Britain’s Bureaucracy Can’t Escape AI’s Growing Paper Trail

🔍 Read the full analysis: Britain’s Bureaucracy Can’t Escape AI’s Growing Paper Trail on ThorstenMeyerAI.com

Written complaints, freedom-of-information requests and employment tribunal claims have risen across several UK services since ChatGPT launched in November 2022, according to a Sky News report published October 20. The report links the timing to the spread of AI-assisted writing, while cautioning that its evidence does not establish AI as the cause; the trend matters because many public services still depend on people submitting written applications, appeals and complaints.

Using data reviewed by Sky News technology correspondent Rowland Manthorpe, the report describes a similar upward pattern across channels including complaints to councils, GP practices, water companies and the Department for Work and Pensions (DWP). It says the increases begin after years of relatively little movement in some measures, with a gradual rise from late 2022 followed by faster growth. Each channel also has its own possible causes, the report notes, so the shared timing alone cannot explain the changes.

The report lists increases since ChatGPT launched of 144% for household complaints about water companies, 78% for complaints to English local authorities and 71% for written complaints to GP practices. It also gives increases of 64% for DWP complaints, 62% for individual employment tribunal claims, 49% for legal-services complaints, 35% for written NHS complaints, 34% for police complaints and 23% for public-service complaints. These are reported growth figures; the source material does not give the underlying totals or a common measurement window for every channel.

Manthorpe also tested 100 recent FOI requests with an AI-text detector. It classified 89% as fully AI-written and 9% as human, while a smaller 2019 sample was classified as entirely human. The report stresses that AI detectors are unreliable and that this small test cannot show whether AI caused the overall rise. Separately, a survey of civil servants cited by Sky News found that 58% said public AI-generated content significantly affected their work.

Paperwork Is Becoming a Capacity Test

The rise could affect how quickly public bodies handle requests, complaints and appeals. If each submission takes staff time to assess, a larger volume can add pressure to services even when the writing is easier to produce. Sky News reports that civil servants described being inundated by correspondence and said some interactions felt like arguing with a chatbot. Those accounts describe staff experience; they do not establish that every AI-assisted submission is low quality or unnecessary.

The public interest is not limited to administrative workload. According to the report, a substantial portion of complaints may be justified. At the DWP, the share of complaints upheld has fallen from the 50s to close to 40%, but the report says the much larger volume means the number upheld in absolute terms is still higher. Researcher Chris Schmitz, whose work is cited in the report, similarly argues that most cases he reviewed involved people claiming something to which they were entitled. These findings complicate any response that treats more submissions as simply spam.

The policy choice is whether to add barriers to written claims or redesign services so they can process more requests without relying on long, individually drafted essays. Either route could affect access: added friction may deter valid claims, while unchanged processes may struggle with growing workloads.

A Shift From Letters to AI Agents

In May, the Information Commissioner’s Office issued guidance after public authorities reported higher volumes and more complex FOI requests, some of which misquoted legislation or needed clarification. The ICO said an AI-written request remains a valid request. Authorities do not need to determine who—or what—wrote it, and they may ask for clarification when a request is unclear or badly formed.

The report also points to possible growth beyond people using chatbots to draft text. AI agents can take actions on a user’s behalf, potentially submitting applications or appeals automatically. Manthorpe demonstrates an agent service from Meta and raises examples such as appealing parking tickets or filing rail delay compensation claims. The report presents these as prospective pressures, not evidence that such automated submissions are already widespread across UK services.

Tom Loosemore, who founded the Government Digital Service and helped establish GOV.UK, argues that services should rely more on structured data and less on essays. His warning is that services dependent on lengthy written applications could struggle if AI makes submitting them nearly effortless. That is a view about future system capacity, not a measured forecast of when a particular service will fail.

““It’s like being bombarded with correspondence at speed.””

— Civil servants quoted in the Sky News report

AI’s Role Is Not Proven

The reported increases show a pattern across several channels, but they do not prove that AI caused it. Complaints can rise for service-specific reasons: the report points to water companies’ difficult recent years as one example. The supplied material also does not provide full underlying counts, consistent time windows or detailed methods for every growth figure, limiting comparisons between services.

The detector test is a small sample, and its classifications are not conclusive. It cannot establish how much of the wider growth consists of AI-written submissions, nor whether AI use made a particular complaint less valid. It is also unclear how many submissions are generated by agents acting automatically, rather than drafted by people using chatbots. The report describes that as an emerging possibility, not a quantified current trend.

Service Design Faces the Next Test

The immediate response from public bodies will be guided by existing rules, including the ICO’s position that AI-written FOI requests remain valid and can be clarified when necessary. The report does not identify a single national policy change or timetable for redesigning complaints, benefits or appeal systems. Further data on volumes, outcomes and processing times would help distinguish workload growth from changes in the number of people seeking valid redress.

The longer-term question is whether government services can replace open-ended written submissions with structured forms and data that capture what officials need to decide a case. Loosemore argues for that direction, while warning against measures that make it harder to submit claims. If AI agents become more common, agencies may face further pressure to adapt; how quickly that happens, and which services are most exposed, remains uncertain.

Key Questions

What did the Sky News report find?

It reported rising complaints, FOI requests and employment tribunal claims across several UK channels since ChatGPT launched in November 2022. The report says the trend is consistent across services, but does not prove AI caused it.

Does the report show that AI is responsible for the increase?

No. It identifies a pattern in the timing and describes a small detector test, but notes that AI detectors are unreliable and the evidence does not establish causation. Other factors may affect individual complaint channels.

Are AI-written FOI requests valid?

According to guidance from the Information Commissioner’s Office cited in the report, AI-written FOI requests are still valid requests. Authorities may seek clarification if a request is unclear or badly formed.

Are the additional complaints mostly frivolous?

The report does not characterize them all that way. It cites DWP data showing a lower share of complaints upheld, close to 40%, while saying the larger overall volume still means more upheld complaints in absolute terms. Researcher Chris Schmitz also says most cases in his review involved people claiming something they were entitled to.

What changes could public services make?

Tom Loosemore argues for services that use structured data rather than relying on lengthy written submissions. The report does not announce a government-wide redesign or set out a timetable, and the effects of any changes remain unknown.

Source: ThorstenMeyerAI.com