The new paperwork crisis starts with an odd imbalance. A person can ask a chatbot to draft a formal complaint, add references to laws and rewrite it in a more assertive tone—often in moments. But the public employee who receives it may need to spend hours untangling what it actually says, checking whether the law is relevant and figuring out what the writer wants.
A November 8 investigation by The Guardian describes that problem across UK local authorities, members of parliament’s offices, employers and other institutions. The phenomenon is sometimes called “AI clog” or “agentic flooding.” The phrases are new; the underlying imbalance is not. Digital tools have made producing a request cheap while leaving much of the responsibility for answering it in human hands.
This matters far beyond the UK. It exposes a design problem at the center of the AI productivity debate: what happens when technology makes it easier to demand someone else’s time than to save it?
The complaint that grew from two pages to twenty
One detail from the Guardian’s reporting captures the problem. According to Deborah Evans of Lawyers in Local Government, complaints that once averaged around two pages are now commonly much longer, with some reaching approximately twenty pages. She cited an extreme example running to about 200 pages.
That does not establish a national average for every complaint in Britain, nor does length prove a message was written by AI. But it illustrates what happens when generating more polished words becomes almost effortless.
A long document can hide a surprisingly small underlying issue. A missed collection, disputed parking charge or planning objection may arrive surrounded by arguments, boilerplate, incorrect legal citations and repeated demands. The recipient must still separate the actual request from the surrounding noise.
The problem becomes especially sharp when the institution has a legal obligation to respond. Public-sector workers cannot simply delete an awkward message because it sounds like a chatbot. People have genuine rights to complain, request information and challenge official decisions—even if they need help expressing themselves.
The other side of the productivity equation
The person sending a letter may have good reasons for using AI. Perhaps they are not confident writing formal English. They might have a disability, limited time, little experience of legal processes or no money for professional advice. A chatbot can help them organize facts and produce a clear request.
Those are meaningful accessibility benefits. It would be a mistake to frame every AI-assisted complaint as malicious or frivolous.
But productivity gains can move costs rather than eliminate them. If one person saves two hours of drafting but creates five hours of work for the recipient, the overall system may become less efficient. More importantly, there is no guarantee the generated letter will be accurate. A chatbot can cite a statute from the wrong country or phrase an uncertain allegation as a settled legal conclusion.
Human officials then have to check precisely those details. A confident tone increases neither reliability nor legal weight.
When automation collides with the right to be heard
The Guardian reports concerns from local-government lawyers and parliamentarians’ staff, as well as observations from the UK’s Information Commissioner’s Office about unusually long, repetitive or misdirected information requests. These are reported experiences and institutional observations, not evidence that every growth trend in correspondence was caused by AI.
Transport for London, for example, recorded 4,769 freedom-of-information requests in 2025–26 and reported an increase in appeals that appeared to be AI-generated. Its figures describe requests to that organization, not all UK public bodies. And an apparent AI writing style is not a reliable forensic test of authorship.
That uncertainty is part of the problem. A system that assumes machine-like writing equals bad faith risks harming the very people it is supposed to serve. Equally, a system that treats every generated paragraph as requiring the same attention can be overwhelmed by volume.
In the middle are public employees and ordinary citizens competing for a limited resource: human attention.
Could AI answer the AI-generated complaints?
The tempting response is to put another chatbot on the receiving end. An automated assistant could group duplicate requests, extract the specific action demanded and draft routine responses. Used carefully, such tools might reduce backlogs.
But a fully automated exchange also creates new failure modes. One system could misunderstand the request; the other could confidently misread the answer; both might continue producing increasingly elaborate documents. Legal deadlines, privacy obligations and accountability still belong to the organization, not to a statistical model.
There are more modest possibilities worth testing: structured digital forms that ask for the requested resolution, plain-language character limits where lawful, machine-assisted triage reviewed by staff, and explicit prompts to identify the key facts and supporting evidence.
None of these fixes should strip away a person’s ability to submit complex evidence when complexity is necessary. A serious complaint about abuse, discrimination or service failure should not be rejected because it does not fit neatly into a text box.
The distinction between easier speech and manufactured volume
What makes this issue important is that the same technology can both democratize access and enable overload. One citizen using AI to challenge a wrongful decision is not the same as an organized campaign automatically generating thousands of near-identical submissions. The effect on institutions can look similar from the outside; the appropriate response may be very different.
The solution therefore cannot be merely “ban AI-written letters.” Such a rule would be hard to enforce fairly and could penalize people who depend on writing assistance. The more useful question is whether systems can distinguish a clear, actionable request from a flood of repetitive or unsupported claims without judging people by their writing style.
The real bottleneck is still human
AI has dramatically reduced the cost of composing text. It has not removed the need for judgment, accountability, evidence-checking or empathy. Those requirements often remain concentrated in the same overworked teams that handled correspondence before chatbots existed.
The result is a strange new kind of congestion: not a shortage of words, but an excess of them. Perhaps the next productivity breakthrough will not be another tool that writes longer letters. It will be a way to help both sides get to the important question faster—and make sure someone can still give a meaningful answer.