The productivity gains that UK businesses have credited to generative AI have a mirror side that rarely makes the enterprise slide decks, detailed in a Guardian investigation published Saturday: British public bodies — local authorities, MPs’ offices, courts, schools and transport agencies — report being swamped by chatbot-generated complaints, freedom-of-information requests and legal challenges that take seconds to make and hours to properly answer. Commenters have labelled the phenomenon “AI clog”, and researchers studying it at the Hertie School’s centre for digital governance, the Cooperative AI Foundation and GovAI call the tactic “agentic flooding”.
The picture from inside is blunt. “Half the emails we receive are some form of AI,” Hollie Wickens, a senior caseworker in a Labour MP’s office, told the Guardian, adding that the volume has made requests harder to triage despite a decade in the role. Deborah Evans, chief executive of Lawyers in Local Government, which represents 5,600 local authority lawyers, says complaints that averaged about two pages before chatbots are now “10 times longer” — and the worst single one she is aware of ran to roughly 200 pages. Alessandro Georgiou, leader of Enfield council, reports a significant uptick in parking appeals written with AI; the leader of Enfield council is far from alone, as the GMB union says casework in MPs’ offices that keeps its own figures has “doubled or more”, leaving staff, in one office manager’s words, “exhausted” and “working more hours than they are paid”.
The numbers are public, and they are going the wrong way
This is not anecdote-only terrain. Transport for London received a record 4,769 freedom-of-information requests in 2025-26, a 5% rise on the previous 2023-24 peak, and the minutes of TfL’s audit and assurance committee record a “noted increase” in appeals that appear AI-generated, “which often contain highly technical and often inaccurate assertions and assumptions”. Employment tribunal claims opened between April and June soared 28% year-on-year to 14,000 single-claimant cases, with 70,000 open single claims by the end of June — a backlog one start-up founder, fighting an AI-polished dismissal claim, expects to cost him £30,000 in legal fees before it is even heard.
UK regulators saw this coming. In May the Information Commissioner’s Office published guidance on handling AI-generated FOI requests — the practical problems it names match the Guardian’s reporting almost clause for clause: requests that are “very long, repetitive and contain unrelated comments or misdirected demands”, misquote the law, or require significant clarification before they can even be processed. As GIP Digital Watch reported, the ICO’s position is that such requests still must be answered: the guidance is explicitly about helping teams “apply existing FOI principles consistently, regardless of how a request is created”, not adding new legal duties. Civil Service World’s coverage frames the same dilemma for stretched authorities — there is no extra funding line that grows when AI multiplies complaints. “Local authorities have one pot of money and don’t get extra money because of more complaints,” as Evans put it.
Why the AI-written complaint is structurally different
The asymmetry is the story. A chatbot turns a task that took hours of research and drafting into seconds of prompting, but the receiving institution still needs a trained human to read, verify and answer — and AI text, as the Guardian’s sources describe it, is simultaneously more to read and worse to read: legalistic, padded with bullet points, citing great case law from Canada that has no bearing on the jurisdiction, occasionally hallucinated outright. Jim Moore, an employee relations expert, told the Guardian that many AI-generated complaints “miss out on the most important point of a complaint — what resolution are you looking for?”, being worded to sound intimidating instead. The ICO, to its credit, has not taken the easy path of dismissing sloppy AI requests — its caution is that requests appearing AI-drafted but containing inaccuracies still must be answered by humans.
The responses proposed so far form a spectrum from template design to outright friction. Some ombudsman services are imposing word limits; Wiltshire council is exploring AI tools to summarise long emails and draft initial replies — in effect meeting machine output with machine processing, which critics warn means “losing the human touch”. Researchers cited in the Guardian go further, floating fees for submitting requests or blocking bots from government websites entirely. Each of those options trades away some access to justice in exchange for staff survival, which is precisely why none has been adopted at scale.
The UK is not an outlier — it is the largest lab that has run the experiment longest. ChatGPT turned four this month, and the UK’s AI adoption is among the strongest in Europe, so what London’s councils are absorbing this year, every English-speaking public service with a web form will absorb next. That includes New Zealand, where public agencies face identical economics — fixed budgets, statutory response deadlines, and a citizenry with the same free chatbots (our own coverage of NZ Police’s AI-driven back-office redesign and Wellington’s Deloitte report on AI-written work being barely detectable in output both point the same direction). A complaint that arrives with fabricated case law is not a productivity gain being shared; it is a cost being transferred from the person with the chatbot to the public budget. The honest accounting is that generative AI’s productivity story, told economy-wide, only balances if the human hours it consumes on the receiving end are counted — and right now, in the filing cabinets of UK public bodies, they are not being counted by anyone.
FAQ
What is “AI clog”? Commenters’ term for the flood of AI-generated complaints, FOI requests and appeals that public bodies must still process, named in the Guardian’s November 2026 investigation. Researchers call the deliberate version “agentic flooding”.
Do public bodies have to answer AI-generated requests? Yes. The UK ICO is explicit that its guidance adds no new law: FOI duties apply regardless of how a request was created, and requests that look AI-drafted still require human answers.
Is this a UK-only problem? No. The UK is simply early — high chatbot adoption plus statutory response deadlines make the volume visible first. Any public body with online forms faces the same dynamics.
