AI-Drafted Complaints and Requests: What UK Public Bodies Are Receiving and How to Handle Them

Title: AI-Drafted Complaints and Requests: What UK Public Bodies Are Receiving and How to Handle Them

Date: 16 September 2026

Type: Paper

Author: SAASiQ (contact@saasiq.ai)

Word count: 2831 words

Reading time: 11 min

Published: 16-09-2026

Tags: #PublicSector #HousingOmbudsman #FreedomOfInformation #ComplaintHandling #AIGovernance #LocalGovernment

A study of 84 cases in 11 jurisdictions, reported by TechCrunch on 10 September, found that AI tools are driving up the number of complaints, petitions and applications reaching public services, and that most of the extra submissions come from people entitled to make them. Its UK example is the Housing Ombudsman, where the study records complaints rising from 2,600 in 2022 to more than 7,000 in 2025. This paper covers what is arriving, how it is produced, which bodies it reaches, the duties that still apply, and what a public body can do without restricting legitimate access.

What the research found

The study is by Chris Schmitz, with Lewis Hammond of the Cooperative AI Foundation and Alan Chan of GovAI, and is due to be presented at the AI, Ethics and Society conference in October. The authors call the effect 'agentic flooding'. To count as a case, a surge had to pass three tests: a plausible way in which AI cut the cost of making the request, evidence that demand actually shifted, and an explicit attribution to AI by officials or a credible third party.

Of the 84 cases, 50 showed a rise in the number of submissions and 76 a change in the submissions themselves, with 42 showing both, according to Dataconomy's account of the paper. Courts and legal services made up 19 cases, regulatory complaints 10, and welfare and social security 9. In the US, complaints to the Consumer Financial Protection Bureau reached five times their 2022 level by 2025. Brazil saw a rise in judicial petitions and Germany in parliamentary petitions, where one submission ran to more than 4,000 pages.

The main finding is that most of the extra volume is legitimate. 'The vast majority of cases we find are people who are entitled to claim for something, claiming for that thing,' Schmitz told TechCrunch. The administrative effort of writing a formal letter had kept eligible people from applying or complaining, and AI has removed much of it. 'People are finding out that this is something one can do, and incrementally, it is just getting easier,' he said, and he suggested this 'could be the moment to say, we need to rethink pretty much everything about how this process looks'.

The Housing Ombudsman's own figures

The Housing Ombudsman handles complaints about social landlords in England, meaning councils and housing associations. Its published series counts determinations, the decisions it makes after investigating a complaint, by financial year: 2,138 in 2021-22, 7,082 in 2024-25 (up 30 per cent on the year before) and 8,783 in 2025-26. The study's figures of 2,600 and just over 7,000 do not map exactly onto that series, although the direction is the same.

The 2025-26 annual complaints review, published on 15 September, reports 20,016 findings, up 39 per cent from 14,420, and an uphold rate of 62 per cent, down from 71 per cent. It also adds a new complaint category covering artificial intelligence. 'As use of AI increases for both landlords and residents, we anticipate more complaints in this category,' the Ombudsman said, according to Inside Housing, and it encouraged landlords to train staff on appropriate use of AI and its risks.

Other things changed over the same years. The Complaint Handling Code became statutory for social landlords in April 2024, and the Ombudsman's 2024-25 review said the rise in findings about property condition showed residents were 'more aware than ever about their rights to redress'. The study counts cases where AI was named as a cause. It does not claim AI accounts for the whole increase.

The same pattern elsewhere in the UK

The Property Ombudsman, which covers letting and estate agents, said on 29 July that complaints in the first half of 2026 were more than 50 per cent higher than in the first half of 2024. It listed AI among the causes, alongside wider awareness of consumer rights and the Renters' Rights Act, and said AI tools 'have enabled consumers to produce detailed submissions that may previously have been difficult to articulate'. Writing in Inside Housing in June, Lesley Horton, the Chief Property Ombudsman, said generative AI 'can present information with great confidence even when the advice is incomplete or lacking context'.

Some letting agents want complainants to declare when they have used AI, and Greg Tsuman, a past president of ARLA Propertymark, proposed it in August. The Property Ombudsman declined on 14 August, on the grounds that its Codes 'set standards for property agents, not consumers'. It also said it uses AI itself to manage its caseload.

At the Financial Ombudsman Service, chief operating officer Marc Harris wrote on 27 March that about a third of responses to initial assessments in a recent sample of cases appeared to be AI-generated or heavily assisted. Some submissions from professional representatives ran to more than 200 pages in reply to a six-page provisional decision. The service reported 'hallucinations', which it described as 'fabricated laws, misquoted regulations or invented past decisions', and said it was sharing what it had found about professional representatives with the FCA and the SRA. It has also said, as Which? reported in May, that AI can help people explain complaints more clearly, including vulnerable and neurodivergent consumers.

The BBC reported on 25 August that councils and schools are seeing the same thing. Deborah Evans, chief executive of Lawyers in Local Government, which represents 5,600 council lawyers, said a complaint that would once have been a side of A4 about bins not being emptied could now run to between 19 and 27 pages, with case law and references to Acts of Parliament, and that complaint handlers might then need to consult lawyers. The Association of School and College Leaders said parents were sending several long complaints at once. The Local Government and Social Care Ombudsman told the BBC that AI needed to be used with 'discretion'.

Freedom of information teams report it too. On 6 May the Information Commissioner's Office published guidance on AI-generated FOI requests after authorities reported rising volume and complexity, including requests that misquote the Act. 'FOI practitioners have told us clearly that the rise in AI-generated requests is changing the day-to-day reality of their work,' said Deborah Clark, the ICO's upstream regulation manager.

Licensing has produced the first known criminal case. In April a resident pleaded guilty under section 158 of the Licensing Act 2003, which makes it an offence to make a false statement in connection with a licence application or review, after submitting fictitious reports and AI-generated emails to Westminster City Council during the licence review of the Heaven nightclub. The court gave him a 12-month conditional discharge, according to the Westminster Times.

How the submissions are produced

Most of what public bodies describe comes from general-purpose chatbots. A few lines describing a problem, given to a chatbot with a request for a formal complaint or an FOI request, come back as a long, structured letter with headings, statutory references and case citations, written in a confident legal register. The cost to the sender of producing that letter has fallen to a few minutes. The cost to the receiving body of reading, understanding and answering it has not fallen, and it rises with the length.

The ICO, the Financial Ombudsman and the council lawyers quoted by the BBC all describe the same errors: legislation quoted wrongly and cases that do not exist. Letting agents told Mortgage Solutions in August that AI tools sometimes draw on bills that were proposed but never passed in that form. The courts have seen it from the professional side: in June 2025 the Divisional Court considered a judicial review claim against the London Borough of Haringey that cited five authorities which do not exist, and used the judgment to warn lawyers about relying on AI research without checking it.

Alongside chatbots there are services built for one kind of submission, such as tools for writing planning objections and a disrepair app, launched by claim.co.uk with the University of Salford, that assesses damp and mould from photographs. Agentic tools go a step further. They can fill in web forms, send emails and follow up on a user's behalf, which removes most of the remaining effort of submitting and chasing. The ICO's guidance deals with high-volume coordinated campaigns and repeated requests for substantially similar information, both of which become cheap to produce with these tools.

Who it reaches

The bodies affected are those whose work starts with a written submission from the public. Councils receive complaints, FOI and subject access requests, planning representations and licensing objections. Housing associations and council landlords run complaint procedures under the Housing Ombudsman's statutory code. Ombudsmen and regulators receive the escalations, and the Financial Ombudsman's experience shows professional representatives using the same tools as consumers. Schools and academy trusts answer FOI requests and run their own complaint procedures.

Procurement teams meet it from bidders. The Cabinet Office's Procurement Policy Note 02/24, published in March 2024, told contracting authorities to plan for 'a general increase in volume of tender responses' as suppliers use AI, to 'factor additional time into procurements for additional due diligence and verification around AI', and to establish 'the accuracy and credibility of tender responses'. It lets authorities ask bidders about AI use, but says the example questions 'should not be scored or taken into account in the tender evaluation' and that AI use by suppliers 'is not prohibited'.

All of these teams work to deadlines set by statute or by a code, with a duty to deal with each submission on its merits.

The legal position

UK law on complaints and information requests does not treat a submission differently because AI helped write it. The ICO's guidance says the Freedom of Information Act is technology-neutral and that authorities 'must respond to these requests unless they are invalid because they don't meet the requirements in section 8', which means a request in writing, with a name, an address for correspondence and a description of the information wanted. It adds: 'Just because a request contains inaccuracies, it doesn't mean it's invalid.'

The existing FOI provisions still apply. An authority has 20 working days to respond under section 10. Where a request is unclear it can ask for clarification, and under section 1(3) it does not have to comply until the clarification arrives, with the 20 days running from then. Section 12 allows refusal where compliance would exceed the cost limit, which is £600 for central government and £450 for other authorities, and similar requests from the same person, or from people acting together, within 60 working days can be aggregated. Section 16 requires advice and assistance, and the ICO expects authorities to help requesters narrow over-broad AI-generated requests.

Section 14 allows an authority to refuse vexatious or repeated requests. The ICO lists AI-related situations where it may apply, including 'repeated requests for substantially similar information', requests intended to disrupt, requests that reopen closed matters and high-volume coordinated campaigns, and says the list is not exhaustive. The test is still applied to the request and its effect on the authority.

Subject access requests under UK GDPR carry a one-month deadline, which can be extended by up to two months where requests are complex or numerous, and a controller can refuse a request that is manifestly unfounded or excessive. The Data (Use and Access) Act 2025 confirmed that a controller need only carry out reasonable and proportionate searches.

For complaints, the timescales come from the ombudsmen's codes. The Housing Ombudsman's Complaint Handling Code has been statutory for social landlords since 1 April 2024, under the Social Housing (Regulation) Act 2023. It requires acknowledgement within five working days, a stage 1 response within 10 working days of acknowledgement and a stage 2 response within 20. The Local Government and Social Care Ombudsman's code for councils uses the same timings, allows extensions of up to 10 and 20 working days where there is good reason and the reason is explained, and councils were expected to comply from 1 April 2026.

Planning and the courts have gone further on disclosure. The Planning Inspectorate's guidance, first published in September 2024 and updated in February 2026, asks anyone who uses AI to draft or substantially rewrite casework material to say so, name the tool, explain what it was used for and confirm the checks made. It warns that improper use could be treated as unreasonable behaviour, which is open to an award of costs. The Civil Justice Council consulted from 23 April to 16 July 2026 on rules for AI in court documents prepared by lawyers. Litigants in person were outside its initial scope, and a final report is expected later in 2026.

Public bodies also have duties under the Equality Act 2010, including reasonable adjustments and the public sector equality duty. People who find writing difficult are among those who gain most from AI drafting, as both the Financial Ombudsman and the study note, so any policy that treated AI-assisted submissions less favourably would need testing against those duties.

Triage

Triage starts with finding the request inside the document. A structured online form, with separate fields for the problem, the outcome wanted and the evidence available, produces shorter and more consistent submissions whatever tool the sender used; Deborah Evans suggested forms to the BBC as a way of limiting length. Where long letters still arrive, the first step is a short written statement of what is being asked, checked with the sender where it is unclear, so that the response answers the actual question. For FOI requests, the ICO treats this as part of the section 16 duty.

Tools on the receiving side can help with the reading. The ICO lists drafting responses, simplifying complicated correspondence, summarising datasets and better redaction as possible uses, with practitioners checking accuracy before anything is disclosed. The Financial Ombudsman and the Property Ombudsman both use AI in case handling while keeping decisions with people.

Grouping near-identical submissions lets a body answer the shared points once. For FOI it also shows where section 12 aggregation or section 14 may apply, provided the statutory tests are met for each request.

Verification

Verification is mostly about citations. A caseworker who checks each statute and case before relying on it, or before rebutting it, avoids accepting a fabricated authority and avoids spending hours answering points with no legal basis. Where a submission misstates the law, the ICO suggests correcting the misunderstanding politely and then dealing with the substance.

Identity matters where weight depends on who is speaking, or how many people are. Licensing and planning representations carry weight partly because they come from affected residents, and the Heaven case turned on letters presented as coming from neighbours; police traced two of the submissions to one person through their IP addresses. Checking that representations come from real and distinct people is proportionate where numbers influence a decision. For an ordinary service complaint it is unnecessary.

A published policy on AI-assisted submissions

A short published statement sets expectations on both sides. It can say that the body accepts submissions drafted with AI, that the sender is responsible for the accuracy of what they send, that decisions rest on the evidence rather than on the length or tone of a letter, and that a summary of the main points at the top is welcome. The ICO's guidance includes example wording that authorities can adapt to encourage responsible use by requesters.

Whether to ask senders to declare AI use is a choice with precedents on both sides. The Planning Inspectorate asks for a declaration in casework, the Property Ombudsman decided against requiring one, and PPN 02/24 lets contracting authorities ask bidders on condition that the answer is not scored. In SAASiQ's view, for a complaints service an optional declaration with no consequence attached gives the body data on how many submissions are AI-assisted without putting anyone off complaining.

Capacity planning

The Housing Ombudsman's new AI category points to the data a body needs. That means counting submissions and recording which were AI-assisted, how long they were, how long they took to handle and how many needed clarification. Deborah Evans told the BBC that council legal teams would need to double in size to keep up; a body's own figures will show whether its position is similar, and they are the evidence a business case for extra staff will need.

Forecasts should assume volumes keep rising. The Property Ombudsman said on 29 July that it 'does not expect complaint numbers to stabilise in the near term', and the Housing Ombudsman expects more complaints in its AI category. Deadlines under FOIA and the complaint codes do not move with demand, which leaves staffing, triage and the codes' extension provisions, used with reasons given to the complainant as the codes require.

The ICO's guidance, Freedom of Information (FOI) and artificial intelligence, is on its website, and the Housing Ombudsman's Annual Complaints Review 2025-26, which introduces the AI complaint category, was published on 15 September 2026.

SAASiQ - Intelligent Solutions for SaaS ©