The end of friction reaches public services: councils and NHS trusts should expect more complaints and compensation claims
Councils, health regulators and tribunals report sharp rises in complaints and requests, and many now link them to people using AI. Personal AI agents that draft, send and follow up for their owners come next, and they put capacity and budgets at risk.
Good Transformer is Patrick Hussey's AI practice. Want help with this in your business? Get in touch.
Dorset Council received 2,406 complaints in the year to March 2026, a quarter more than the year before and more than three times as many as in its first year, 2019/20. A report to its audit committee in April said there may be several reasons, but "one thing that is for certain is that the public's increasing use of artificial intelligence is having an impact."
Dorset is not alone. Complaints about English councils that reached the Local Government and Social Care Ombudsman rose by a third, to 27,625, the sharpest rise in ten years. Written complaints in the NHS in England, including those to GP and dental practices, rose by almost a fifth, to 305,264, according to figures published on 15 October. In June the judges who run the employment tribunals reported that one kind of urgent application, which used to arrive about 20 times a year across Great Britain, now reaches most tribunal offices that often each month, "often indicating use of artificial intelligence".
That is the first step: people using AI to write. The next step has begun in the US. Since September, Meta, OpenAI and Anthropic have each released a personal AI agent, an assistant that acts for one person rather than only answering their questions. Meta's agent, Muse, can send emails, fill in forms and negotiate, asking its owner's approval before it sends a message, and Meta says it "keeps working after people close the app". None of these agents is yet available to UK consumers. The government is exploring agents of its own: it said in August 2025 that agents it builds could deal with public services on people's behalf, and could be rolled out across the country from late 2027.
Last week we described what personal agents mean for businesses, a change we called "the end of friction". For a council, an adult social care department or an NHS trust, the end of friction is a risk to capacity and to budgets, and three numbers will show it first: complaints, compensation requests and compensation payouts. To judge the risk, a leader needs to know what friction has been doing, what the rise so far shows, what an agent adds, where the extra demand will arrive, and what it would cost.
What "the end of friction" means for a public service
Friction is the time, effort and small irritations that stand between a person and what they want: the form, the queue, the eligibility rules, the complaints procedure nobody reads. In shops and banks it keeps customers loyal out of habit. In public services it keeps many people from asking for what they are entitled to.
Benefits show the scale. The Department for Work and Pensions estimated on 22 October that up to 900,000 families entitled to Pension Credit do not claim it, leaving up to £2.4bn a year unclaimed. Up to 270,000 pensioners entitled to Housing Benefit, which councils run, do not claim it either, leaving up to £1.3bn unclaimed (DWP).
Complaints show the same gap. In a YouGov poll for Healthwatch England, fewer than one in ten people who said they had received poor NHS care in the past year made a formal complaint (Healthwatch England). So more than nine in ten people with cause to complain did not.
Appeals show it most clearly. Families registered 25,000 appeals against council decisions on special educational needs in 2024/25, up 18% on the year before. Of the appeals the tribunal decided, 99% went at least partly in the family's favour (Ministry of Justice). When appeals almost always succeed, the number of appeals depends on how many families have the time and confidence to bring one.
So effort has kept the demand on public services lower than the need for them. Some of that effort protects nothing and costs people what they are owed. Some of it is deliberate: proof of identity, consent before personal data is shared, a fair queue. AI takes away the effort without telling the two apart, so a public service has to decide which checks it keeps on purpose, and remove the rest of the effort itself.
What the numbers show so far
A growing number of public bodies now report rising complaints and requests and link them, at least in part, to AI. Some say so in their own reports, and others have said so to journalists. None has measured how much of its rise AI explains.
| Organisation | What rose | By how much | What it says about AI |
|---|---|---|---|
| Employment tribunals (Presidential Guidance, June 2026) | Applications for interim relief, an emergency order that is heard urgently | From "perhaps 20 such applications a year" across Great Britain to "a similar number each month" in most offices | The applications are "often indicating use of artificial intelligence" |
| Dorset Council (audit committee report, April 2026) | Complaints, 2025/26 | Up 25%, to 2,406 | "the public's increasing use of artificial intelligence is having an impact" |
| General Pharmaceutical Council (annual report, 2025/26) | Concerns raised with the pharmacy regulator | Up about a third, from 7,378 to 9,928 | Its chief enforcement officer told the Pharmaceutical Journal: "We are also experiencing a sharp increase in lengthy AI-generated complaints" |
| Optical Consumer Complaints Service (reported by Optometry Today, July 2026) | Complaints about opticians | Up 31%, from 1,679 to 2,201 | Its head said consumers use AI to write complaints, and "The presentation of the complaint will often be lengthier than perhaps we may have seen in the past" |
| Lincolnshire County Council (BBC, May 2026) | Freedom of information requests, 2025/26 | Up 18%, to 1,992 | Its head of business support: "You can tell by the way complaints or requests are worded that they have used AI technology" |
| European Ombudsman (European Ombudsman, April 2026) | Complaints handled, 2025 | Up 54%, from 2,264 to 3,490 | The rise "is believed to be in part due to AI tools suggesting the Ombudsman when people look for help" |
| The national bodies say the same. The Parliamentary and Health Service Ombudsman wrote in its annual report this summer that "AI tools like ChatGPT and Copilot help complainants create and submit complaints to us." On 25 October Cllr Richard Clewer of the Local Government Association told The Independent that longer, more complex complaints written with AI are "increasing administrative burdens at a time of significant resource pressure". Caseworkers in MPs' offices, who pass many residents' cases on to councils and trusts, told the paper that around half of the cases they receive are now at least partly written by AI (The Independent). The Financial Ombudsman Service, which handles complaints about banks and insurers, is the only body we found that has estimated AI's share of what it receives. From a small sample, it found that AI "is likely to have contributed to up to a third (35%) of responses to initial assessments" (Financial Ombudsman Service). |
GP practices report it too. In a Medical Protection Society poll of 158 practice managers in September, 45% said AI-written complaints had a "negative impact on staff morale due to increased workload", and only 7% had never received one (Medical Protection Society).
Two cautions apply. First, rises are also happening where no one names AI. The council ombudsman points to services "under sustained and serious strain", and NHS England gives no cause for its rise. Dorset's own report lists several possible factors. So AI is one cause among several, and nobody yet knows its size. Second, many of these complaints are justified. The journalist Sam Freedman told The Independent: "You've got a situation where people have struggled to write a complaint, or have the means to do so, and their complaint is genuine."
What a personal AI agent adds
Today a person uses a chatbot to write a letter, then sends it themselves. An agent does the rest of the work that puts most people off.
It reads every page. Cloudflare, whose network sits behind more than a fifth of websites, said in July that "more than 50% of traffic on the Internet is now non-human" (Cloudflare). Most of that is search engines and other crawlers rather than personal agents, but it shows that machines already read more of the web than people do. An agent working for a resident reads your eligibility rules, charging policy and complaints procedure in full, and acts on what they say.
It tracks your deadlines. Adult social care and NHS complaints in England fall under regulations that say the organisation "must acknowledge the complaint not later than 3 working days" after it arrives (legislation.gov.uk). For other council complaints, the ombudsman's Complaint Handling Code says councils should reply to a first-stage complaint within 10 working days of acknowledging it and to a second-stage complaint within 20, with limited extensions (LGSCO). The Freedom of Information Act requires an answer "promptly and in any event not later than the twentieth working day following the date of receipt", with some exceptions (legislation.gov.uk). Few residents know these dates. An agent can note each one and write again when it passes.
It finds named people. Contact pages, committee reports and organisation charts often give the name and email address of the officer responsible. An agent can write to that person directly as well as to the main inbox. Messages sent to individual officers can bypass the complaints log, so they may never be counted.
It writes at length. Dr Sarah Townley of the Medical Protection Society said of AI-written complaints to GP practices: "Letters can be 20 pages, sometimes longer." The Financial Ombudsman Service has received submissions from lawyers and claims firms using AI that run to "approximately 200 pages in response to a six-to-eight-page provisional decision".
And it follows up. An agent does the chasing that put most people off. Meta says Muse "comes back when something changes or when it needs approval" (Meta), and its own examples include lowering a bill. One person's agent can also work on several matters at the same time.
Who gains and who loses when friction falls
Residents and patients who were put off by the effort gain most. Some will receive benefits, assessments and remedies they were always owed. For a public service that exists to meet need, that is part of the job, even when it is expensive.
The companies that make agents gain users. An agent that can lower a bill or win a refund is one people will pay for and keep using, so its makers have every reason to make it good at dealing with organisations, public bodies included.
The people who lose are those waiting in a longer queue. Freedman described the problem: "for any individual it's easier to use AI, but if every individual uses AI, everyone's outcomes get worse because it takes so long to get a decision." The employment tribunals already show it: to fit in the extra urgent hearings, the judges wrote, "other hearings are often postponed or delayed." Residents who cannot pay for an agent, or would not use one, may wait behind those who can. That is a fairness question every public body will have to answer. Staff also lose time to reading and checking AI-written letters, some of which contain errors. Hertfordshire County Council warns residents that AI can "make mistakes or make up incorrect 'facts'" (Hertfordshire County Council).
Where the end of friction will run through your organisation
The extra demand will not arrive in one place. It will move through an organisation in the order a resident's agent works, through seven places, and each has a cost and a measure.
1. The website. Agents read first. Every eligibility rule, charge, opening time and procedure on your site becomes something an agent relies on and quotes back. AI tools also decide where people send their message. Dorset found that AI search summaries were pointing residents to its complaints email address rather than its online form. When an agent acts on out-of-date information, it produces a wrong application or a complaint, and the complaint will be justified.
2. The inbox. Agents then write, to the main inbox and to any named officer whose address they find. Freedom of information and data protection requests arrive here too, each with a deadline set in law. Volume is only half of it. Durham County Council received fewer freedom of information requests in the 12 months from June 2025, but said they took longer to answer, which it suggested was because many were written using AI (BBC). The question is whether you can still answer within the times you publish.
3. Demand for services. As applying gets easier, more people who are entitled to help will ask for it: assessments, adaptations, appointments, benefits, school places. GP practices show how fast demand grows when the door gets easier to open. Patients submitted a record 8.6 million online consultation requests in February 2026, up 85% on a year earlier, or 53% when measured per 1,000 patients (NHS England). Online forms are exactly what agents fill in. This demand lands on assessment teams and waiting lists, not on the complaints team.
4. Complaints. When a service is late or refused, an agent can turn that into a formal complaint in minutes, cite your procedure and track your deadline. First-stage complaints rise, and more of them go on to the second stage.
5. Escalation. Agents can find and use the routes beyond you: the ombudsmen, the Information Commissioner's Office and the tribunals. The council ombudsman upheld 86% of the council complaints it investigated last year (LGSCO), so a rise in escalations is likely to bring a rise in findings against councils. AI is reaching the stage after that too. The ombudsman, Amerdeep Clarke, told the BBC in August: "We have also seen a rise in the use of AI in legal challenges to our decisions" (BBC).
6. Compensation and legal claims. An agent can ask for a financial remedy every time, and draft a legal claim. For councils that own homes, the Housing Ombudsman ordered £4.8m in compensation in 2025-26 and recommended a further £1.2m (Housing Ombudsman). In the NHS, trusts pay for clinical negligence claims through their contributions to NHS Resolution. It received 15,236 new claims and reported incidents in 2025/26, up 6%, and paid out £3.24bn, of which £811m went on legal costs (NHS Resolution). The National Audit Office found that low-value claims make up "three quarters of all claims", and that legal costs on claims of £25,000 or less were "3.7 times higher than the damages awarded" (National Audit Office). So a rise in small claims would add far more in legal costs than in damages.
7. The budget. Every stage above costs staff time, and the later stages cost cash. The government has agreed exceptional financial support in principle for 36 councils for 2026-27 (GOV.UK). Healthwatch found that NHS trusts had an average of 4.4 full-time staff working on complaints in March 2024, against 4.66 in 2019. Few organisations have spare capacity to absorb a sustained rise.
The table puts the seven places together, with the measure to watch at each and the latest national figure where one exists.
| Where | What agents change | What to measure monthly | Latest national figure |
|---|---|---|---|
| Website | Every page is read and acted on | Pages with out-of-date information; complaints that cite your own published information | Over half of internet traffic is non-human, mostly crawlers (Cloudflare, July 2026) |
| Inbox | More messages, longer messages, more sent to named officers | Contacts received; average handling time; share answered within your published time; FOI and data requests | No national figure |
| Demand for services | More eligible people apply | Applications and assessment requests; waiting times | 8.6 million GP online consultation requests in February 2026, up 85% (NHS England) |
| Complaints | Complaints are easy to write and to track | First-stage complaints; share escalated to the second stage; share answered on time | 305,264 written complaints in the NHS, up 18.9% (2025-26) |
| Escalation | Agents use every route beyond you | Ombudsman referrals and findings; tribunal appeals; legal challenges | 27,625 council complaints to the ombudsman, up 33% (2025-26) |
| Compensation and claims | A financial remedy can be requested every time | Compensation requested; compensation paid; legal claims received | £3.24bn paid on NHS clinical negligence claims (2025/26) |
| Budget | Staff time and payouts rise together | Cost of handling per complaint; total remedies paid | 36 councils with exceptional financial support agreed in principle (2026-27) |
If you would like to work through this table with your own organisation's figures, that is the kind of work our AI Strategy service does with leadership teams. Get in touch and we can start with a free 30-minute call.
What a rise would cost
No one can yet say how much agents will add, because they are not yet available to UK consumers. But a leader can work out what a rise would cost, using figures most organisations already hold.
Take an example. Suppose a council handled 2,000 first-stage complaints last year, and each took an officer about half a day to investigate and answer. If its complaints rose by a quarter, as Dorset's did last year, that would be 500 more complaints and about 250 more days of officer time. At about 220 working days a year, that is more than one full-time post, before any rise in second-stage complaints, ombudsman cases or compensation. If each complaint also takes longer to read because it is longer, as Durham found with its requests, the cost rises again.
Run the same sum with your own figures: last year's complaints, your average handling time, and the compensation you paid. Then ask what a rise of a quarter, or a third, would mean for each.
The council ombudsman has given its own view of where the money goes: "The cumulative expense of complaints, legal challenges and emergency interventions far exceeds the cost of investment in effective, well-managed services" (LGSCO). In other words, money spent getting a service right costs less than money spent dealing with complaints about it.
What to do in the next six months
- Start counting now. Set a monthly baseline for the measures in the table, before agents reach UK residents. If you count only three things, count complaints received, compensation requested and compensation paid. Add a rough count of contacts that look AI-written, and the average time each takes to handle. Without a baseline you will not see the change until it reaches your budget.
- Check what agents will read. Ask each service to review its eligibility rules, charges, procedures and published response times. Remove pages that are out of date, and make sure no two pages give different answers. Ask an AI assistant how to complain to your organisation, and check where it sends people. An agent will quote your own pages back to you in a complaint.
- Settle more complaints at the first stage. The council ombudsman saw a 20% rise last year in complaints it was satisfied had already been put right locally. Each one settled early stays out of the escalation and compensation figures. Give first-stage staff the authority to put things right quickly, including small payments where they are due.
- Set rules for AI-written and agent-sent contact. Do not reject a complaint because AI helped write it: the ombudsmen tell residents that AI can help them. Decide how you will handle very long submissions. Croydon Council may ask for a long complaint in a different form, or write a short statement of it for the resident to approve (Croydon Council), and Dorset's report proposed taking most complaints through its online form only. It noted that this cannot apply to statutory social care complaints, which must still be accepted in several ways. Decide how you will confirm that an agent is acting with the person's consent, and what proof of identity you need before releasing anyone's personal data. Where a system holds personal data or payments, you may decide, as HMRC has for its Government Gateway, that automated tools may not sign in as a person (HMRC).
- Use AI on your side, with a person deciding. AI can summarise a twenty-page letter, check it against the case file and draft a reply. Use a tool your organisation has approved for personal data, after a data protection impact assessment, not a public chatbot. A named officer should still decide the outcome and sign the response.
- Put a scenario in next year's budget. Use the calculation above for complaints, compensation and legal costs, and agree in advance what you will do if complaints rise past that level.
- Tell the regulator what you are seeing. The Information Commissioner's Office is collecting evidence on how data protection law applies to AI agents until 20 November 2026 (ICO). Public bodies that receive AI-written and agent contact are well placed to describe it.
These are decisions for a leadership team, not only for a complaints manager. Our AI Strategy work helps a leadership team decide where AI should handle incoming demand and sets the rules for it in a 90-day plan and a plain-English AI policy. Book a free call to talk it through.
Common questions
Can a council or NHS trust refuse a complaint because AI wrote it?
The ombudsmen tell residents that AI can help them write a complaint, so a complaint should be judged on what it says. Some councils, such as Croydon, ask for a very long complaint in a different form, or write a short summary for the resident to approve.
How can we tell whether AI is behind our own rise?
You cannot measure it directly. Some people who would have complained anyway now use AI to write, and long complaints that misquote the law were not invented by AI. What you can do is track signs month by month: average length, handling time, and the share of complaints that cite law or policy that does not apply. If those rise together with volume, and nothing in the service itself has changed, AI is a likely part of the reason.
When will personal AI agents be available in the UK?
The agents from Meta, OpenAI and Anthropic are not yet available to UK consumers, and none of the companies has given a date. Separately, the government has said agents it builds could help people with public services from late 2027. Many residents already use AI chatbots to write their letters.
Is the same thing happening in the US?
Yes. In a survey of 193 US federal freedom of information staff earlier this year, 46% had received AI-generated requests, and one reported receiving 50 to 100 requests in a matter of minutes (US National Archives). In Medicare Advantage, the private plans that deliver much of the US government's Medicare programme, only 11.5% of refused requests were appealed in 2024, but 80.7% of appeals succeeded at least in part (KFF). Personal agents are available in the US now, so US public bodies will see agent contact first.
Sources
- Dorset Council, Corporate complaints policies, report to the Audit and Governance Committee, 27 April 2026
- Local Government and Social Care Ombudsman, Annual Review of Local Government Complaints 2025-26, July 2026
- NHS England, Data on Written Complaints in the NHS, 2025-26, 15 October 2026
- Employment Tribunals, Presidential Guidance on interim relief, 22 June 2026
- General Pharmaceutical Council, Annual report 2025/26, September 2026, and Pharmaceutical Journal, Sharp increase in AI-generated complaints, 18 September 2026
- Optometry Today, Five insights from the General Optical Council, 9 July 2026
- BBC, Lincolnshire County Council freedom of information requests, 30 May 2026; Durham County Council freedom of information requests, 6 October 2026; AI and council complaints, 25 August 2026
- European Ombudsman, Annual Report 2025, 22 April 2026
- Parliamentary and Health Service Ombudsman, Annual Report and Accounts 2025 to 2026, July 2026
- The Independent, How AI turned Britain into a nation of super-complainers, 25 October 2026
- Medical Protection Society, Almost half of GP practice managers say AI-generated complaints affecting staff wellbeing, 12 October 2026
- Financial Ombudsman Service, Response to the Mills Review, February 2026
- Meta, Introducing Muse, 8 September 2026
- GOV.UK, AI helpers could coach people into careers and help them move home, 16 August 2025
- Department for Work and Pensions, Income-related benefits: estimates of take-up, financial year ending 2025, 22 October 2026
- Healthwatch England, A pain to complain, January 2025
- Ministry of Justice, Tribunal Statistics Quarterly, July to September 2025, 11 December 2025
- Cloudflare, The agentic Internet, 1 July 2026
- The Local Authority Social Services and National Health Service Complaints (England) Regulations 2009, regulation 13
- Local Government and Social Care Ombudsman, Complaint Handling Code
- Freedom of Information Act 2000, section 10
- Hertfordshire County Council, Using AI when you contact the council
- NHS England, Record GP access figures, 26 March 2026
- Housing Ombudsman, Annual Complaints Review 2025-26, 2026
- NHS Resolution, Annual Report and Accounts 2025/26, July 2026
- National Audit Office, Costs of clinical negligence, 17 October 2025
- GOV.UK, Exceptional Financial Support for local authorities for 2026-27, updated 18 August 2026
- Croydon Council, Using AI to help you with your complaint
- HMRC, Use of HMRC sign-in details, automation and accessing HMRC's web services, 27 May 2026
- Information Commissioner's Office, agentic AI call for evidence, October 2026
- FOIA Advisory Committee, Volume and Frequency Subcommittee final report, 8 June 2026
- KFF, Medicare Advantage insurers made nearly 53 million prior authorization determinations in 2024, 28 January 2026
This is general information, not legal advice.
Work with Patrick
Want a clear AI plan for your business?
AI Strategy is two to three weeks with your leadership team: a 90-day plan, the use cases worth doing and a plain-English policy. Email, call or book a free 30-minute call.