal to cost him £30,000 in legal fees.

Meanwhile, the regulator ICO points out that even a request showing signs of AI use and containing errors must be processed manually. It cannot simply be dismissed.

What authorities are doing, and what it means

The measures so far are basic. Ombudsmen are introducing word limits and online templates that restrict the volume of attachments. One MP’s office has stopped responding to automated “write to your MP” campaigns. Researchers propose tougher options: making submissions harder, charging for requests and blocking bots. Some councils, including Wiltshire, are exploring AI to summarise long letters and draft replies. The result is machine versus machine, and experts warn of a loss of human contact.

For product builders, the lesson is practical. If your service accepts user submissions, whether for support, refunds, moderation or compliance, expect the same wave. It is worth planning ahead for structured forms with a mandatory “what do you want?” field, length limits, LLM-assisted triage and deduplication. This also creates a market for products that help organisations filter incoming submissions.

There are limitations, too: the evidence largely comes from industry accounts and interviews, rather than rigorous measurements of the share of AI-written text. Identifying a generated letter by its style is unreliable. But the trend is clear: when the cost of sending falls almost to zero, recipients have to defend themselves.